FutureCentral > Sustainable Agribusiness in the AI Era Sample Chapter 13

FUTURECENTRAL PRESS · BOOK SAMPLE

Sustainable Agribusiness in the AI Era

Chapter 13: Value Chains and Market Linkages

Who sets the terms in an agri-food chain, what moving up one is known to pay, and why the producer sometimes takes a smaller share of a larger number

Learning Outcomes

Six outcomes. Each is demonstrated on a chain the reader can draw, not on a diagram somebody else drew:

Map an agri-food chain to a published method, naming every element that method requires the map to carry.

Classify the governance of a single link on the five-type scheme, and state what that scheme claims about a whole chain and what it does not.

Identify the condition under which a producer is absent from a chain, and separate it from the condition of being badly placed inside one.

Separate the four upgrading paths, and state for each what the published evidence establishes about its return, including where that evidence does not exist.

Read a market-access requirement as an instrument of chain governance, naming who funds compliance and in which year.

Judge a claim made for a digital trading platform against the randomized evidence on what such platforms do to producer prices.

Opening Vignette

Two numbers describe the same grape grower in the Nashik belt, and they point in opposite directions. Selling into the domestic chain, that grower keeps about 35 percent of what the consumer in Delhi pays. Selling into the export chain that ends on a Dutch shelf, the same grower keeps about 21 percent. Both figures come from one central bank working paper, from the same fieldwork, in the same season.¹

A manager reading only those two numbers would keep the crop at home. That would be the wrong decision, and the reason it is wrong is the whole subject of this chapter.

The absolute price the grower receives is higher in the export chain. What the missing 79 percent bought was packaging, grading labor, laboratory certification, inland transport, and ocean freight, none of which the domestic chain requires and all of which somebody funds. The share fell because the chain got longer and costlier to run. The money the grower took home went up.

Then a third number changes the question again. Only 50 to 60 percent of the crop sorts to export grade in the first place, so the export chain takes a quality-selected slice, and it takes it after the grading table has already spoken. Most of the crop leaves by other routes whatever the grower prefers. The grower is therefore in both chains at once, and the decision was never which chain to join.

That is what a value chain question looks like when it is posed properly. Who governs each link, what a producer must fund to stay in it, what share the structure leaves behind, and what the alternative pays. This chapter builds the instruments for all four, and reports how little of the received wisdom about moving up a chain has ever been measured.

Section 1 maps a chain and says what the map is for. The first framework takes the governance of a single link and finds the case the framework treats as absence. Section 2 sets out the four upgrading paths against what is known about their returns. The second framework turns the paths into a decision. Section 3 reads the buyer’s government as a chain governor. Section 4 takes digital platforms and asks who captures what they create.

1. Mapping a chain, and what the map is for

Four published methodologies exist and a manager can follow any of them. The German international-cooperation agency GIZ’s ValueLinks 2.0 runs eleven modules and specifies what a chain map must show. The European Commission’s Value Chain Analysis for Development method has framed more than sixty-five country studies since 2016 around four questions: what the chain contributes to economic growth, whether that growth is inclusive, whether the chain is socially sustainable, and whether it is environmentally sustainable. The Making Markets Work for the Poor toolbook supplies eight tools and treats income distribution and employment distribution as two separate exercises. A fourth guide, written for agri-food specifically, sits alongside them.²

They all draw the same picture. The German manual is the most explicit about what a map must contain, asking for the stages, the operators at each, the business linkages between them drawn as arrows, the end markets, the support service providers, and the institutions the chain runs inside. What separates the four is what each obliges the analyst to count once the picture is drawn, and that obligation is where the value sits. A map with no arithmetic underneath it is a diagram for a slide.

A usable map carries five things, and most published maps carry three. The actors at each stage and how many of them there are. The physical flow, in tonnes, with the loss at each transfer. The price at each handover, so the margin between stages can be computed. The service providers and regulators standing alongside the chain without being in it. And the direction in which specification travels, because the actor who writes the specification is rarely the actor who owns the goods, and a map drawn only in the direction of physical flow will miss that entirely.

The arithmetic that follows is the farm share, and the global figure is worth memorizing. Standardized input-output accounting across sixty-one countries for 2005 to 2015, covering roughly 90 percent of global output, puts the farmer’s average receipt at 27 percent of consumer spending on food eaten at home.³ The range for middle-income and high-income countries sits between 16 and 38 percent, and the share falls as an economy gets richer. Most of what a consumer pays for food is paid for something that happens after the farm gate.

Indian figures straddle that band and vary by how perishable the crop is. The same central bank release that produced the grape numbers puts the farmer’s share of the consumer price at 35 percent for grapes, 31 percent for bananas, and 43 percent for mangoes, against 75 percent for gram and about 70 percent for milk. Those Indian numbers are a price share for one commodity and the global figure is an expenditure share across a basket, so they sit beside each other and are not comparable. The spread across crops is the finding. An earlier survey put the range at 28 to 78 percent on the same pattern.⁴ Perishability predicts this better than anything else.

One caution belongs with all of these figures. A share is a ratio, and Chapter 12 showed how a denominator can do the moving. A falling farm share can mean the chain is extracting more, and it can equally mean the chain is doing more. Telling them apart takes the absolute price, which is why the map carries both.

Framework: Chain Governance and the Excluded Cell

The standard account of how a link in a global value chain is governed comes from G. Gereffi, J. Humphrey and T. Sturgeon, “The governance of global value chains,” Review of International Political Economy 12, no. 1 (2005): 78–104.⁵ It compresses a large case literature into three questions and five answers. Two decades of testing have qualified it and extended it, and no replacement typology has been proposed.⁶ This section states it as the authors state it, then adds the qualification one of them published later with a co-author.

Three variables decide how a link is governed, and all three are properties of the transaction. The paper names them as the complexity of transactions, the ability to codify transactions, and the capabilities in the supply base. Complexity asks how much information must move. Codifiability asks whether that information can be written down in a specification a stranger could execute. Supplier capability asks whether the firms available can meet the specification without help. Each takes a high or a low value, and that deliberate crudeness is what makes the framework usable in a meeting where nobody has three months to characterize a transaction properly.

Five governance types follow, and their names are worth using precisely. In markets, the specification is simple, the price does the coordinating and switching costs are low on both sides. In modular chains, the supplier makes to a customer’s specification using general-purpose equipment, so the relationship is close, the information flowing through it is dense, and the assets on the supplier’s side are not locked to any one buyer. In relational chains, complex mutual dependence is managed through reputation, family, or ethnic ties, and the specific assets on both sides are real. Captive coordination is the fourth type, in which small suppliers are transactionally dependent on much larger buyers and face significant switching costs. In hierarchy, the lead firm stops buying and starts owning.

Eight combinations exist, five occur, and the missing cases are the most useful part of the paper. Two combinations are excluded because low complexity with low codifiability is unlikely to arise at all. It is the third exclusion that matters here. Where transactions are simple, the specification is fully codified, and supplier capability is low, the paper’s result is that this leads to exclusion from the value chain, which is a very different statement from saying that such a supplier gets poor terms. The cell describes absence from the chain.

