The Economics of Completing an Order

In March 2026, Worcester, Massachusetts, asked suppliers to quote four sizes of water-main coupling, with estimated quantities of two each. A week later, the city corrected two part numbers. Bidders had to use the revised pricing sheet, provide manufacturer and test documentation, and include delivery in a firm price. The city lists five bidders and names Core & Main as the awardee.[1]

Four lines of physical goods required a much longer chain of commercial decisions. A seller had to understand the requirement, obtain usable supply terms, follow the amendment, calculate its exposure, and decide what it could responsibly promise. The work began before the seller knew whether it would earn a dollar.

Applied Commerce is being built around that work. We intend to become an AI-native merchant that bids on and fulfills physical-goods purchases for U.S. governments and public institutions. We will source products, compete for orders, and take responsibility for the delivery we agree to provide. Our customer will buy goods. Our software and agents will help us operate the business.

At Colos, I first found prospective buyers and then requested supplier quotations against their requirements. I am bringing that demand-first discipline to Applied Commerce, together with my experience as a software engineer. The question is how much of the work needed to make a transaction executable can become repeatable software.[2]

Demand first does not mean an order is secured. A competitive government purchase still has to be won, contracted, and fulfilled. But it gives the company a concrete starting point: a buyer's stated need, against which every proposed action can be judged.

Our thesis is that reducing the information work around a transaction could change which orders a merchant can economically serve. The merchant must remain accountable for the result.

The work around the goods

The price of an item is only one part of the cost of supplying it.

Someone must distinguish a required manufacturer from an illustrative reference, check quantities and units, confirm delivery conditions, and obtain a supplier quotation that remains usable through the purchasing process. A source price without freight, availability, or valid resale terms may be insufficient to support a binding customer commitment. After selection, the seller still has to place the purchase, coordinate shipment, resolve discrepancies, invoice correctly, and collect payment.

Much of this effort attaches to the transaction rather than the number of units. A short order can require nearly as many decisions as a larger one. If the dollars remaining after merchandise and delivery cannot support that effort, a supplier has a reason to decline the opportunity, charge more, or concentrate on accounts it can serve repeatedly.

That does not establish that public buyers lack capable suppliers. Worcester attracted five bidders. Core & Main describes an operation built on supplier relationships, municipal-specification knowledge, local inventory, and technology supporting quotation and delivery. It also attributes favorable purchasing terms and product access to its scale. These are substantial commercial advantages.[3]

We should expect established distributors to adopt better agents, too. Applied Commerce will have to compete against experienced suppliers using modern tools.

The opening we see is therefore conditional and specific. Some transaction classes may contain enough repeatable information work for a different operating structure to improve their economics. That could support a more competitive offer on an ordinary purchase, or make a smaller, less familiar order worthwhile to pursue.

The central constraint is the amount of commercial attention an order can afford.

A different operating structure

Software has long recorded orders, calculated prices, and moved structured information. People often supply the missing continuity: interpreting an unfamiliar document, reconciling a vendor's response, noticing a changed requirement, and deciding which action should happen next.

By 2026, developers can build agents that use tools across multiple steps and preserve a record beyond a single interaction. Anthropic's April 2026 description of Managed Agents, for example, separates durable session history from the software directing the agent and the environments in which it acts. That is useful infrastructure for work that must resume when new information arrives.[4]

Our inference is that more of a merchant's digital coordination can now be designed as executable work. An agent could extract requirements, prepare supplier inquiries, compare responses, identify missing terms, and monitor a shipment. The commercial test is whether the complete process becomes less expensive while remaining accurate. Independent evaluation cautions that performance on software tasks does not establish reliability across other kinds of work.[5]

We intend to organize the company around a persistent transaction record. It will retain the original requirement and amendments, supplier quotations, cost assumptions, approvals, purchase documents, delivery evidence, and payment status. When a part number changes, the system should identify which matching decisions and draft offers require review. It should preserve uncertainty until someone resolves it.

Ordinary code should perform arithmetic and reconcile quantities, units, dates, and document versions. Agents can interpret and prepare work. People must approve consequential commitments, resolve ambiguous specifications, and decide whether the business can accept an obligation. Suppliers must confirm availability. Carriers must move the goods. Capital must fund the payment gap.

This architecture places routine work and its supporting evidence in one operating process. People should spend less time reconstructing what happened between inboxes and more time making decisions that require judgment. That is a design objective, not a description of a deployed ACC system.

A merchant also has a direct reason to care about the whole result. A faster quotation has little value if it produces the wrong purchase. A cheap shipment is no bargain if it violates the delivery promise. By owning the transaction, we intend to make selection, pricing, execution, and error costs answer to the same commercial outcome.

The threshold for a worthwhile order

The economics become clearer when we count the work spent on orders we never win.

Consider an illustrative $12,000 order for a basket of maintenance goods. This is a constructed example, unrelated to the Worcester procurement. Assume merchandise costs $10,200 and the other costs below total $480. Those amounts are modeling assumptions, not supplier quotations, industry averages, or ACC results.

Assume that one in ten comparable submitted quotations becomes a completed, paid order. At that rate, each completed order carries the average pursuit cost of ten quotations, including the unsuccessful pursuits. Screening opportunities that we reject must also be allocated to the quotations we pursue.

For the human-led illustration, assume $90 of fully costed pursuit work per submitted quotation. For the proposed agent-assisted process, call that cost x. It remains unknown. Both figures must include active human work, supervision, retries, software and model spending, and allocated screening. Waiting for a supplier is elapsed time, not automatically labor saved.

