Outcome
Sign
The numbers hold. You sign with a short list of checks to hold the vendor to during the build.
Independent AI proposal review
Check the AI quote before you sign.
You have a proposal from a consultancy, integrator or vendor. We test the build cost, the monthly running cost and the payback claim, then tell you whether to sign, renegotiate or walk away. Five business days, fixed fee.
Five questions every AI quote has to survive.
Answered by someone who isn't selling you the build.
- Build cost
- Is the effort realistic, or padded, or missing the hard parts?
- Run cost
- Tokens, hosting and human review at your real volumes, not demo volumes.
- Payback
- Does the value case hold once you swap in your own numbers?
- At scale
- What breaks at ten times the volume, and how they'll know it's working.
- Lock-in
- What you own at the end, and what it costs to switch models or vendors.
Why a second opinion
The firm quoting you is also selling the build.
That's normal, and most proposals are made in good faith. But the cost to run an AI system often shows up only after go-live, when tokens, review time and upkeep land on your budget. A review gives you a clear call before money is committed.
Outcome
Renegotiate
The idea is sound but the quote isn't. You get the specific points to change: scope, milestones, run cost or ownership.
Outcome
Walk away
Payback doesn't work at your volumes. You get the reasons in writing, and a cheaper route if one exists.
What we look for
Where AI proposals usually go wrong.
We rebuild the cost model from your volumes and test each claim against how these systems behave in production.
- Running cost priced on demo volumes. Token and hosting bills sized for a pilot, not for every document or ticket you handle each month.
- Human review left out. The value case assumes no one checks the output. Someone always does, at least at first.
- No way to prove it works. No test set, no accuracy target, no plan for measuring quality once it's live.
- One model, no exit. The design depends on a single model or platform, with no plan for price changes or better options.
- No stop rule. Nothing in the plan says when to pause if payback slips.
€12,000Fixed fee per proposal, 5 business days
Fee credited in full if you book a Blueprint Sprint or Fractional AI adviser within 90 days.
What you take away
- A decision memo
- Sign, renegotiate or walk away, with the reasons on one page.
- A rebuilt cost model
- Build cost, monthly run cost, token budget and payback at your volumes. Every assumption labelled.
- Questions for the vendor
- The points to raise before you sign, and what a good answer looks like.
- Contract checks
- Milestones, acceptance tests and ownership terms worth adding.
How the 5 days run
- Day 1: Read
- The proposal, your volumes, and a walkthrough with your sponsor.
- Days 2 to 3: Rebuild
- Cost model, payback and run cost from your numbers.
- Day 4: Stress test
- Architecture, scale, quality checks and lock-in. An optional call with the vendor's architect.
- Day 5: Readout
- The call, the memo and the cost model are yours to keep.
See the work
The same method, on real numbers.
A proposal review uses the same cost and payback model as the AI Blueprint Sprint. The clock starts once we have the proposal and your volumes.
Sample decision memo
One yes, one no, and the numbers behind both.
Four AI proposals from a fictional finance team. One funded, one turned down, every assumption labelled.
Case study
Build or buy an AI assistant for pay and HR questions?
A 1,200-person IT services firm. The call: build in-house, in three gated releases, with payback in about 19 months.
Common questions
Before you send us the proposal.
- Do we have to tell the vendor?
- That's your call. Most reviews work from the proposal and any supporting documents you share. One call with the vendor's architect helps, but we can do the review without it.
- Will you recommend a different vendor?
- Only if you ask. The review tells you whether this proposal is worth signing and what to fix in it. If you want alternatives, we can shortlist them, and we have no ties to any model, platform or reseller.
- Can you review a project that has already started?
- Yes. If the build is under way, we check the running cost and quality against what was promised. That usually fits better as Ongoing support or a Fractional AI adviser engagement.
For build firms
Building AI for clients?
We don't build, so we never compete with you. A proposal that passes an independent review is easier for a CFO to approve. Talk to us about partnering.
Got a quote on your desk?
Tell us what you've been offered. In 30 minutes you'll know whether a full review is worth it.
Prefer email? shekhar@asksarthi.ai · LinkedIn