Read that against a smallholder who cannot meet a written residue limit, and the framework says something sharper than any account of unfair terms. A grower failing a codified specification is not being governed badly by the chain. That grower is not in the chain, and no improvement in the terms of trade reaches somebody the terms of trade do not apply to. Entry and terms are separate problems and they need separate instruments.

One of the original authors later returned to fix the framework’s scope, and the correction matters for how it is taught. S. Ponte and T. Sturgeon wrote in the same journal in 2014. They record that most value chains contain a range of linkage types at once and that it is risky to characterize a whole chain, still less a whole industry, from research on inter-firm linkages alone.⁷ They keep the five types and place them at the level of the individual node. The types survive. For the chain as a whole they introduce polarity: unipolar where one actor sets terms throughout, bipolar where two do, multipolar where terms are contested. An agri-food manager asking who governs the dairy chain or the oilseed chain is asking a polarity question, and the five types will not answer it.

A later extension supplies what the original left out, which is where pressure comes from when it comes from outside the chain. S. Ponte, T. Sturgeon and M. Dallas separate four kinds of power on two dimensions, whether it travels directly or diffusely and whether it is exercised by two parties or by many: bargaining power, demonstrative power, institutional power, and constitutive power.⁸ Sustainability pressure in agri-food chains almost never arrives as bargaining power between a buyer and a supplier. It arrives as a standard, a regulation, or a norm, which is to say institutionally and constitutively, and a manager watching only the buyer will not see it coming.

Using the framework takes two passes. Classify each link separately, because the same chain routinely runs captive at the farm gate, modular at the processor, and market-like at the retailer. Then ask who sets terms across the whole thing, which is the polarity question, and where the specification originates, which is the power question. The instrument earns its place when those answers disagree, since that is where a chain is about to change shape.

2. What upgrading is known to pay

Four upgrading paths are named in the literature and two sources are cited for their origin. Process upgrading makes the same thing more efficiently. Product upgrading makes a better or a different thing. Functional upgrading takes on a stage somebody else used to perform, which for a producer means grading, packing, processing, branding, or exporting. Chain upgrading, also written inter-sectoral or inter-chain upgrading, moves the whole capability into a different chain. Two candidate origins circulate for the taxonomy, a 2000 working paper and a 2002 article, and this book cites both and asserts neither, so a reader who needs the origin should read the two documents side by side.⁹

Now the part that ought to be uncomfortable. Searching the peer-reviewed and institutional record to August 12, 2026, located no randomized controlled trial whose treatment is functional upgrading, and no meta-analysis pooling the economic return to any single upgrading path.¹⁰ The taxonomy is a quarter of a century old, it is taught everywhere, and the path this book is about to call the most consequential has never been tested against a counterfactual.

The synthesis literature is larger than the causal literature by an order of magnitude. A scoping review published in Nature Sustainability in 2020 screened 12,320 records and included 202 studies of market links between chain actors and small-scale producers, reporting that 83 percent showed a positive result on at least one outcome assessed.¹¹ Read that as a success rate and the reading fails. It is what a mostly observational literature produces when a study may report on any of several outcomes, and the review records that few studies examine a primary outcome such as income alongside an intermediate one.

The most rigorous synthesis says it with more discipline. A Campbell systematic review published in 2024, covering value chain interventions and women’s economic empowerment, included 118 studies of which four are randomized trials.¹² Its finding is that such interventions do improve income, assets, productivity, and savings, that the effects are small, and that confidence is limited by methodological quality. Its authors add the sentence this chapter has to carry: the gains accrue more through skill acquisition and better inputs than through improvements in market access, which is a finding directly at odds with the premise of most market-linkage programs. Four randomized trials out of 118 studies is the honest measure of how much is actually known here.

What has been measured properly is channel upgrading, a producer moving from a traditional to a modern or export buyer, and there the best-identified work is encouraging. An endogenous switching regression on Kenyan vegetable growers associates participation in supermarket channels with a 48 percent gain in average household income. A quasi-experimental study of supermarket procurement in one Central American country finds that selling to supermarkets raises household productive assets, and adds the qualification that participation is largely confined to farmers with advantageous endowments of geography and water.¹³ The gain is real and the entry condition is doing much of the work.

The reframing this chapter takes from the literature came from a study of one national wine chain over a decade. S. Ponte and J. Ewert published it in World Development in 2009.¹⁴ They found better product quality, improved processes, and some functional upgrading coexisting with downgrading, higher risks, and limited rewards in traditional export markets, which is a combination the taxonomy on its own has no way of describing. Their proposal is to treat upgrading as reaching a better deal, assessed on rewards net of risk. That move converts a taxonomy into a decision rule. It does so by forcing the question every taxonomy suppresses, which is better for whom, and who is carrying what if the thing goes wrong.

Framework: The Upgrading Trajectory Map

This is the chapter’s own instrument, and the rest of Part III leans on it.¹⁵ It takes the four paths as given and adds the two columns the taxonomy omits, which are what the move costs and who is left holding the risk. Every row is answered for a named party, because upgrading is something a specific balance sheet does.

Process upgrading is the cheapest path and the one with the shortest payback. What changes is yield, loss, or conversion per unit. What it costs is capital inside a stage the firm already operates, which means the firm keeps the asset if the buyer leaves. The risk stays where the investment is. Chapters 9 and 12 supply the arithmetic. Process gains are also the easiest for a buyer to capture at the next negotiation, so a firm should model what happens if the whole saving is competed away.

Product upgrading buys a specification and pays for it in advance. What changes is grade, variety, presentation, or a certified attribute. What it costs is a compliance investment made before any premium is earned, and the third case gives the measured order of magnitude. The critical property is that most of the spend is specific to one market, so its resale value is close to zero if that market closes. A firm should ask how many buyers accept the specification it is about to fund, and treat a single-buyer answer as the risk it plainly is, since a specification with one customer is an asset with one exit.

Functional upgrading is the path everybody recommends and nobody has tested. What changes is which stage the firm operates, so the firm takes on a margin it did not previously earn and a set of fixed costs it did not previously carry. The evidence position is the one Section 2 records, which is an absence. What can be said is structural: taking on a stage converts a variable cost into a fixed one, and a fixed cost has to be covered in a bad season as well as a good one. The instrument here is a break-even volume.

Chain upgrading is a change of business and should be priced as one. What changes is the market the capability serves. Almost nothing carries over except the physical asset and the workforce, and both were configured for a chain the firm is leaving. This is the path most often taken in a crisis and least often modeled beforehand.

Two columns finish the map. The risk ledger asks who funds the change, who can walk away from it, and how long the funded asset lasts if they do. The reward test asks whether the party making the investment ends up better off net of risk, which is the reframing the 2009 wine chain study proposed. A path passes only when both are answered for the same named party.

The test is one comparison and it is unkind. Name the party writing the check, and name the party who can end the arrangement without breaching anything. Where those are different parties, the upgrade is being funded by whoever has the least power to keep it going, and the chapter’s third case is what that looks like when it fails.

3. The buyer’s government as a chain governor

A great deal of what looks like commercial power in an agri-food chain is public regulation wearing a buyer’s clothes. The specification a European importer sends an Indian exporter is mostly a restatement of what the importer’s own government requires, and an exporter who reads it as a negotiating position has misread it.

