Settle the attribution before the money moves. That's what our co-founder Anita Rajdev would put alongside the outcome and the end-to-end cost as the things to pin down before anyone spends anything. She set the test out to journalist Rayanne Harmon in an interview with Raconteur, published on 11 November 2026. It's the one of the three that tends to get skipped, and two teams can both claim the same saving in good faith.
Some of the highlights are below, and there's a link to the full interview.
Attribution is a decision for day one
Before anything gets spent, you need agreement on what the work is meant to produce. After that the questions are the ones you'd expect. Was the finished work worth doing, what did it really cost, was the output usable, did decisions improve, did the business get quicker. Anita's point is that a case held to those questions gets a lot clearer.
Attribution belongs in that same group, and she wants it sorted at the outset.
"Attribution also needs to be agreed at the beginning. An AI project may involve internal teams and other third parties, so companies must be clear about who contributed to the result. You can only save a dollar once." Anita Rajdev, speaking to Raconteur
We'd add one thing. The split belongs in the approval paper, with a named owner against each share, signed off alongside the funding.
Hours saved is only the floor
The mistake Anita sees most often is a familiar sum. Take a process you already run, work out the hours AI might take out of it, and book that as the whole return. What your team then does with the time it gets back is the part left out. If people use it to get through more work, or to move on to something more valuable, she argues the benefit can run well past the initial saving.
AI isn't a silver bullet, in her words, and weak data quality, unreliable ingestion and records that contradict each other may all need fixing first. Start with the problem and work backwards, rather than assume the technology will sort it.
Why we tie our fee to the evidence
We built Valliance to deliver measurable value, and we committed early to not selling time and materials. We only get paid in full when the agreed result arrives, which means we and the client have to define that result properly at the outset.
The model has two parts. There's a base fee, priced on how risky the work looks, how confident we are of reaching the result and what delivering it will take, which also gives the client a stake in the outcome. On top sits a share of the value, tied to the result. Anita models the case where the success fee lands in full and the case where none of it does, and those two numbers are the guardrails in her reporting to investors.
It's harder than billing by the hour, as she says plainly. What holds it together is the shape of the fees and the forecasting behind them. It narrows the client list too. If you're set on a conventional time-and-materials contract, you're probably better off elsewhere.
First published as CFO on the Spot: Anita Rajdev, co-founder of Valliance by Rayanne Harmon, Raconteur, 11 November 2026.
FAQs
Does tying fees to outcomes make revenue harder to forecast?
Yes, and Anita says so plainly. Her answer is the structure of the fees plus forecasting across scenarios, including the one where no success fee turns up at all. If you're buying, ask whoever is pitching to show you that forecasting.
How should attribution be handled when several teams contribute?
Before the work starts. Name the contributors and agree the split in writing. Anita's line that a dollar can only be saved once is the test, and if two business cases claim the same saving then one of them is wrong.
Is time saved a good enough measure of AI value?
No. The hours a process gives back are only the starting figure, and Anita's argument is that what your team does with those hours carries the rest.
What financial evidence should a CFO demand before funding the next AI use case?
An agreed definition of the outcome, a cost that covers the work end to end, a judgement on whether the output is usable, and a written attribution split naming who gets to count which part of the benefit. Anita's point is that a case held to those questions is much clearer. We'd press hardest on the attribution split, because it decides whether your reported saving holds up when someone checks it later.
















