The direct-trade math: why AI-matched sourcing beats broker-mediated for specialty roasters

A specialty roaster costing out a new green-coffee contract usually starts from a landed price and works backward, trying to guess what happened to the value between the farm gate and the port. Most of that guessing is unavoidable, because most of the chain between origin and roastery is opaque by default — not through any single actor's bad faith, but because a multi-hop trade structure was never built to be legible to the buyer at the end of it. Understanding where that opacity sits, and what it costs, is the actual case for a direct, AI-matched sourcing model — not a promise of a lower number on an invoice, but a shorter, more visible chain to reason about.

The chain behind a landed price

A conventional lot of specialty-grade coffee or a botanical ingredient can pass through several hands before it reaches a roaster or formulator: a farm or cooperative, a local buying agent or wet-mill, an exporter, an importer, and sometimes a broker sitting between the importer and the buyer. Each of those hands performs a real function — aggregation, quality sorting, financing, logistics, documentation — and each one is entitled to be paid for it. The question a buyer rarely gets a straight answer to is not whether intermediaries add value, but how much of the price they're paying reflects value added versus the simple fact that information had to move through that many parties to reach them at all.

Where information asymmetry actually lives

Information asymmetry in a broker-mediated chain runs in both directions, and that is easy to miss when a buyer only feels their own side of it.

• A producer or cooperative usually cannot see what a roaster in Europe or North America is willing to pay for a differentiated lot — cup score, origin story, and compliance evidence attached. They see the price the next buyer up the chain offers, which is a compressed signal of everything downstream demand actually looks like.
• A buyer usually cannot see the cost structure at origin — what a producer actually received, what a mill charged for processing, what financing or pre-payment cost the exporter along the way. They see a quoted landed price, which is a compressed signal of everything that happened upstream.
• Neither side can verify the other's compressed signal without cost, so both default to trusting the intermediary in the middle to have priced fairly — which is a reasonable default, but not the same thing as a transparent price.

That double compression is what makes margin stacking hard to audit from either end. It is not that anyone is hiding a number; it is that no single participant in a long chain has visibility into the full chain, so no one is well positioned to tell a buyer what a lot "should" cost.

Where margin enters a layered chain

It is worth reasoning through this qualitatively rather than reaching for a number, because any specific percentage would be presented as if it were measured across the industry, and it isn't — margins vary by origin, crop year, exchange rate, and the specific relationships involved. But the structural point holds regardless of the exact figures: every intermediary layer between farm gate and roastery is a place where a margin gets taken, and in a chain with several such layers, those margins compound rather than simply add.

Suppose, purely as an illustration, that a local buying agent takes a margin to cover aggregation and transport, an exporter takes a margin to cover milling, grading, and export documentation, an importer takes a margin to cover financing, warehousing, and quality control on arrival, and a broker takes a margin to cover the relationship and the deal itself. None of those margins is unreasonable in isolation — each corresponds to a real cost or a real service. But a buyer paying a single landed price at the end of that chain is paying all of them at once, stacked, without a clean way to see which layer contributed what. A buyer who could see the chain broken into its parts might still choose to pay for all of it — full-service intermediation is a legitimate model for many buyers, particularly ones without the capacity to manage origin relationships directly. The problem is not that intermediation exists; it's that most buyers never get to see the breakdown well enough to make that choice deliberately.

The compliance bill that arrives late

A newer cost has entered this picture with the EU Deforestation Regulation, and it tends to surface at the worst possible point in a broker-mediated chain: after a shipment is already committed, when a buyer discovers that the geolocation and origin documentation needed for a due-diligence statement either doesn't exist for that lot or has to be reconstructed after the fact.

In a long chain, traceability data degrades with every handoff unless someone deliberately preserves it. A cooperative's own farm-plot records may be solid, but by the time a lot has passed through a mill, an exporter's consolidated container, and an importer's warehouse, the link back to the specific plot of origin can thin out or vanish entirely. The buyer who eventually has to file a due-diligence statement is the one left holding that gap, typically at the point furthest from where the gap could have been prevented — which is exactly backward from where compliance cost is cheapest to absorb. Data captured once, at origin, at the time of harvest, is inexpensive. Data reconstructed after a shipment has already changed hands multiple times is not, and sometimes it simply cannot be reconstructed to a standard that satisfies a due-diligence statement at all.