The residue regime is the sharpest example, because it converts an agronomic practice into a market-access condition. Regulation (EC) No 396/2005 sets maximum residue levels for pesticides in food and feed across the European Union, and it remains the governing instrument as of August 2026.¹⁶ Its default matters more than any individual limit. Where no specific level has been set, Article 18(1)(b) applies a default of 0.01 milligrams per kilogram unless a different default is fixed for that substance in Annex V, so an exporter can read the applicable figure off the regulation itself. A third-country exporter is bound through one simple mechanism: non-compliant produce may not be placed on the market, so the rule reaches every plot supplying that market with no legal relationship between regulator and grower.

Enforcement sits in a second instrument. Regulation (EU) 2017/625 governs official controls on food entering the Union, and an implementing regulation attached to it lists products from specified origins facing a heavier regime, reviewed at intervals not exceeding six months.¹⁷ The current list as of August 2026 was adopted on June 9, 2026, and took effect on June 30. Its structure sits in three annexes.

One annex raises the frequency of identity and physical checks at the border. A second requires an official certificate and a pre-export analysis for every consignment, which moves the cost of proof from the importing state to the exporting one. A third suspends entry altogether. A product moves between these tiers on evidence, twice a year.

Above the public rule sits a private one, and an exporter meets both. GLOBALG.A.P.’s Integrated Farm Assurance standard, at version 6, is the principal private farm assurance scheme in this space and functions as a buyer-side entry condition.¹⁸ Its group route matters as much as its content. Group certification runs under the scheme’s Option 2, in which a producer group is certified as one entity with an internal quality management system and the certification body samples the square root of the number of member farms. Square-root sampling is what makes certifying four hundred small growers cost something other than four hundred times certifying one, and it is the most practically useful arithmetic in the scheme’s regulations. An entry-level route sits below full certification, and the scheme describes it as a capacity-building step.

Group certification relocates the cost of proving compliance from the farm to the institution above it. The requirement itself is unchanged by that route, and the buildings, records, and equipment it calls for are funded separately. Who funds those, and on whose land they stand, is what the second framework’s risk ledger was built to ask. A specification written by a buyer and applied through an importing state’s rules has to be funded somewhere in the chain, and naming where is the second framework’s first question.

4. Platforms, matching and who captures what

A digital trading platform promises to shorten a chain by matching a seller to a buyer directly. The promise is coherent, the technology works, and the measured effect on producer prices is close to nothing. That gap between mechanism and outcome is the most reliable finding in this literature and the least reported.

Two randomized evaluations bracket the result. A randomized evaluation of a commercial market information service in India reported no significant effect on the prices farmers received, on storm losses, on crop choice, or on cultivation practices, with take-up never exceeding 0.5 percent of farmers in the study districts and 41 percent of those given free access never using it.¹⁹ A randomized rollout of e-commerce infrastructure across 100 villages delivered a real gain of roughly 5 percent in the cost of living among households that used the terminal, and about 1 percent for the average household in a village that had one.²⁰ The gain went to rural consumers, and producers gained nothing measurable. In both, the measured effect turns on how few of those enrolled went on to transact.

One mechanism works and the other does not, and confusing them is where most business cases go wrong. Price information reaches a producer who already has a buyer and improves the bargaining position inside an existing relationship. Matching asks a producer to find a new counterparty, complete a transaction with a stranger, arrange logistics, and bear settlement risk, and the drop-off between registering and trading is where a platform’s theory of change dies.

Value capture in platform businesses runs upward, and the agricultural ones are no exception. A 2024 analysis of digital agricultural platforms records that the underlying cloud infrastructure market was more than 97 percent concentrated among a small number of providers in 2023, and that sector-specific agricultural platforms have generally not reached profitability.²¹ The economics point one way: the platform captures less than the infrastructure it runs on, and the producer captures less than the platform. A manager evaluating a platform partnership should ask which layer the enterprise is being invited to occupy.

Data rights are the second question and they are unsettled. Peer-reviewed work published in 2025 on agricultural data governance reports that in the jurisdiction it examines there is no legal definition of agricultural data, and that ownership, access, and control are left to contractual agreement, which means the position will be set by the terms of service somebody signs.²² Chapter 5 gave the Indian statutory position on farmer data, and nothing in it was written with a trading platform in mind.

On artificial intelligence specifically, the record is empty in a way worth stating. Nothing in the published or institutional literature, checked on August 12, 2026, evaluates a deployed artificial intelligence system for matching, routing, or price discovery in an agri-food value chain against measured business outcomes.²³ Architecture papers exist and pilot descriptions exist. Evaluations do not, and a manager funding one in 2026 is funding a first.

India Case: The plot, the laboratory and the pack-house

Nothing here is disguised. This case reproduces a published procedure of a public authority, and every requirement in it is one somebody currently meets.

Five gates stand between a vine in Nashik and a Dutch shelf, and one published procedure sets out all five. The governing document is trade notice APEDA-QCT/13/2024-27 of February 17, 2026, Procedures for Export of Fresh Table Grapes to the European Union, which governs the 2025–26 season.²⁴ Gate one is plot registration with the district agriculture or horticulture officer, producing a six-part number that encodes state, district, taluka, product, farm, and plot. A registered plot may not exceed 1 hectare. A boundary adjustment extends it to 1.2. Registration runs three years with renewal every year between September 1 and December 31, the number is displayed on the farm where an inspector can read it, and the grower keeps a written record of every chemical applied to the plot.

Gate two is inspection: two visits by the state officer, one at registration and one within twenty days before sampling, at which spray records are verified. Gate three is residue testing. A recognized laboratory draws at least 5 kilograms per hectare from fifteen primary spots, packs it into two boxes so a counter-sample survives, delivers within twenty-four hours, and certifies within six days, against a list of 163 agrochemicals set annually by the National Referral Laboratory at the ICAR-National Research Centre for Grapes, Pune, and read against harmonized European limits.

Gate four sets where the fruit may be processed and packed. The notice permits export only where the fruit is processed and packed in a pack-house recognized by the authority, which then holds representative samples in cold storage at 0 to 1 degree Celsius and 90 to 95 percent relative humidity for sixty days. Gate five is scope: the twenty-seven member states, the United Kingdom, and other countries following the same norms. Read the five together. Every registered plot is capped at 1 hectare and every consignment must pass through a recognized facility. The specification behind both requirements is a published one.

One result belongs beside all that cost. Indian table grapes appear nowhere on the European instrument setting temporarily increased official controls on food of non-animal origin.²⁵ A chain-level compliance architecture, funded largely by growers and exporters, buys that absence, and absence from a control list is worth more to an exporter than any premium.

India Case: Two platforms, and which one the evidence belongs to

No company appears in this case. Two public electronic market platforms do, both from their own published material, and the case is about an error of attribution that has traveled a long way.

Two public electronic platforms share an architecture, and a price effect measured on one belongs to the one it was measured on. One electronic system links regulated market yards across states. Figures reported as of March 2026 put integrated market yards at 1,656, registered farmers at about 1.80 crore, and cumulative trade since 2016 at roughly ₹4.84 lakh crore, which is about $58 billion at approximately ₹84 to the dollar.²⁶ Any figure of that kind should travel with the release it came from and the date on it, since integration is a continuing process and the counts are levels at a date.

