This post assumes you have already decided to buy rather than build. If you have not, should a mid-market company build or buy its AI tools covers that decision. Once you are buying, the question becomes which vendor, and the hard part is that every demo looks good.
Why is the sales demo the wrong place to decide?
Because a demo is designed to be impressive, and it is run under conditions you will never have. The data is clean. The task is the one the tool does best. The person driving has done it two hundred times. None of that is dishonest. It is just not evidence about how the tool will behave in your company, on your customer records, with your team using it on a Tuesday afternoon.
The owners I speak with who regret an AI purchase almost never say the tool did nothing. They say it did less than the demo, cost more than the quote, and was harder to leave than they expected. Each of those is a question you can ask before signing. The rest of this post is the list.
What should you ask about your data?
Three questions. Ask them first, because the answers can end the conversation.
- What data does the tool need access to, and can we limit it? A tool that drafts proposals does not need your payroll. Ask for the list of systems it connects to and whether access can be set by role. If the answer is “it works best with everything,” ask what it does with the parts it does not need.
- Is our data used to train or improve your models? Then ask where that is written. Some vendors say no in the meeting and yes in the terms of service. The contract is the answer. The meeting is not.
- Where is our data stored, who at your company can see it, and what happens if you are breached? You want a named region, a named access policy and a notification timeline. Your one-page AI usage policy should already say what data may leave the building. Hold the vendor to the same line.
What should you ask about reliability?
AI tools fail differently from the software you are used to. A spreadsheet gives a wrong answer when a formula is wrong. An AI tool can give a confident wrong answer with nothing visibly broken. So the questions are about failure, not uptime.
- How does the tool fail, and how will we know? Ask for examples of wrong outputs from real customers. A vendor who says it does not make mistakes has either not watched it or is not telling you. The good answer describes the kind of mistake, how often, and what the tool does to flag it.
- Who reviews the output before it reaches a customer? If the answer is nobody, you are being sold automation, not assistance, and the reliability bar is much higher. Decide which one you want before you decide which vendor.
- What happens when the underlying model changes? Most AI tools sit on top of a model built by someone else. When that model is updated, the tool’s behavior can change overnight. Ask whether you are told, whether you can test before the change reaches you, and whether you can hold a version.
What should you ask about cost?
The quote on the slide is the smallest number you will pay. How much an AI rollout costs a mid-market company walks through the full picture. For the vendor conversation, three questions get most of it.
- What is the total for three years, including usage? Many tools charge a platform fee plus a charge per message, per document or per seat. Ask the vendor to price your expected volume for years one, two and three. If they cannot, model it yourself before signing.
- What is the renewal increase, and is it capped? Introductory pricing is common. Ask for the cap in the contract. A vendor who will not cap renewals is telling you the first-year price is not the real price.
- What will it cost us in people? Someone has to set it up, maintain the connections, review output and train the team. Ask the vendor how many hours a week their typical customer of your size spends on this. Then ask one of those customers.
| Cost line | What to ask for | Where it usually hides |
|---|---|---|
| Platform fee | Three-year total at your volume | Quoted monthly, for year one only |
| Usage charges | Price per unit and your expected units | “Fair use” language with no numbers |
| Renewal | Written cap on the annual increase | Absent from the order form |
| Implementation | Fixed scope and fixed price | Hourly, open ended |
| Internal time | Hours per week from a reference customer | Never on the vendor’s paperwork |
What should you ask about leaving?
Ask this while you still have leverage, which is before you sign.
- How do we get our data back? You want the format, the method and the time window. “Contact support” is not an answer.
- What happens to the prompts, workflows and training we build inside the tool? Much of the value in an AI rollout is the configuration your team builds. Ask whether any of it can be exported. Often the honest answer is no, and that should affect how long a contract you sign.
- When do you delete our data after we leave? Get a number of days, in writing.
How do you test a vendor instead of trusting the demo?
Run a two-week test on your own data, with the people who will actually use the tool, measured against three results you wrote down before the test started. Pick the process the tool is meant to change, measure how it performs today, and compare. How to measure the ROI of AI in a mid-market company explains how to set the baseline.
During the test, keep a shared log of every wrong output. At the end you will have something no demo can give you: a count of how often the tool was wrong, on your work, and what kind of wrong it was. That number, more than any feature, should decide the purchase.
A five-person team can generate $32 million in revenue with AI-driven systems, as mine did, but only if the systems are chosen against real work and not against a demo. The tools that earn their place are the ones that hold up on your data, with your people, on an ordinary week.
Which answers should stop a deal?
- The vendor cannot show you the contract clause on training with your data.
- The vendor says the tool does not make mistakes.
- The vendor will not price three years at your volume or will not cap renewals.
- The vendor will not agree to a test on your data before signing.
- The vendor cannot name a customer your size you may call.
Any one of these is reason to pause. Two is reason to move on.
What should be in writing before you sign?
The data clause, the renewal cap, the exit terms and the results from your two-week test, attached to the order form as the standard the tool is expected to meet. Have your attorney review the agreement; this post is general guidance, not legal advice.
If you want help choosing the first process, running the test and scoring vendors against it, the engagements and FAQ page explains how the AI Clarity Sprint works. The prompts for drafting vendor questions and scoring the answers are in ChatGPT for Profit, Volumes 1 and 2.
Frequently asked questions
- What is the most important question to ask an AI vendor?
Whether your company’s data is used to train or improve the vendor’s models, and where that is written in the contract. The answer decides whether your customer lists, pricing and internal documents stay yours. A vendor that cannot point to the clause has not answered the question.
- Should a business owner trust an AI sales demo?
Treat it as a preview, not as evidence. A demo runs on data the vendor chose and on tasks the tool is known to do well. Ask for a two-week test on your own data, run by the people who will use the tool every day, with a short list of results you will measure before you start.
- How long should an AI contract be for a mid-market company?
Start with the shortest term the vendor will offer, usually twelve months, even if a longer term is cheaper. AI tools and prices are changing quickly, and the discount for a three-year commitment is rarely worth losing the ability to switch. Ask for the renewal price cap in writing.
- What should an AI vendor contract say about getting your data back?
It should name the format your data will be returned in, how long you have to export it after the contract ends, and when the vendor deletes its copies. If the contract is silent on any of the three, ask for the language before you sign. Leaving a tool is much harder to negotiate after you depend on it.
- Who should be in the room when a company evaluates an AI vendor?
The owner or executive who holds the budget, the manager who runs the process the tool will change, and whoever is responsible for IT or data at the company, even if that is an outside firm. The process owner asks what the tool does. The IT or data person asks where the data goes. The owner asks what it costs over three years and how the company gets out.
Choosing an AI vendor and want a second set of eyes?
The AI Clarity Sprint gives mid-market companies, law firms, wealth managers and insurance agencies an AI audit, deployment recommendations with ROI models, and a 90-day plan in three weeks.