What changes in a direct, AI-matched model

This is the actual case for direct trade specialty coffee AI matching, stated plainly rather than as a pitch: it does not promise to make any given layer of intermediation disappear by decree, and it does not promise a specific savings figure. What it changes is where information sits and who can see it before a deal is struck.

1. 1Matching runs on structured lot attributes — origin region, cupping data, certifications, EUDR compliance status — rather than on which broker happens to have a relationship with which buyer. That widens the set of buyers who can find a given lot on its evidence, and the set of lots a given buyer can find on their criteria, independent of an existing personal network.
2. 2Traceability and compliance documentation are attached to the batch at origin, at verification time, rather than assembled retroactively when a shipment is already at the border. The cost of gathering that evidence is paid once, early, by whoever is closest to the source of it — instead of paid late, expensively, by whoever is holding the shipment when a regulator asks for it.
3. 3Producer identity stays anonymized until a request-for-quote is accepted, so early comparison happens on the lot's own evidence — quality, certifications, coarse region, compliance tier — rather than on relationship or reputation. That is a deliberate design choice: it lets a buyer evaluate a lot on its merits before either side commits, which is a different kind of information symmetry than a chain built on introductions.
4. 4Fewer intermediary handoffs between the party who grew or aggregated the lot and the party who is buying it means fewer places where a margin is taken without the buyer seeing which layer took it. That does not guarantee a lower total price on any individual contract — quality, volume, and market conditions still set the price — but it does mean more of what a buyer pays is visible as either origin value or platform service, rather than disappearing into an unstated chain of markups.

Direct trade doesn't promise a smaller number on the invoice. It promises fewer places where the number gets bigger without you being able to see why.

What direct trade does not remove

None of this is a case for direct-only sourcing across the board, and it would be dishonest to present it that way. A few things are worth being explicit about.

• Aggregation, financing, and logistics still have to happen somewhere. A direct model shifts who performs some of that work and how visible the cost of it is; it does not make the underlying work free. A buyer sourcing direct from many small lots may find they are now doing coordination work an importer used to absorb.
• Due diligence is a buyer obligation either way. A verified lot with strong compliance evidence gives a buyer a better starting point, but under EUDR the operator placing goods on the EU market remains the one who has to file the due-diligence statement, run their own risk assessment, and apply mitigation where it's called for. Better evidence at the input does not discharge that responsibility.
• A fair-price band shown alongside a lot on this marketplace is an AI-generated estimate, produced using Claude (Anthropic), built from the lot's attributes and available market context. It is a reference range to reason from, not a live market quote, not a trained pricing model, and not a guarantee of what a lot will actually transact at. Treat it the way you'd treat any well-informed estimate: useful for framing a negotiation, not a substitute for one.
• Smaller producers and smaller buyers may have less negotiating leverage without an intermediary who historically absorbed some of that risk on their behalf — a factor, invoicing, or credit relationship a broker or importer sometimes provided is not automatically replaced by a matching platform.

Reading a verified lot with this lens

The practical use of all this for a buyer evaluating a specific lot is less about chasing a theoretical savings number and more about asking a sharper set of questions than a landed-price quote usually invites. Where did the traceability data originate, and how far back does it actually reach? Is the compliance tier attached to the batch itself, or inferred from a producer's general reputation? Is the price band, if shown, clearly labeled as an estimate, and does it come with the underlying attributes that produced it? A buyer who can answer those questions about a lot is a buyer who has more of the chain visible to them than a landed price alone would ever reveal — regardless of how many hands the lot ultimately passes through before it ships.

That is the honest version of the direct-trade pitch: not a discount, but a shorter distance between the evidence and the person deciding whether to buy on it. For a specialty roaster or ingredient buyer weighing a new contract, that distance is worth pricing in on its own terms.

Verified lots on the marketplace carry their compliance tier, certifications, and coarse region up front, with full producer identity revealed only after an RFQ is accepted. Browse the verified listings to see what that evidence looks like on an actual lot, or request buyer access to start comparing against your current sourcing chain.