The price gains usually cited in this context are increases of about 5.1 percent for paddy, 3.6 percent for groundnut, and 3.5 percent for maize. Those figures come from a study published in the Proceedings of the National Academy of Sciences in 2020, and they were measured on Karnataka’s Unified Market Platform, a state-backed system launched two years before the national one.²⁷ They belong to that platform, in that state, over that period.

The management point is not pedantry about footnotes. Two platforms sharing an architecture and separated by a state boundary are not interchangeable evidence, and an effect size measured on one becomes a forecast when applied to the other. A firm routing volume through an electronic mandi should ask which platform the number in the deck came from, over which crops, and against which counterfactual channel.

The counterfactual is the neglected half. In the grape belt of the first case, growers sell an estimated 85 to 90 percent of output to informal buyers, and that channel is what any platform has to beat.²⁸ It is fast and it is close, and it clears fruit a graded channel would turn away. What it charges for those services is published nowhere this book could trace, which is itself why a platform business case built against it rests on an assumption. Chapter 8 sets out the formal instruments a producer organization can reach for in its place.

Global Case: The certification the chain required and nobody funded

What follows is a randomized design, its measured effects, and a sequence of events the researchers recorded after the study closed and published as an epilogue.

The intervention worked, and then the chain requirement arrived. Thirty-six self-help groups were randomized into three arms of twelve: brokerage services with credit, covering 373 individuals, the same services without credit, covering 377, and a control group of 367.²⁹ Baseline data came in April 2004 and follow-up in May 2005, with 86 percent retention. Adoption of the export crop rose 19.2 percentage points. The share of land under cash crops rose 4.3 points. Marketing expenditure fell by 3,528 shillings a household. Formal deposit holding rose 7.8 points. Household income across the full sample rose without reaching significance, and among first-time growers of the export crop it rose about 32 percent.

A mandatory European private food-safety standard took effect in January 2005, in the middle of that follow-up year. The exporter kept buying until the middle of 2006 and then stopped, because none of the groups had obtained certification. The brokerage lost money on its loans and closed. Growers went back to intermediaries, some produce was left in the field, and most of them returned to the crops they had grown before the program arrived, which is the outcome the intervention had been designed to prevent. In 2007 the exporter resumed with two of the groups, after those two had built a grading shed with a charcoal cooler.

The thing nobody bought had a published price. Compliance under the group option was estimated at about $581 per farmer in the first year, of which roughly $446 was infrastructure with an average life of 7.8 years and about $134 a recurrent annual expense. The authors’ conclusion is the number to carry into any upgrading discussion: first-year compliance cost more than twice the net gain of first-time adopters. The arithmetic was published, the requirement had a commencement date, and the parties still met it after the chain had already broken.

Run the second framework’s test over it and the failure is visible before the first season. The party who would have written the check was the grower. The party who could end the arrangement without breaching anything was the buyer. Those were different parties, the funded asset was specific to one market, and nobody in the chain had a commercial reason to finance a shed standing on somebody else’s land for somebody else’s specification. This is what the risk ledger is for.

Global Case: What a second mill did to a working relationship

The commodity is coffee, the identification is careful, and the finding runs against what competition is normally expected to deliver. The paper behind it is cited in full at the endnote.³⁰

More buyers arrived, and the relationships between farmers and mills thinned out. Researchers censused every coffee mill in one country for the 2012 harvest, surveyed matched farmers, graded a sample of each mill’s output, and measured the relationship in three parts: inputs and loans given to farmers before harvest, cherries sold on credit at harvest, and second payments and assistance afterward. Entry was instrumented using an engineering model of where a mill could physically be sited, scored across the country from aerial imagery, which separates competition from the local conditions that attract it.

An additional mill sited within 10 kilometers cut the overall relational score by about 0.283 standard deviations on the study’s instrumented estimate, against a simple correlation of 0.116 the authors decline to read as conclusive. The effect runs through proximity and through the margin a competing mill can offer, which is why the study reports both channels. Mills ran at lower and more irregular capacity, so unit processing cost rose. Farmers sold a smaller share of their cherry to mills and were likelier to process at home to get cash at the end of the season. Prices did not rise. An index of farmer satisfaction fell, and coffee quality slipped where farmer effort did.

The mechanism is precise and it is why the case earns its place. Competition did not reduce anybody’s appetite for the relationship. It reduced their ability to sustain one, by raising the farmer’s temptation to sell elsewhere after taking an advance and by cutting the mill’s margin, which was the collateral behind everything the mill had promised. In the first framework’s vocabulary this is a link sliding from relational toward market governance, and the specific assets on both sides stop being worth what they were.

Regulatory and Standards Landscape

The instruments below govern access; conduct inside the firm is Chapter 12’s subject. Section 3 explained how they work; this section states what they are, and what changed in the two years to August 2026.

Global

Two European regulations and one private scheme set the terms most Indian agri-food exporters actually meet. Regulation (EC) No 396/2005 fixes maximum residue levels and supplies the default of 0.01 milligrams per kilogram where no specific level exists and Annex V fixes none. That default does most of the work. Regulation (EU) 2017/625 governs official controls, and the implementing regulation listing origins and products under a heavier regime was last adopted on June 9, 2026, effective June 30, with review at intervals not exceeding six months. Its three tiers run from raised border check frequencies, through a requirement for an official certificate and pre-export analysis on every consignment, to suspension of entry. Above them sits GLOBALG.A.P.’s Integrated Farm Assurance standard at version 6, whose Option 2 group route certifies a producer group as one entity and samples the square root of its member farms.

Trade policy moved in the same window. The European Union and India concluded negotiation of a free trade agreement, and as of August 2026 it is neither signed nor ratified. Its sanitary and phytosanitary chapter reaffirms existing World Trade Organization rights and contains no article on residue limits and none on import tolerances. The agreement in force is the India and United Kingdom comprehensive economic and trade agreement, effective July 15, 2026.³¹

India

The Indian architecture regulates the plot and the facility, and leaves the transaction to the states. Export of a regulated horticultural commodity runs through plot registration, sampling by a recognized laboratory, and packing in a recognized pack-house, administered by the export development authority under the commerce ministry, with state agriculture departments inspecting. The transaction itself sits with the states, and Chapter 1 lists the model instruments available to them. Adoption is voluntary and proceeds state by state, so what is in force at any moment is a state-level question. The adoption counts that have been published are carried in Chapter 2, each with the body that issued it and the date it was issued. Chapter 2 owns the constitutional position and maps the institutions. Chapter 1 sets the chain stages a transaction moves through.

The national electronic market operates as a platform layered over state-regulated market yards and does not replace them, which is why the second case treats it as a channel choice and not as a reform. Chapter 14 takes producer organizations as market-facing institutions, and Chapter 15 the contractual forms through which a firm reaches a smallholder.

SDG Connection

SDG 2 is the goal this chapter serves most directly, and target 2.3 commits to secure and equal access to markets for small-scale food producers and to their productivity and incomes. Access is the operative word here, and the first framework gives it a precise meaning: a producer who cannot meet a codified specification is outside the chain, so market access is a question about entry before it is a question about price. Chapter 4 sets out target 2.3 as a doubling commitment on productivity and income, and the access clause quoted here is the part of that target Chapter 4 does not use.

