You had the conversation with the sales guy. Their new AI platform is going to solve all of your problems, and get your company running so smoothly... hell, you won't even have to work anymore!
You're thinking about signing on to the exciting new magical AI platform that will do everything for you... but HOLD UP. Practice your due diligence. Here's the first question you need to ask before you sign: if we cancel tomorrow, what do we actually lose?
If they can't answer it cleanly, it's a big red flag.
That's just one of a few questions that you have to ask. There's a few more you ask down below.
What's under the hood
When I look at these platforms closely, I keep finding the same thing. The pitch is a custom AI workspace that connects to Microsoft 365, Google Drive, and Slack. The model underneath, Claude, in the cases I know best, already connects to all of those natively. Anthropic's enterprise tier handles Drive, Gmail, Slack, and Microsoft 365 with no extra layer, and its connector marketplace launched with more than 2,000 integrations, with CrowdStrike and Snowflake among the partners.
So the buyer is paying a premium for a slower copy of a connection they already hold. They also get a new contract and one more thing that can break or lack features.
That last part is a procurement problem more than an AI problem. The pilots that actually worked stalled at scale because every new use case needed its own approvals and contracts. Each wrapper adds another round of that. We all love being slowed down my red tape and contracts, right? Nope.
The numbers around all of this are ugly. MIT's NANDA initiative, in its State of AI in Business 2025 report, found that 95% of enterprise generative AI pilots showed no measurable P&L impact, against an estimated $30 to 40 billion in spending (via Fortune, August 2025). On the vendor side, a SimpleClosure analysis cited by buildmvpfast puts 60 to 70 percent of AI wrapper products at zero revenue. Zero, not low. A thin shell over someone else's model is a hard thing to build a business on, and the customer is the one holding the contract.
Five questions
Fair is fair: some platforms truly earn their price. Real orchestration, real workflow logic, real domain knowledge sitting on top of the model. They exist. The sales deck just won't tell you which kind you're looking at, so you have to ask. Here's a few important questions to ask before you put a dent (many times, a very big dent) in your budget...
- Do you use the model's native connectors, or did you build your own integration layer? Show me the difference.
- What does this do that my current subscription can't?
- Where does our data go once it leaves your interface? Through your servers, or straight to the model provider?
- How do your uptime and security compare to the native enterprise offering? Put the SLA in writing for me.
- FInally, the one from the top. We cancel tomorrow. What do we lose?
A vendor who stumbles on that last one might be cooked, in my eyes. Buyer beware.
If you decide to leave the platform, and a lot of the work is housed within... how difficult will it be to get back out? What happens to the templates that you painstakingly built with all of the prompt engineering you tested and iterated on? Are you going to just get a folder of files and a bunch of xml/JSON/data that you'll need a consultant and a team of programmers to port to another platform? Is it worth the risk?
It's a new AI world, speeding faster than anyone can handle.
Everybody is building the plane mid-flight right now, and it's important to be mindful.
Before you buy somebody else's half-built one, go check what's already in your hangar. You may have the connectors already, and it might be easier to stay on the plane and tweak it than to strap on your parachute and jump out.