Choosing an AI vendor: what actually matters
In most companies, nobody really chooses an AI tool. One person starts using ChatGPT or Claude, shows a colleague, and six months later half the team is on it, some on free accounts and some on a company card. That's a perfectly normal way for it to start. It's just not a great way to end up with the tool your business depends on.
If you’re at the point of picking one properly, this is how I’d go about it.
The model and the product are different things
When people compare AI vendors, they usually compare models: which one writes better, which one is smarter at math, which one topped a ranking this month. The model matters, but it’s only part of what you’re buying.
What your team actually uses is the product around the model. That’s the app they log into, the admin settings, the connections to your email and files, the contract and the bill. Two products built on the same model can treat your data very differently. Some platforms also let you use models from several providers in one place, so you don’t have to bet everything on whichever model happens to lead today.
So instead of asking which model is best, ask which product your business can use safely and consistently.
Start with the data question
I’d rule vendors out on data handling before looking at anything else. For the plan you’d actually buy, not the free version, find out:
- whether your prompts and files are used to train their models (business plans usually exclude them; free and personal plans often don’t unless someone changes a setting)
- how long conversations and uploaded files are kept, and whether you can shorten that
- where the data is stored, if your clients or your industry care about that
- whether they’ll share an independent security report, such as SOC 2 Type II or ISO 27001
A good vendor answers all of this in writing without much fuss. If you get vague answers, take that as your answer.
Then look at the admin side
This is the part people skip, and it’s where most problems show up later. You want people to sign in with the accounts they already have, such as Microsoft 365 or Google Workspace. You want access to end automatically when someone leaves, and you want to control who can share what. And you want some record of how the tool is being used.
None of that is exciting in a demo. But without it, you can’t answer simple questions like “who has access to this?” or “did anything get shared outside the company?” Those are exactly the questions a client or an insurer will eventually ask.
Test it on your own work
Online rankings measure general ability. They won’t tell you which tool writes better proposals for a 20-person engineering firm, or which one summarizes your contracts without missing the payment terms.
The only test I trust is a short pilot. Pick two or three options. Pick a handful of people from different roles, and make sure at least one of them is skeptical. Have them use each option for two to four weeks on tasks they actually do every week, with realistic examples that don’t contain anything sensitive. Then ask them which one they kept reaching for, and look at what they produced.
A simple scorecard helps keep the comparison honest:
| What you’re checking | What good looks like |
|---|---|
| Data handling | No training on your data, clear retention, an independent audit report |
| Admin controls | Company sign-in, roles, usage and audit logs |
| Fit with your tools | Works with your email, documents and CRM, and respects existing permissions |
| Quality on your tasks | The pilot group prefers its output most of the time |
| Real cost | Predictable when the whole team uses it, with key controls included |
| Getting out | You can export your data, and the contract isn’t a trap |
Price it for the whole team
AI pricing is per person, per use, or a mix of both, and the number on the pricing page is rarely the number you end up paying. Ask what happens when someone hits a usage limit. Ask which features only come with the higher tiers, because admin and security controls often do. Then work out the cost with everyone using it, not just your early adopters.
Compare that against the time it saves. A cheap tool that nobody uses is still money wasted.
Keep a way out
AI tools change every few months. The vendor that’s ahead now might not be next year. Before you sign anything long, check that you can export your conversations, files and any custom assistants you build. Also make sure your processes don’t depend on a feature only one company offers. Keeping the option to switch costs you very little today and can save you a lot later.
If you take one thing from this: check data handling and admin controls first, then let your own team’s work decide between whatever is left. The best AI vendor for your business is the one your people can use every day without anyone worrying about it.
We'll compare the options against your tools, your data and the work your team actually does.