Decent work and economic growth, Goal 8, arrives here as target 8.3 on micro, small and medium enterprises and their access to financial services, which is where the second framework’s risk ledger bites. The compliance figure in the certification case, roughly $581 per farmer in a first year, is the size of the entry ticket an enterprise of that class is being asked to buy.

Responsible consumption and production, SDG 12, carries target 12.3 on food losses along production and supply chains, and every handover a chain map records is a place those losses occur. Chapter 10 carries the loss arithmetic and the measurement problem underneath it; this chapter contributes only the map on which the losses are located.

Sustainability Decision Box: An aggregator taking on the grading line

The decision. Picture an aggregator moving about 9,000 tonnes of fruit a year from roughly 3,400 growers and selling to processors and wholesalers.³² A grading and packing line would let it sell graded fruit directly to two modern retail buyers at a higher price. The board is treating this as a margin calculation, and the pricing is the least uncertain thing about it.

Economic lens. Functional upgrading is the path with no randomized evidence behind it, so the case has to be built from structure and from adjacent measurement. Two adjacent studies are instructive. A randomized trial across 167 producer organizations found that paying members 30 percent of the final price at delivery raised the price they received by 24 to 26 percent and the share selling through the organization by 24 to 28 points. Telling members what the organization eventually sold for did nothing.³³ A study of coffee found a washed-coffee export premium of 28 to 33 percent that was largely passed through to farmers, and the share of coffee sold to wet mills still reached only 19 percent. The higher-grade route needs 52 percent more harvest labor per hectare and returns 18 percent less income per labor hour.³⁴ Upgrading in both cases turned on working capital and labor allocation. The price was never the binding constraint.

Environmental lens. A grading line creates two physical flows the aggregator does not have: a reject stream and a refrigerated load. Both need a boundary statement before the first season, on the rule Chapter 9 set and Chapter 12 carried into the accounts. The reject stream is a business decision disguised as waste, since fruit failing the grade goes somewhere, and the somewhere is usually the informal channel the aggregator is trying to leave.

Social lens. A grading line formalizes exclusion. Sorting to a written specification decides which of the 3,400 growers are inside the new chain and which are not, and the first framework’s excluded cell arrives as an operating procedure long before it arrives as a policy. The aggregator should know its sort-out rate by grower before it knows its margin, because the answer determines how many of its suppliers the upgrade leaves behind.

Named trade-off and verdict. This buys a margin the aggregator does not currently earn, and it pays with a fixed cost that must be covered in a bad season and with a supply base that shrinks to whoever grades out. The verdict is proceed, staged. Toll the volume through an existing recognized facility for two seasons and take the function without the asset. Measure the actual sort-out rate by grower and the actual volume that clears the specification. Fund working capital first, since the adjacent evidence says that is what moves behavior. Buy the line when the break-even volume has been observed, and not when it has been forecast.

AI in Practice: Matching, and the outcome nobody has measured

The capability offered here is matching: a system that reads offers, bids, grades, locations, and logistics costs and proposes trades, sometimes with a price forecast attached. Chapter 5’s test is the right instrument, and this proposal fails at an unusual place.

The decision changes cleanly, the data exists, and the incumbent is the problem. Naming the decision is easy. Somebody chooses a buyer for each lot every day. The data rung is respectable, because bids, grades, and settlement records are captured and reconciled by the trading operation itself. What defeats the business case is the incumbent, which is a trader with a phone, an advance in cash, and twenty years of knowing who pays, none of which a matching engine reproduces on the day a lot has to move. Section 4 gave the measured record against that incumbent.

Error visibility is the honest reason to be careful. A bad match announces itself at settlement, which is fast feedback and good design. A systematically bad match, meaning one that is slightly worse than what a trader would have found, never announces itself at all, because nobody observes the counterfactual trade. A firm deploying this has to run a holdout: a share of lots allocated the old way, every season, as the only way it will ever know.

Both cautions from Section 4 apply here without change. No evaluated deployment exists as of August 12, 2026, so a firm funding one is funding a first and should price it that way, and the ownership of the data such a system generates will be decided by the terms of service somebody signs.

Practitioner’s Lens: The chain manager at a fruit exporter

Vikram Rathore was put together here and his exporter ships nothing. He was written to give the map, the governance classification, and the risk ledger to somebody who signs a purchase order.

Vikram Rathore buys from about 1,100 registered plots through nine field officers, packs at two recognized facilities, and sells to four European importers and eleven domestic wholesalers. He owns the grower list and the price at every handover, and treats them as one document.

His first rule is that the chain map is a working file and not a slide. It carries the actors at every stage with a count, the tonnage moving between them with the loss at each transfer, and the price at each handover so a margin can be computed on any line. He redraws it after every season and keeps the previous version, which is how he noticed that the number of growers between him and the fruit had risen by two while his margin had not moved. A map that exists in one version is a picture, and a map that exists in six is a record.

His second rule is that every clause in a buyer specification gets traced to the rule behind it. He marks each requirement as law, as scheme, or as preference. Law is not negotiable and he stops arguing about it. Scheme is not negotiable this season and may be next, when the version changes. Preference is a commercial position and he prices it. Three seasons of this produced his most useful finding, that most of what his importers send is law they have restated, and the negotiable part is smaller and more valuable than either side assumed.

His third move is a register of who is one requirement away. Every grower outside the specification is listed with the one thing keeping them out, and each season he funds a few of the cheapest fixes. What he likes least is that large growers clear new requirements faster. What he trusts most is a price series he collected himself.

Applied Exercise: Draw the chain and count the rupee

Objective. Map one chain your organization sells into, compute the share of the final price reaching the producer, and classify the governance of every link.

Time required. Four hours, with your purchase records, your sales records, and one hour of a field officer’s time.

Steps. Draw every stage from producer to final buyer, with a count of actors at each. Write the tonnage in and out of each stage. Write the price at every handover and compute the producer’s share of the final price. Classify each link on the five governance types, naming the evidence for your classification. Then mark the link where the specification originates, because that is the link whose owner decides what everybody upstream of it has to become, and it is rarely the link holding the goods at the moment the specification is written.

Deliverable. One map carrying actors, tonnages, losses, and prices; one line giving the producer’s share as a percentage with the date it was computed; and one paragraph naming the link you would change first and the party who would have to fund the change.

Summary and Bridge

The chapter opened on one grower holding two shares of two consumer rupees, and the smaller share sat on the larger payment. Nothing in a chain is judged on a ratio alone. The first framework classifies a link on three properties of the transaction and returns five governance types, and its most valuable output is the case it refuses to classify, where a producer who cannot meet a codified specification is simply absent, whatever the terms of trade inside the chain happen to be. The second framework adds to those paths the cost of the move and the identity of the party left holding the risk.

Teaching has run well ahead of evidence here. No randomized trial of functional upgrading exists, no meta-analysis pools the return to any single path, and the field’s most rigorous synthesis rests on four randomized trials out of 118 studies. What is measured is that channel upgrading pays well for producers who can enter, that entry is conditioned on endowments, that first-year compliance can cost more than twice the first-year gain, and that digital platforms have so far moved producer prices very little.

