How to Evaluate an AI Implementation Proposal or Quote

Evaluate an AI implementation proposal on six things: whether it states your problem in numbers, lists deliverables you can test, shows the full cost including monthly running fees, puts everything in accounts you own, explains what happens to your data, and says what support and exit look like. Query anything vague in writing before you sign.

Most quotes for small AI projects fall down on the same point: they price the build carefully and say little about what it costs to run, or who controls it once the builder has gone. The checklist below catches that, and a worked example shows how a cheaper-looking quote can cost two and a half times more over three years.

Follow me on Instagram@sagnikteaches

The 22-point proposal checklist

Go through it with the proposal open beside you. For each item, mark it as clear, vague or missing. You don't need every item to be perfect for a small job, but anything marked missing in the running-cost or ownership groups deserves a written answer.

Connect on LinkedInSagnik Bhattacharya

The problem and the result

1. It describes your problem in your numbers. Hours a week, enquiries a month, errors per quarter. Check that the figures match what you told them; if they're rounded up to make the project look bigger, ask why.
2. It names a test of success. For example, "in a two-week trial, at least 8 in 10 drafted replies need only light edits". Without one, you can't hold anyone to anything.
3. It says what's out of scope. A proposal that lists what it won't do is easier to trust than one that seems to cover everything.

Deliverables and acceptance

4. Each deliverable is something you can see. "A working enquiry-drafting automation connected to your inbox" is a deliverable. "Discovery", "optimisation" and "AI enablement" are activities.
5. It explains how you'll accept the work. Who signs off, against which test, and what happens if the result fails it: a fix at no charge, a partial refund, or nothing.
6. Testing uses your real examples. Ask how many past cases will be run through it before go-live. Twenty is a sensible minimum for anything customer-facing.

Here are items 4 and 5 in practice: a line from an illustrative quote to a small architecture practice, and the version worth asking for instead.

Subscribe on YouTube@codingliquids

AS QUOTED
Phase 2: AI enablement and optimisation of client
onboarding. 5 days.

AS IT SHOULD READ
Deliverable 2: when the new-client form is submitted, an
automation creates the project folder, drafts the fee
letter from the form answers for a partner to approve,
and adds the client to the CRM.
Acceptance: 15 test submissions from last year's clients
run correctly, with any failure sent to the office inbox.
If 3 or more fail, we fix and re-test at no charge before
the final invoice.

The first version can be "delivered" by five days of meetings. The second can be tested in an afternoon.

Price and payment

7. The pricing basis is stated. Fixed price, hourly with an estimate, or a monthly retainer, and what triggers extra charges under each. Fixed-scope versus hourly billing explains which protects you in which situation.
8. Payments follow milestones, not dates. I'd be wary of paying more than half before you've seen anything working on your own data.
9. Change requests have a rate and a process. Small AI projects always change once people see the first draft. Know the price of that in advance.

Running costs (the group most quotes leave out)

10. Every subscription and usage fee is listed, with who pays it. The automation platform tier, AI seats or API usage, any add-ons, and any fee the builder charges for hosting it.
11. Usage estimates show their working. Volumes in, tasks or credits or tokens used, cost per month. Ask what the bill looks like if your volume doubles in your busy season.
12. Price-rise exposure is clear. If the builder resells a tool or charges a platform fee, can it rise mid-contract, and with how much notice?

An AI assistant can do the first pass on a long proposal, provided you make it quote. Paste the proposal (and any terms attached to it) into a business-plan assistant with a prompt like this:

Below is a quote for an AI automation project. List every cost it
mentions, one per line: amount, one-off or recurring, who is paid,
and the exact sentence it comes from. Then list every sentence that
describes an activity rather than something we will receive.
Do not summarise or estimate. Write NONE if a list is empty.

The reply (illustrative) for a platform-fee quote might list "$1,500 setup, one-off, agency" and "$249 a month, recurring, agency", then flag "ongoing optimisation of your AI workflows" and "discovery and alignment sessions" as activities. That's a good start, but check it against the document with your browser's find function before relying on it. In this example a search for increase finds a line in the attached terms, "fees may be adjusted annually in line with our pricing", which the assistant didn't list because no amount was attached. That sentence goes straight into item 12 as a question.

Seasonal businesses are where item 11 bites. Picture an illustrative garden-furniture shop whose quote says running costs are "Zapier Professional, $19.99 a month billed annually". That covers 750 tasks. The builder based the estimate on 100 enquiries a month at five action steps each, 500 tasks, which is true from September to March. In April to June the shop gets around 400 enquiries a month, which is 2,000 tasks and well past the plan's allowance. The quote isn't dishonest; it priced an average month. Ask for the running cost in your busiest month, and budget for that tier.

Ownership and access

13. It's built in accounts your business owns. You hold the admin login; the builder is invited as a user and removed at the end.
14. You own the workflows, prompts, code and documents. If the builder reuses their own components, you should get a licence to keep using them. Who owns the workflows a consultant builds covers the wording.
15. Access is handed back. The proposal says when the builder's logins and API keys will be removed.

