Why I split payment 40/30/30 and start with discovery rather than a “turnkey” contract

Implementing an AI agent is not buying a boxed product. It is rebuilding a piece of a working process, and at the outset neither side knows everything: I don’t know the quirks of your 1C (the ERP/accounting system standard in Russia), and you don’t know how the agent will behave on live data. The payment format has to reflect that uncertainty honestly — otherwise it turns into shifting the risk onto one of the parties.

Why not 100% upfront — and not payment on completion either

Full upfront payment shifts all the risk onto the client: they pay before they have seen the system work. Full payment on completion shifts it onto the contractor: months of work on credit against a promise. Both schemes breed the same disease — haggling instead of work: with upfront payment the contractor has every incentive to wrap the project up fast; with payment on completion the client has every incentive to expand the requirements endlessly.

The 40/30/30 split ties the money to verifiable checkpoints:

• 40% at the start — covers discovery and the pilot phase. This is work that gets done in any case, even if you decide not to go further.
• 30% after the pilot — when the agent is already running on your data in “read-only and reports” mode and you have seen with your own eyes what it finds. You pay for what you have verified.
• 30% after go-live — when tasks are being assigned, digests are arriving, and the system has lived through its first working weeks.

Why discovery is a separate phase

The most expensive implementation mistakes are made before the first line of code — at the “we’ve roughly understood what you need” stage. Discovery turns “roughly” into specifics: what data actually exists in 1C, what shape it is in, which checks will pay off in the first week, and which are a pretty idea with no data underneath.

A side effect of discovery is that you get a useful result before implementation. The very first run over historical data usually surfaces specifics: in the wholesale company case study it was 18 debtors and 143 missing stock items — before the system had the right to do anything at all.

What this gives the client in practice

An exit point after every phase. If after the pilot you decide not to continue, you have lost 40% of the budget but gained discovery, a working pilot and a list of findings from your own data. That is an unpleasant outcome, but not a catastrophe. An upfront-payment scheme in the same situation leaves you with a stack of promises.

Details of the format are on the pricing page, and what the result of each phase looks like is in the case studies with “before → after” figures.