Four cases carried it. A regulator built five gates and required a recognized pack-house at gate four. Two electronic platforms showed how an effect size migrates from the market where it was measured to a market where it was not, arriving in a deck as though the two were the same place. A randomized success collapsed when a certification requirement arrived that nobody in the chain would fund. And an extra mill within 10 kilometers dissolved a working relationship without raising a single price.

Part III now turns from the chain to the institutions inside it. Chapter 14, Cooperatives, FPOs and Collective Enterprise, takes the producer-owned firm as the instrument through which growers occupy a stage of the chain themselves, which is functional upgrading attempted collectively. Who funds a move and who can walk away from it arrives there as a question about members.

Endnotes

1. Reserve Bank of India, Department of Economic and Policy Research, Price Dynamics and Value Chain of Fruits in India: A Study of Grapes, Bananas and Mangoes, Working Paper WPS (DEPR): 06/2024, October 2024, for the grape grower’s 35 percent share of the domestic consumer price and 21 percent of the export consumer price, for the higher absolute price in the export channel, for the finding that 50 to 60 percent of production sorts to export grade, and for the farmers’ shares quoted in Section 1 for bananas and mangoes. Method: focus group discussions and interviews with traders, grower-exporters, and farmers in the Nashik belt, January 2023. The 35 percent is traced on Thompson Seedless into the Azadpur wholesale market in Delhi and the 21 percent on Sharad Seedless into the Netherlands, so the two shares describe two chains observed in one study, and no single grower was quoted for both. The share for gram is from the companion paper in the same October 2024 release, Reserve Bank of India, Department of Economic and Policy Research, Pulses Inflation in India, Working Paper WPS (DEPR): 07/2024, and the share for milk from Working Paper WPS (DEPR): 05/2024 in the same release.

2. On chain mapping method, four published guides: Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ) GmbH, ValueLinks 2.0: Manual on Sustainable Value Chain Development, by Andreas Springer-Heinze, January 2018, in eleven modules, for the eight generic elements of a value chain map, five of which constitute the basic map at micro level; the European Commission’s Value Chain Analysis for Development methodology, funded by its Directorate-General for International Partnerships and implemented by the Agrinatura consortium, framing more than sixty-five country studies since 2016 on its four questions; Making Markets Work Better for the Poor (M4P) Project, Making Value Chains Work Better for the Poor: A Toolbook for Practitioners of Value Chain Analysis, Agricultural Development International, Phnom Penh, 2008, funded by the UK Department for International Development, for the eight tools and for income distribution and employment distribution as Tools 6 and 7; and Food and Agriculture Organization of the United Nations, Developing Sustainable Food Value Chains: Guiding Principles, Rome, 2014. The five elements this chapter requires are the book’s own compilation from the four.

3. J. Yi, E.-M. Meemken, V. Mazariegos-Anastassiou, J. Liu, E. Kim, M. I. Gómez, P. Canning and C. B. Barrett, “Post-farmgate food value chains make up most of consumer food expenditures globally,” Nature Food 2, no. 6 (2021): 417–425, doi:10.1038/s43016-021-00279-9, for the farm share of 27 percent on average, the 16 to 38 percent range for middle-income and high-income countries, and the finding that the share falls as incomes rise. Method: input-output accounting standardized across sixty-one countries for 2005 to 2015, covering approximately 90 percent of global output.

4. Reserve Bank of India Bulletin, “Supply Chain Dynamics and Food Inflation in India,” October 11, 2019, for the survey of 9,403 respondents across 85 markets in 16 states and 16 crops, and for the 28 to 78 percent range of farmers’ shares of retail price.

5. Gereffi, Humphrey and Sturgeon, “The governance of global value chains,” cited in full in the body of Framework 1, doi:10.1080/09692290500049805, for the three determining variables, the five governance types, and the exclusion logic. The three variables and the eight-combinations passage are paraphrased here and not quoted. The exclusion reading is restated by Ponte, Sturgeon and Dallas, cited below. Any four-type rendering of the typology is a secondary simplification and this chapter does not use one.

6. An independent multi-disciplinary review reaches the same position on the framework’s standing: L. Kano, E. W. K. Tsang and H. W. Yeung, “Global value chains: A review of the multi-disciplinary literature,” Journal of International Business Studies 51, no. 4 (2020): 577–622, doi:10.1057/s41267-020-00304-2, characterizing the field’s direction as extension of the 2005 typology. That the field has extended the 2005 typology and proposed no replacement for it is this chapter’s own search finding, dated August 12, 2026, and it is no statement of that review’s.

7. S. Ponte and T. Sturgeon, “Explaining governance in global value chains: A modular theory-building effort,” Review of International Political Economy 21, no. 1 (2014): 195–223, doi:10.1080/09692290.2013.809596, for the observation that most value chains contain a range of linkage types and that characterizing a whole chain from inter-firm linkage research alone is risky, and for the micro, meso, and macro levels with polarity at the macro level.

8. S. Ponte, T. Sturgeon and M. Dallas, “Governance and power in global value chains,” in S. Ponte, G. Gereffi and G. Raj-Reichert, eds., Handbook on Global Value Chains, Edward Elgar, 2019, 120–137, for the two dimensions of transmission and arena and the four resulting power types: bargaining, demonstrative, institutional, and constitutive. The reading that sustainability pressure in agri-food chains travels institutionally and constitutively is this chapter’s application, not the authors’ claim.

9. On the four upgrading paths, this chapter cites both candidate origins and asserts neither: Governance and Upgrading: Linking Industrial Cluster and Global Value Chain Research, IDS Working Paper 120, Institute of Development Studies, Brighton, 2000, by J. Humphrey and H. Schmitz; and J. Humphrey and H. Schmitz, “How does insertion in global value chains affect upgrading in industrial clusters?”, Regional Studies 36, no. 9 (2002): 1017–1027, doi:10.1080/0034340022000022198. A peer-reviewed review of the field, P. Gehl Sampath and B. Vallejo, “Trade, Global Value Chains and Upgrading: What, When and How?”, The European Journal of Development Research 30, no. 3 (2018): 481–504, doi:10.1057/s41287-018-0148-1, sets out the development of the taxonomy across both sources.

10. On the absence of causal evidence for functional upgrading: the dated search in Section 2 reached no such trial and no meta-analysis pooling the economic return to any single upgrading path. Section 2 depends on that absence, which is why the date is printed beside it.

11. L. S. O. Liverpool-Tasie, A. Wineman, S. Young and others, “A scoping review of market links between value chain actors and small-scale producers in developing regions,” Nature Sustainability 3, no. 10 (2020): 799–808, doi:10.1038/s41893-020-00621-2, for the 12,320 records screened, the 202 studies included, the 83 percent positive on at least one outcome, and the authors’ observation that few studies examine a primary outcome alongside an intermediate one. The warning against reading that share as a success rate is this chapter’s.

12. S. K. Malhotra, S. Mantri, N. Gupta, R. Bhandari, R. N. Armah, H. Alhassan, S. Young, H. White, R. Puskur, H. S. Waddington and E. Masset, “Value chain interventions for improving women’s economic empowerment: A mixed-methods systematic review and meta-analysis,” Campbell Systematic Reviews 20 (2024): e1428, doi:10.1002/cl2.1428, for the 118 included studies of which four are randomized trials, for the finding that effects on income, assets, productivity, and savings are small and limited by low confidence in methodological quality, and for the authors’ statement that benefits accrue more through skill acquisition and improved inputs than through market access improvements. No pooled effect size is printed here.