Data

16. It says which data goes to which service. And whether that service may use it to train models. Business plans from the major AI vendors don't train on business content by default; consumer plans need a setting switched off.
17. It covers what the builder sees during the work. Which customer records they'll handle while building and testing, and how those copies are deleted afterwards. Where personal data is involved, ask whether a data processing agreement is needed; your data-protection adviser can tell you.

Silence on item 17 is how test data ends up in the wrong place. Take an illustrative estate agency whose builder, wanting realistic test cases, exported the full contacts list of 3,000 buyers and vendors into a spreadsheet on their own cloud drive. Nothing bad happened to it, but the agency only learned where it was at handover, when it asked for the test log and got a link to a file it didn't own. One sentence in the proposal would have prevented it: "Testing will use 25 past enquiries with names, emails and phone numbers replaced; any copies will be held in your account and deleted at go-live, confirmed in writing."

Timeline and your part

18. It lists what you must provide, and by when. Samples, access, and hours from your staff. Projects slip because of the client far more often than the builder, and a good proposal says so.
19. The timeline states its assumptions. "Four weeks, assuming access to the inbox and 30 sample enquiries by day three" is honest. "Four weeks" on its own is a hope.

What that looks like in an illustrative four-vet practice: the proposal said four weeks, assuming 30 sample appointment enquiries by day three. The practice agreed, then found every sample email carried pet owners' names, phone numbers and addresses that had to be removed first. The reception team did it between appointments and the samples arrived on day eleven. Go-live moved by just under two weeks, and because the assumption was written down, nobody argued about whose delay it was. When you read this item, count the staff hours it asks of you and put them in someone's diary before you sign.

After go-live

20. Support has a length and a response time. And a definition of what counts as a fault (fixed free) versus a change (charged). Ask for examples in writing. A fault: drafts stop arriving because the email connection expired, or the AI starts quoting last year's prices when the price list it reads was updated correctly. A change: adding a second service to the replies, or sending drafts to a new member of staff. Agreeing two or three examples of each up front heads off most arguments in month two.
21. Handover documents are a named deliverable. How it works, how to pause it, what it costs, who owns it internally.
22. Exit is described. What happens if you stop paying, how much notice either side gives, and what you keep. "You keep everything" is only meaningful if it's in a form you can move: an export of each workflow (Make, for example, can export a scenario as a blueprint file), the prompts as text, and the reference documents the AI reads. What you pay after an automation goes live helps you judge whether an ongoing fee is fair.

Phrases in AI proposals that need pinning down

Worked example: a bakery compares two quotes

Say a bakery takes about 80 custom-cake enquiries a month by email and web form. Each needs a reply with a price range, available dates and questions about allergies and design, and the owner spends around six hours a week writing them. The job itself is described in taking custom cake orders with an AI enquiry assistant. Two quotes arrive. The figures are illustrative; the tool prices are real list prices as of September 2026.

Quote A: $3,200 fixed. Built in the bakery's own Zapier and email accounts; the AI drafts each reply and the owner approves it before it's sent. Includes 30 days of fixes after go-live and a handover document. Running costs, which the bakery pays directly: Zapier Professional at $19.99 a month billed annually, plus AI usage.

Quote B: $1,500 setup, then a $249 monthly "platform fee" on a 12-month minimum term. Runs on the agency's own platform. "Unlimited tweaks" included. No mention of what happens at the end of the term.

First, check Quote A's running-cost claim. Zapier counts each successful action step as a task (triggers and filters don't count). At 80 enquiries with about five action steps each, that's roughly 400 tasks a month, inside the 750 the Professional plan includes. For the AI itself: if each enquiry uses about 4,000 tokens of input (the enquiry, the price list, the instructions) and 800 of output, 80 enquiries is 0.32 million input and 0.064 million output tokens. At Claude Sonnet 5's API price of $2 input and $10 output per million tokens, that's about $1.28 a month. Call it $5 to leave room for longer threads.

Quote B looks cheaper on the day and costs about two and a half times as much over three years. That doesn't make it a bad offer in every case: if the agency's fee covered real monitoring, weekly quality checks and fast fixes during wedding season, it might be worth paying. But the proposal doesn't say so, and "unlimited tweaks" is not the same thing. The bakery's next step is an email asking the agency the questions from the phrase table above, and asking the author of Quote A what the 30 days of fixes excludes.

A scoring sheet for comparing quotes

Score each proposal 0 (missing), 1 (vague) or 2 (clear) on these ten lines. Anything under 12 out of 20 needs a written clarification round before you decide. A quote that scores 0 on either ownership line should be ruled out, however cheap it is.

