Page 14 of UFIDA's 2026 semi-annual report disclosed that the company has established an FDE team to co-create and provide continuous services with customers around specific scenarios, and explore charging based on results. The report also separately describes the resale of Token and other resources and AI business operation services. This shows that the disclosed materials involve several types of business arrangements, and they cannot be combined into the same quotation, let alone infer that a certain customer has completed a transaction at a certain unit price.
Let’s first explain what can be counted as one result.
The following are independent design suggestions for small teams and are not a public contract from UF. Assuming that you are organizing inquiry materials for a customer, the billing object can be a data package that completes the specified fields, is attached with the original source, and is confirmed by the customer. A model call, an automatically generated draft, and a package accepted by the client are three different units of counting. Choose one of them first before you can discuss the price.
Counting units also have boundaries. When an inquiry is suspended due to lack of data, the customer cancels midway, or the same material is uploaded repeatedly, whether it will be included in the service volume should be agreed in advance. Putting these situations into a sample table and demonstrating normal completion, return and supplement, and termination processing one by one can reduce misunderstandings about work progress by both parties. The examples here are for scope confirmation and should not be packaged as completed customer orders.
Leave separate basis for result cost and resource cost.
The deliverer can establish two corresponding records: one records which service items were completed, who confirmed it, and based on which version; the other records the actual investment in models, storage, third-party interfaces, and manual processing. The customer purchases an agreed service, but the team still needs to know what it costs to complete each task, otherwise the increase in business volume may just increase costs that cannot be recovered.
For tasks that are reworked repeatedly, system errors and new requirements should also be distinguished. If the originally confirmed field extraction is incorrect, it can be processed according to the correction mechanism; if the customer adds a new data source or changes the business rules, the scope needs to be re-judged. In order to make the completion rate look better, you cannot delete failed tasks from the ledger, nor can you use automatically generated quantities to replace the number of results approved by the customer.
Performance indicators cannot be separated from the evidence that customers can provide
If the agreement is related to saving time, both parties must at least clarify the starting and ending points of the original process, the observation period, sample type and whether manual review will be included. If the customer is unable to provide these records on an ongoing basis, a fixed-range implementation fee and maintenance fee can be adopted instead of directly committing to a share of the savings. This article does not provide a unified pricing conclusion. The specific arrangement depends on the project, cost and agreement between both parties.
Before a small team makes a quotation, they can use a one-page description to list the service unit, confirmer, dispute examples, resource cost caliber and stop conditions. Let customers understand how bills are generated first, and then judge whether it is worth trying. UF's disclosures provide research leads but do not disclose data on individual project contracts, net customer revenue, or small-team profits that would prove the model is generally applicable.
Continue reading: How to agree on paid pilots and Understanding AI startup numbers, you can understand the pilot scope and business records together.