13. E. J. O. Rao and M. Qaim, “Supermarkets, Farm Household Income, and Poverty: Insights from Kenya,” World Development 39, no. 5 (2011): 784–796, doi:10.1016/j.worlddev.2010.09.005, for the 48 percent gain in average household income associated with participation in supermarket channels, estimated by endogenous switching regression correcting for self-selection. H. C. Michelson, “Small Farmers, NGOs, and a Walmart World: Welfare Effects of Supermarkets Operating in Nicaragua,” American Journal of Agricultural Economics 95, no. 3 (2013): 628–649, doi:10.1093/ajae/aas139, for the increase in household productive assets and for the finding that participation is largely confined to farmers with advantageous endowments of geography and water.

14. S. Ponte and J. Ewert, “Which Way is ‘Up’ in Upgrading? Trajectories of Change in the Value Chain for South African Wine,” World Development 37, no. 10 (2009): 1637–1650, doi:10.1016/j.worlddev.2009.03.008, for the reframing of upgrading as reaching a better deal assessed on rewards net of risk, and for the trajectory findings Section 2 reports. This chapter takes the reframing.

15. The Upgrading Trajectory Map originates here. The four paths are taken from the sources at n. 9, the reward test from n. 14, and the risk ledger appears in no source. Chapter 9 supplies the process arithmetic and the boundary rule the map depends on, and Chapter 12 carries that rule into the management accounts.

16. Regulation (EC) No 396/2005 of the European Parliament and of the Council on maximum residue levels of pesticides in or on food and feed of plant and animal origin, for the residue regime and for the default of 0.01 milligrams per kilogram at Article 18(1)(b), which applies where no specific level is set and where Annex V sets no different default for the substance. It remains the governing instrument as of August 12, 2026, and the article number is given as it appears in the published text.

17. Regulation (EU) 2017/625 on official controls, and Commission Implementing Regulation (EU) 2019/1793 listing products and origins subject to a temporary increase in official controls, emergency measures, or an official certificate requirement, as amended. The amending regulation current as of August 2026 is Commission Implementing Regulation (EU) 2026/1206, adopted on June 9, 2026, published in the Official Journal on June 10, 2026, and applicable from June 30, 2026, and the instrument requires review at intervals not exceeding six months. This chapter names no country’s products on any annex of it.

18. GLOBALG.A.P., Integrated Farm Assurance standard version 6, in the editions dated at Chapter 1, endnote 24, and the scheme’s General Regulations for the Option 2 group certification route and the square-root sampling rule applied to member farms. An entry-level route sits below full certification under the scheme’s Primary Farm Assurance product, which the scheme itself describes as a capacity-building step and not an accredited certification valid for global trade. Everything stated here is descriptive and taken from the scheme’s own published documents.

19. M. Fafchamps and B. Minten, “Impact of SMS-Based Agricultural Information on Indian Farmers,” The World Bank Economic Review 26, no. 3 (2012): 383–414, for the absence of significant effects on prices received, storm losses, crop choice, and cultivation practices, and for the take-up and non-use figures. Chapter 5 carries the full citation and the surrounding randomized literature on digital advisory.

20. V. Couture, B. Faber, Y. Gu and L. Liu, “E-Commerce Integration and Economic Development: Evidence from China,” National Bureau of Economic Research Working Paper 24384, first issue, March 2018, which the bureau now lists under the title “Connecting the Countryside via E-Commerce: Evidence from China”, for the randomized rollout of rural e-commerce infrastructure across 100 villages, the real gain of roughly 5 percent in the cost of living among households using the terminal and about 1 percent for the average household in a treatment village, and the absence of measurable producer gains. A later journal version appeared as American Economic Review: Insights 3, no. 1 (2021): 35–50, doi:10.1257/aeri.20190382, and is not the version cited here.

21. M. Sauvagerd, M. Mayer and M. Hartmann, “Digital platforms in the agricultural sector: Dynamics of oligopolistic platformisation,” Big Data & Society 11, no. 4 (2024), doi:10.1177/20539517241306365, for the concentration of the underlying cloud infrastructure market at over 97 percent among a small number of providers as of 2023, and for the general absence of profitability among sector-specific agricultural platforms. No provider and no platform is named in this chapter.

22. S.-L. Ruder and H. Wittman, “Agricultural data governance from the ground up: Exploring data justice with agri-food movements,” Big Data & Society 12, no. 1 (2025), doi:10.1177/20539517251330182, for the finding that ownership, access, and control of agricultural data are left to contractual agreement in the jurisdiction studied, with no legal definition of agricultural data. Chapter 5 carries the Indian statutory position under the Digital Personal Data Protection Act, 2023, which contains no agricultural provision.

23. On artificial intelligence in chain matching, routing, and price discovery: no independently evaluated deployment carrying measured business outcomes could be located anywhere in the published record as of August 12, 2026. This chapter prints no performance figure for any such system and rests the AI-in-Practice callout on that absence.

24. Agricultural and Processed Food Products Export Development Authority, Procedures for Export of Fresh Table Grapes to the European Union, trade notice APEDA-QCT/13/2024-27, February 17, 2026, governing the 2025–26 season, for every requirement in the first case: the six-component plot registration number, the 1 hectare cap extendable to 1.2, three-year validity with renewal between September 1 and December 31, display of the number on the farm, spray records, two state inspections with the second within twenty days of sampling, sampling of at least 5 kilograms per hectare from fifteen primary spots into two boxes, delivery within twenty-four hours, the certificate within six days, the list of 163 agrochemicals set annually by the National Referral Laboratory at the ICAR-National Research Centre for Grapes, Pune, export only through a recognized pack-house, retention of samples for sixty days at 0 to 1 degree Celsius and 90 to 95 percent relative humidity, and the destination scope. The suffix in the notice number is a file series and not a season.

25. The absence stated in the India Case was verified against the instrument at n. 17 on August 12, 2026. This chapter states that fact and states no country’s listings.

26. Figures for India’s national agriculture market are as reported at March 2026 by the Ministry of Agriculture and Farmers Welfare: 1,656 integrated market yards, about 1.80 crore registered farmers, and cumulative trade since 2016 of about ₹4.84 lakh crore. Chapter 1, endnote 7 reads the same series a month earlier and at a slightly lower cumulative figure. Rupee figures are glossed at an indicative ₹84 to the dollar. The conversions are the author’s.

27. R. Levi, M. Rajan, S. Singhvi and Y. Zheng, “The impact of unifying agricultural wholesale markets on prices and farmers’ profitability,” Proceedings of the National Academy of Sciences 117, no. 5 (2020): 2366–2371, doi:10.1073/pnas.1906854117, for the modal price increases of about 5.1 percent for paddy, 3.6 percent for groundnut, and 3.5 percent for maize. Those estimates were measured on the Karnataka Unified Market Platform and on no other system. The correction the second case makes is about attribution and carries no claim about any platform’s performance.

