When AI Sells Outcomes, Define What Counts

The invoice is where the promise gets specific

Try a simple thought experiment. A hypothetical food distributor is considering an AI service to handle delivery exceptions. The supplier demonstrates that an agent can contact a carrier, obtain a revised arrival time and update the shipment record. The demonstration ends. The invoice discussion begins.

What would count as a paid outcome: the message sent, the record updated, the exception accepted as resolved, or the order arriving on time and in full? Each describes something different. A beautifully updated record can still document a late truck.

This is not a criticism of FourKites's product behavior, which this article has not tested. It is a proposed buying exercise: put one realistic case beside the proposed fee schedule and ask both sides to identify the event that earns payment. If their answers differ, they have discovered work to do before signing.

Deloitte's June 4 Technology Spotlight provides a useful distinction. It separates a promise to keep an agent available from a promise to deliver a specified quantity of successful outcomes. Its purpose is supplier revenue-accounting analysis, not buyer procurement advice. My application is to ask whether the commercial promise is access, completed work or a defined business result before comparing prices. [3]

Choose a result the parties can actually observe

There is external evidence that the measurement problem survives a change in pricing vocabulary. HFS Research's September 21 report, produced in partnership with Capgemini, draws on a survey of 101 financial-services firms and senior-leader interviews. It describes outcome terms being added to existing arrangements while established pricing structures remain important. Baselines and attribution are central difficulties. This is sector-specific, commercially partnered research, not a representative census of AI contracts or evidence about FourKites. [4]

For the fictional distributor, I would distinguish accepting a corrected shipment record from improving delivery performance. The first can be checked against agreed information. The second requires a comparison over time and an explanation of what else changed. Suppose the company simultaneously changes carriers and relaxes its delivery window. A better on-time score would not, by itself, identify the AI service's contribution.

The practical response is to agree the comparison before the pilot: which shipments are included, what counts as on time, which period supplies the baseline and how material operating changes will be handled. Those choices should be visible to operations and finance as well as the supplier. This is my proposed application of HFS's measurement argument, not a validated attribution method. [4]

Precision does not require pretending that every effect can be isolated. If reliable attribution is beyond the available data, narrow the paid outcome to something observable and assess the wider business benefit separately. Better to price a modest promise clearly than negotiate a grand promise that nobody can verify. [4][5]

An outcome fee does not make the remaining work disappear

Return to the distributor. Perhaps the agent handles the routine carrier exchange, but an employee still reconciles conflicting updates, checks sensitive orders and corrects mistakes. In this hypothetical arrangement, the buyer should compare the supplier's fee plus that retained work with the current process. Moving a task off one invoice does not make it vanish from the business.

A March 6 study in npj Digital Medicine offers a useful, bounded perspective. Interviews with 17 healthcare decision makers in German-speaking countries identified interest in predictable, understandable pricing and explicit integration, training and support. The authors recommend adding outcome-linked components when measurement is ready. This qualitative study did not test pricing structures in prospective procurement settings, and it does not establish what works in logistics. Two authors disclose employment by medicalvalues. [5]

I would apply that lesson by asking for a complete operating-cost comparison before choosing a fee model. Include implementation, supplier charges and the human effort the buyer retains. A predictable base fee with a narrowly defined performance component may deserve consideration alongside a pure outcome offer. The right answer depends on the work and evidence, not the modernity of the pricing label. [3][5]

Bring a disputed case to the buying meeting

My recommendation is a short commercial rehearsal with the operating owner, finance, procurement and the supplier. Use a completed case and an awkward one. The purpose is to expose disagreement while it is still inexpensive to resolve. These are author-designed questions informed by the pricing analysis and research, not a tested negotiation method. [3][4][5]

• What earns the fee? Specify the unit of success, the acceptance evidence and who checks it. Walk through a case that needs human correction or later reopens.
• What explains the improvement? Agree a baseline and measurement period, record other operating changes and decide when the comparison must be reset.
• What remains ours? Estimate integration, review and exception-handling effort. Compare the complete cost with the existing process and clarify how disputed results are handled.

Buy the promise you can evaluate

The distributor might choose a smaller initial commitment: accepted exception resolution for a defined shipment group, with delivery performance tracked separately. It might instead pay for access while gathering enough evidence to negotiate a performance component later. Neither fictional choice proves the technology works. Each makes the buying decision easier to explain.

FourKites's announcement gives leaders a timely opening to ask suppliers for a closer connection between their offering and the customer's operating result. Take that opening seriously. Then make the result concrete enough that operations, finance and the supplier can recognize the same success when the bill arrives.

Questions for executives

1. Would our operating team and supplier identify the same billable success in an awkward case?
2. What evidence would separate the service's contribution from other changes?
3. Which costs and responsibilities would remain with us under the proposed arrangement?