A proposal tells you what a vendor plans to build. Before you sign, you need to know what the price leaves out.
I ran Upstack Studio from 2017 to 2026, and I wrote its proposals. I know what goes into one, and what often gets left out.
A proposal is a sales document, so it shows the best case. That doesn't make it dishonest. But you still need to check the gaps before you commit.
These five checks help you read an AI or software quote. You'll know what to ask for, what to compare and when to walk away.
First, is the problem worth paying to solve?
Before you compare vendors, work out what the problem costs you. Think lost sales, staff hours and missed follow-ups. A rough monthly cost gives you something to judge the price against.
If you can't name that cost yet, start there. This guide to choosing where AI goes first shows how to work it out.
Then check whether you need someone to build anything. You could fix the process, buy a ready-made tool or do nothing. All three are real options. The build vs buy guide can help you make that call.
Once the cost makes sense, look at what each vendor is offering.
Five checks before you sign
Five checks
- 1The resultyou're paying for
- 2What's left outof the price
- 3Who does the work
- 4What counts as done
- 5What you keepafter launch
Get each answer in writing before you sign.
1. What result are you paying for?
A good proposal starts with the problem you need to solve. It uses the number you care about. A weak one jumps straight to features and screens.
Find one number you gave the vendor. Does the quote show how you'll use it to judge the work?
If it only lists what they'll build, ask them to fill that gap. Get the result you're paying for in writing before work starts.
2. What's left out of the price?
Every quote has limits. Check data clean-up, links to your other systems, content, training and testing. If a task isn't listed, ask whether the price covers it.
Look for a section called assumptions, exclusions or out of scope. These tell you what the vendor expects from you and what they won't do.
Ask how they'll charge for changes once work starts. Get the gaps and change rules in writing before you compare prices. Two quotes can cover very different work.
3. Who will do the work?
The person who sells the project often isn't the person who builds it. That's normal. You still need to know who will lead the work each day.
Ask for the team's names and roles, and when you can meet them. If the team isn't named yet, ask when it will be. Make naming the lead a condition of signing.
4. How will you know it's done?
You and the vendor need to agree what counts as finished. Write down what the system must do at each stage. These checks are called acceptance criteria.
Read each milestone with those checks in mind. Could someone outside both teams tell whether the work passed?
Tie each payment to work you've checked and accepted. If a milestone can't be tested, rewrite it before you sign.
5. What do you keep after launch?
Someone has to fix bugs, update connections and keep the system running. Agree who will do that work and what it will cost.
Then check who owns the code and data. If you part ways in a year, what can you take with you?
Get ownership, support costs and exit terms into the contract. Don't leave them for after you've signed.
Add these checks for AI work
AI systems get some answers wrong. Ask how the vendor tests them and what happens when they fail.
- What data does the system use?
- Will your data be used to train the vendor's model?
- How did they test accuracy, and what were the results?
- What will a person check before the work reaches a customer?
Ask to see it run on your own data too. Use a sample you're allowed to share, with private data removed.
A demo using their data doesn't tell you how it handles yours. Don't sign on a demo alone.
Ask for proof you can check
A vendor may tell you about a great result for another client. That can be useful, but you still need their plan for your work.
Use this list to ask for something you can check. The replies below are examples, not quotes from a real client or vendor.
| If they say… | Ask to see… |
|---|---|
| “Our clients usually see big gains.” | Your goal and how you'll check it, in writing |
| “We'll handle whatever comes up.” | A list of what's left out and how changes are priced |
| “We have a great team.” | Names, roles and the person leading your work |
| “You'll know when you see it.” | Written checks for each stage of the work |
| “We're in this for the long term.” | Terms on code, data, support and what happens if you leave |
| “It's very accurate.” | Test results from your own data, where possible |
Good vendors can tell good stories and put their promises in writing. Ask for both, so you can check the work against what you agreed.
How to compare two quotes
Read the scope before you compare the price. A lower quote may leave out work you'll still need to pay for.
Two quotes, same problem
You'll still pay for this work
Compare the scope first. Then compare the price.
Put both quotes side by side and check these points:
- Does each one name your problem and its cost?
- Does each list what's left out?
- Do you know who will do the work?
- Can you test what counts as done at each stage?
- Are code, data, support and exit terms clear?
- For AI work, can you see a test on your own data?
Ask each vendor to fill in the gaps before you choose. If they won't put an answer in writing, use that in your decision too.
When to walk away
Walk away if:
- They won't write down what the price leaves out.
- They won't say who will do the work.
- Who owns the code or data is still unclear after you've asked twice.
- The price only makes sense if nothing goes wrong.
- They push you to sign before you've checked these points.
Questions owners ask next
Should I hire a team or build in-house?
Start with your own team. Do they have the skills and time to build it? Will the tool face customers? The build vs buy guide walks through the four questions to help you choose.
How many vendors should I talk to?
Get enough detail to compare the work each price covers. A second proposal for the same problem can help you spot gaps. It won't help much if the two quotes describe different jobs.
Fixed price or pay by the hour?
A fixed price suits work you can describe fully today. Paying for time suits work where you'll learn as you go. With either choice, check what's left out and how changes are charged. A fixed price only covers the agreed work.
Should I pay to scope the work first?
A short paid discovery phase can help when the scope is unclear. Agree what you'll get before it starts. Ask for a written scope, a cost range and tests for the finished work. Those should help you make a decision, even if you don't hire that vendor.
What should I ask past clients?
Ask what went wrong and how the vendor handled it. You'll learn more from that than a general thumbs-up. Find out what happened next and whether the issue was fixed.
Who on my team should manage the vendor?
Choose one person who has time to check the work. They check milestones, approve changes and raise problems early. If nobody has time, account for that before you sign.
Check your next quote
Take the next quote on your desk and mark the gaps. Ask the vendor to fill them before you agree to the work or price.
If you're still deciding where to spend, the AI Profit Opportunity Audit can help. I find and rank the options. Your team, or people you choose, do the building. So I'm not pricing the work I recommend.
Get the newsletter for simple AI ideas. I share what I build, try and learn.