28. That growers in the Maharashtra grape belt sell an estimated 85 to 90 percent of output to informal buyers comes from A. Gaikwad and T. Schrader, Mapping domestic grape value chains in Maharashtra, India: The connector roles and price buffering effects of informal local buyers, Wageningen Centre for Development Innovation, Report WCDI-24-403, November 2024, doi:10.18174/683629, which also traces a domestic per-kilogram chain from a farm gate of ₹30 to a consumer price of ₹60.

29. The randomized trial in the third case is cited in full at Chapter 8, endnote 16: N. Ashraf, X. Giné and D. Karlan, in the American Journal of Agricultural Economics, 91, no. 4 (2009): 973–990, doi:10.1111/j.1467-8276.2009.01319.x. It supports the cluster-randomized design across 36 self-help groups in three arms, the arm sizes, the April 2004 baseline and May 2005 follow-up with 86 percent retention, every intent-to-treat effect reported in the third case, the epilogue, and the compliance-cost figures, which that paper draws from earlier work by Asfaw, Mithöfer and Waibel. That conclusion is the authors’ own.

30. R. Macchiavello and A. Morjaria, “Competition and Relational Contracts in the Rwanda Coffee Chain,” The Quarterly Journal of Economics 136, no. 2 (2021): 1089–1143, doi:10.1093/qje/qjaa048, for the mill census and matched farmer survey for the 2012 harvest, the three components of the relational contract, the instrument built from an engineering model of mill siting scored from aerial imagery, the instrumented 0.283 standard deviation effect of an additional mill within 10 kilometers against the 0.116 simple correlation, the two channels the study separates, the capacity and unit cost findings, the reduced share of cherry sold to mills, the absence of any price increase, and the fall in farmer satisfaction.

31. On trade policy: negotiations concluded January 27, 2026; status verified as neither signed nor ratified on August 12, 2026. Its sanitary and phytosanitary chapter reaffirms rights and obligations under the World Trade Organization Agreement on the Application of Sanitary and Phytosanitary Measures and contains no article on maximum residue levels and none on import tolerances. The India and United Kingdom Comprehensive Economic and Trade Agreement entered into force on July 15, 2026.

32. The aggregator in the Sustainability Decision Box, Vikram Rathore in the Practitioner’s Lens, and his exporter were built for teaching and have no counterparts. Their tonnages, grower counts, plot counts and buyer counts were chosen to make the instruments visible. Every study, instrument, and figure surrounding them is real and is cited above. Wording for targets 2.3 and 12.3 is cited in full at Chapter 4, endnote 21; wording for target 8.3 is cited at Chapter 7, in the SDG Connection there.

33. R. V. Hill, E. Maruyama, M. Olapade and M. Frölich, “Strengthening producer organizations to increase market access of smallholder farmers in Uganda,” Agricultural and Resource Economics Review 50, no. 3 (2021): 436–464, for the randomized design across 167 producer organizations, the cash-on-delivery treatment paying members 30 percent of the final price at delivery, the resulting 24 to 26 percent price increase and 24 to 28 percentage point increase in the probability of selling through the organization, and the null result on the information treatment.

34. S. Tamru and B. Minten, “Value addition and farmers: Evidence from coffee in Ethiopia,” PLOS ONE 18, no. 1 (2023): e0273121, doi:10.1371/journal.pone.0273121, for the washed-coffee export premium of 28 to 33 percent, its substantial pass-through to farmers, the 19 percent adoption share, the 52 percent higher harvest labor requirement per hectare and the 18 percent lower income per labor hour for red-cherry sellers. Sample: 1,600 households across five coffee zones, February 2014.

Key Terms

Value chain. The full sequence of activities that brings a product from input supply to final consumer, with the actors performing each and the margin each takes. It differs from a supply chain in the question it asks, since a supply chain asks how goods move and a value chain asks where the money stops. Chapter 6 uses the term in Porter’s narrower sense, for the activities inside one firm.

Chain governance. The way terms are set in a link of a chain, classified on the complexity of the transaction, whether it can be codified, and whether suppliers can meet the specification. It is a property of a link and not of a whole chain, which is the qualification most secondary accounts drop.

Captive coordination. A link in which small suppliers depend transactionally on a much larger buyer and face significant switching costs. It can be created commercially, and in regulated export chains it is often created by a rule that funnels many producers through a small number of approved facilities.

Exclusion. The outcome when a transaction is simple and fully codified and supplier capability is low. It is the case the governance framework refuses to classify, and it matters because improving the terms of trade does nothing for a producer to whom the terms of trade do not apply.

Polarity. The concept for governance at the level of a whole chain: unipolar where one actor sets terms throughout, bipolar where two do, multipolar where terms are contested. A manager asking who governs an entire commodity chain is asking about polarity and will not be answered by the five link-level types.

Process upgrading. Producing the same output more efficiently, through yield, loss, or conversion gains. It is the cheapest path and the most exposed to capture, since a buyer who learns of the saving can ask for it at the next negotiation.

Product upgrading. Producing a better or differently specified output, usually by funding a compliance investment before any premium is earned. Its risk is that the investment is specific to one market and has little resale value if that market closes.

Functional upgrading. Taking on a stage of the chain somebody else used to perform, such as grading, packing, processing, or export. It converts a variable cost into a fixed one, so its correct instrument is a break-even volume, and it is the upgrading path with no randomized evidence behind it.

Chain upgrading. Moving an existing capability into a different chain. Almost nothing transfers except the physical asset and the workforce, both of which were configured for the chain being left, so it should be appraised as entry into a new business.

Farm share. The proportion of final consumer expenditure that reaches the producer, on standardized input-output accounting across sixty-one countries. It averages about 27 percent globally for food eaten at home and falls as an economy gets richer, and it must always be read beside the absolute price, because a smaller share of a larger number can be the better outcome.

Maximum residue level. The highest concentration of a pesticide residue legally tolerated in a commodity placed on a given market. Where no specific level is set, the European default of 0.01 milligrams per kilogram applies under Article 18(1)(b), subject to any different default fixed in Annex V.

Group certification. Certification of a producer group as a single entity with an internal quality management system, with the certifier sampling the square root of the member farms. That sampling rule is what makes certifying hundreds of small producers cost something other than hundreds of times certifying one.

Discussion Questions

1. Draw the chain for one product your organization buys or sells, and compute the producer’s share of the final price. Then state what the absolute producer price is in each channel available, and say which of the two numbers should drive the channel decision.

2. Classify each link in that chain on the five governance types, and name the evidence behind each classification. Where two links have different types, say what happens at the boundary between them.

3. Identify the codified specification in your chain that determines who is inside it. Estimate how many of your current suppliers would be excluded if that specification tightened by one requirement, and say who would fund the fix.

4. Your firm is considering functional upgrading into a stage a buyer currently performs. Write the break-even volume, the fixed cost it must cover in a bad season, and the resale value of the asset if your largest buyer leaves.

5. Take one buyer specification you currently receive and mark each clause as law, as scheme requirement, or as commercial preference. Report how much of it turned out to be negotiable.

6. A vendor proposes a matching platform that will connect your producers directly to buyers. Design the holdout that would let you know in two seasons whether it beat your existing traders, and state what you would measure.

7. Your chain’s compliance requirement costs a producer more in year one than the producer gains in year one. Using the financing instruments Chapter 8 sets out, name three ways that gap could be closed, say who bears the risk under each, and state which one you would propose to your board.