The $120 post-award allowance comprises $100 of human fulfillment and review work and $20 of tools. The loss allowance must also cover net unrecovered costs of awarded orders that fail to complete or collect; their pursuit costs are counted separately. All allowances require transaction-specific estimates before an actual commitment.[6]

Under these assumptions, the proposed process produces positive transaction contribution only if x is below $132. It improves on the illustrative human-led process only if x is below $90.

The general relationship is simple: contribution per completed order equals what remains after merchandise and post-award costs, minus pursuit cost divided by the rate at which quotations become completed, paid orders. A lower conversion rate makes every dollar of pursuit more expensive. More generated bids can therefore make a company worse if their selection quality falls.

This also explains how the minimum viable order size could change. A smaller order generally produces fewer dollars to cover work that still has to happen. Reducing that work lowers the contribution required to justify pursuing the order, provided purchasing terms, fulfillment quality, and conversion remain comparable. The threshold falls only as far as the remaining costs allow. Freight minimums, technical review, or service obligations may still make a small order uneconomic.

The economic insight is that a winning order has to pay for the work of losing, too. At the assumed conversion rate, removing $10 of real pursuit cost releases $100 per completed order. Paying an additional $100 for the merchandise consumes the same amount. Superior purchasing terms can outweigh superior automation.

Working capital imposes a separate limit. If suppliers require payment before the institution pays us, cash remains tied up even when contribution is positive. Faster document preparation cannot finance that gap. Supplier credit, customer acceptance, payment timing, and funded exposure must be understood before the commitment is made.

Nor is contribution company profit. It must ultimately support engineering, management, insurance, and other fixed costs. The full thesis requires reliable execution and enough repeated business to carry the organization.

Building the supplier

Government and public-institution purchasing is our proposed starting market because some buying channels expose a concrete requirement before a supplier commits resources. Specifications, deadlines, purchasing conditions, and historical results can give us a defined commercial task to evaluate. The Worcester record illustrates that starting point; it does not establish a market of unserved buyers.

Access still has to be earned and verified. Public visibility does not establish eligibility to bid. A quotation does not necessarily create a contract. The applicable purchasing method determines the legal sequence, and the transaction record must preserve it. Buyers may also use existing contracts: Sourcewell, for example, enables eligible institutions to purchase through competitively solicited supplier agreements, subject to applicable requirements.[7]

Our initial focus will be a narrow, repeatable purchasing pattern involving goods with verifiable identifiers, legitimate supply routes, manageable shipping, and limited installation or service obligations. Maintenance, facilities, plumbing, and waterworks products are starting candidates. Actual supplier terms and transaction economics will determine where we begin.

For each opportunity, we intend to establish eligibility, confirm the specification, obtain actual supplier terms, and calculate the delivered economics and cash exposure. An authorized person will approve the price and commitment. If selected and properly contracted, Applied Commerce will arrange supply, track delivery and acceptance, resolve problems, invoice, and collect. The transaction is only economically understood when its actual costs and cash outcome are known.

Procurement search and bid-writing software can support this work. Buyer-side agents can help institutions decide what to purchase. We intend to be the seller that stands behind the goods. Traditional distributors already carry this responsibility; the difference we seek is a lower cost of coordinating and completing a defined class of orders.

Applied Commerce is at inception and pre-revenue. We have no verified government awards, completed orders, supplier contracts, profitable transactions, or measured agent productivity. Those are the results this operating model must earn.

A company that learns by delivering

The first valuable asset would be a repeatable transaction that works: eligible demand, competitive supply, reliable fulfillment, and positive contribution after the cost of pursuit. The next would be evidence that completing it makes another transaction easier.

A repeat buyer could reduce the work of understanding receiving requirements and permitted purchasing routes. A dependable supplier relationship could improve response quality, access, or payment terms. A verified product history could reduce matching errors. Actual delivery and collection records could improve how we price risk and allocate capital. Each advantage would have to be earned through performance.

Commercial memory matters when it changes the next decision. A stored quotation is useful if it helps identify an appropriate source, but its price and validity still need checking. A recorded exception is useful if it prevents a repeated error. More data alone does not confer an advantage; better selection and execution might.

Expansion should follow that reuse. An adjacent product for a familiar institution may share a supplier, approval process, or delivery pattern. A similar institution may require nearly the same operating work. These are stronger reasons to expand than adding unrelated categories to make the catalog look large.

If the thesis works, a plausible 2027–2030 progression is from a small set of viable transactions to repeat institutional accounts, deeper supplier relationships, and a wider range of orders handled by each operator. The company could become a substantial merchant without becoming a marketplace or selling its software. Its scale would come from repeatedly fulfilling more profitable demand with controlled operating effort and cash exposure.

That future depends on how the constraints move. If agent reliability improves quickly, incumbents will benefit as well; we would need to convert early learning into useful relationships and dependable execution. If improvement is slower, the initial business must still work with more human involvement. If capital becomes the binding constraint, processing more quotations will not solve it. Growth should follow the capacity to perform.

The larger ambition is a supplier whose commercial reach can grow faster than the labor needed to coordinate it. We intend to build toward that result one class of transactions at a time.

What we believe

Commerce requires someone to make a promise and carry it through.

We believe software can increasingly perform the information work behind that promise. We also believe the party making it must remain responsible for the goods, the terms, and the outcome.

Bringing those capabilities together could change the economics of supplying public institutions. More demand could become worth serving. A merchant could grow through better execution, with people directing the decisions that deserve their judgment.

Applied Commerce intends to build and test that company.

Our ambition is to make more worthwhile orders possible—and to be the supplier that fulfills them.