AI PROPOSAL SCORING SHEET                    Quote A   Quote B
1. Problem stated in our numbers               __        __
2. Success test we can measure                 __        __
3. Deliverables we can see and accept          __        __
4. Testing on our real examples                __        __
5. Full running costs, with working            __        __
6. Built in accounts we own          (0 = out) __        __
7. We own workflows, prompts, docs   (0 = out) __        __
8. Data handling explained                     __        __
9. Support period and fault definition         __        __
10. Exit terms: what we keep, notice           __        __
                                     TOTAL /20 __        __
Three-year cost (build + 36 months running):   $____     $____

Filled in for the bakery's two quotes, going only on what each proposal actually says, the sheet reads like this:

AI PROPOSAL SCORING SHEET                    Quote A   Quote B
1. Problem stated in our numbers               2         1
2. Success test we can measure                 1         0
3. Deliverables we can see and accept          2         1
4. Testing on our real examples                1         0
5. Full running costs, with working            1         1
6. Built in accounts we own          (0 = out) 2         0
7. We own workflows, prompts, docs   (0 = out) 1         0
8. Data handling explained                     0         0
9. Support period and fault definition         1         1
10. Exit terms: what we keep, notice           1         0
                                     TOTAL /20 12        4
Three-year cost (build + 36 months running):   $4,100    $10,464

Quote B is out on line 6 before the totals matter. The more useful finding is that Quote A, the clear favourite, scores only 12 and has a 0 on data handling: the proposal never says what the AI service does with customers' allergy details. That becomes an extra line in the clarification email ("Which AI service will read customers' allergy details, and may it use them for training?"), and question 4 covers the missing success test.

An email to send before you accept

Most builders will happily answer these; the ones who bristle are telling you something. Adapt the wording and send it to every shortlisted supplier so the answers are comparable.

Subject: A few questions before we decide on your proposal

Thanks for the proposal. Before we decide, could you confirm in writing:

1. Every monthly cost we'll pay once this is live (subscriptions, usage,
   any fee to you), and what it would be if our volume doubled.
2. Whose accounts it will be built in, and when your access is removed.
3. That we'll own the workflows, prompts and documentation, and can keep
   using them if we stop working with you.
4. What test the work must pass before we sign it off, and what happens
   if it doesn't.
5. What your support covers after go-live, for how long, and what you
   would charge as a change rather than a fix.
6. Which of our customer data you'll see while building it, and how
   those copies are deleted afterwards.
7. What you need from us (samples, access, staff time) and by when.

Thanks,
[Name]

When to walk away instead of negotiating

Price is negotiable, and negotiating an AI consulting quote covers how to do it without losing the parts that matter. These, though, are reasons to stop rather than haggle:

• The builder won't build in your accounts, or won't say what you keep if you leave.
• Running costs are "to be confirmed" after you've signed.
• No testing on your real examples is planned before customers see the output.
• The proposal promises a specific saving in hours or revenue without having looked at how you work.
• You're asked to pay most of the fee before anything is shown working.

If none of those apply and the scores are close, pick the builder whose proposal described your business most accurately. Understanding the problem is the part you can't buy back later. And if the quotes differ wildly in what they include, it may be worth scoping the work yourself first; how to scope an AI project shows how to set deliverables and acceptance criteria before anyone prices them.

Further reads

• Questions to Ask an AI Vendor Before You Sign Anything — The questions to put to a software vendor rather than a builder.
• AI Consulting Contracts: 9 Clauses to Check Before You Sign — Nine clauses to check once the proposal becomes a contract.
• Hidden Costs of AI Implementation Most Small Businesses Miss — The costs that surface after the build is paid for.
• How to Write a Request for Proposal for an AI Project — Write the request first so quotes arrive comparable.
• Scope Creep in AI Projects: How to Handle Change Requests — How to handle change requests once work starts.
• AI Automation Agency Pricing: What You Pay For and What's Extra — What agency fees include and what they bill as extra.
• Do I Need an AI Consultant? 8 Signs It's Time to Get Help — Eight signs a small business needs outside AI help, each with a test you can run this week, plus the signs that it doesn't need a consultant yet.
• How to Brief a Developer on a Custom AI Workflow You Need — The eight parts of a developer brief for an AI workflow, a copyable requirements template, and a worked example from a dog-grooming salon.
• What AI Implementation Actually Involves for a Small Business — The six pieces of work behind a small-business AI implementation, with rough hours for each and one workflow followed from baseline to handover.
• Restaurant AI Tools: 12 Questions to Ask Before You Sign Up — Twelve written questions for any restaurant AI vendor, with red flags, a scoring sheet and five test calls to make before you commit.
• Can AI Help a Small Business Respond to Tenders and RFPs? — Turn a tender pack into an evidence checklist, a checked draft and a realistic decision about whether to bid.
• How to Write a One-Page AI Business Case for a Small Business — Five stages to a one-page AI business case, a finished example from a small charity, and a drafting prompt that stops AI inventing figures.
• How to Vet an AI Consultant's Case Studies and References — A grouped vetting checklist, a case study taken apart claim by claim, a reference-call script with sample notes, and a church office choosing a consultant.
• 10 AI Implementation Mistakes Small Businesses Make — Ten setup, billing and upkeep mistakes that turn a sensible AI project into a year-long cost, each with how it shows up and the fix.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Zapier pricing page and task-counting help documentation; Make Help Center (scenario blueprints); Anthropic API pricing page (checked September 2026).