Understand Your Customer's AI Adoption Readiness Like an FDE
By the time you've deployed your product into a few dozen customer environments, you start to notice a pattern. Some deployments go from kickoff to production value in three weeks. Others limp along for six months and quietly die. And the difference almost never comes down to how good your model is. It comes down to how ready the customer actually is to adopt AI - and whether someone assessed that honestly before the building started.
As a Forward Deployed Engineer at an AI-native company, assessing readiness isn't a phase you run. It's something you do constantly, from the first call, mostly by watching. So here's what you should actually be looking for. This framework will save you from a lot of doomed projects - whether you're an FDE, a founder selling into enterprises, or a buyer trying to figure out if your own org is ready.
Readiness is not enthusiasm
Get this out of the way first: excitement is not readiness. Some of your most painful deployments will start with your most enthusiastic customers. The VP who's read every AI newsletter, who says "we want to be an AI-first company," who has already picked out three use cases - treat that person as a warning sign, not a green light. Enthusiasm without infrastructure just means you'll build fast and break in production.
Real readiness is quieter. It looks like a team that can hand you a clean dataset on day two without a three-week legal detour. It looks like someone who, when you ask "how will we know if this is working?", has an actual answer instead of a blank stare. Readiness is boring, operational, and unglamorous. Learn to love the boring signals.
The five things to check in the first two weeks
Boil your early assessment down to five dimensions. Don't run them as a questionnaire - customers can sense a checklist and start performing for it. Surface them through normal working conversations and, more importantly, through small tests.
1. Can you actually get to the data?
This is the one that kills more AI projects than anything else. Not "do they have data" - everyone has data. The question is whether you can access it, whether it's coherent enough to be useful, and whether getting a sample requires one Slack message or a quarter of procurement and privacy review. On your second or third day, ask for a small, real sample of the exact data the product will run on. How that request goes tells you more than any architecture diagram. If it takes two weeks and comes back as a screenshot of a spreadsheet, you already know what you're dealing with.
2. Do they know what "good" looks like?
AI-native products live and die on evaluation. In traditional software, "working" is binary - the button submits the form or it doesn't. With your product, "working" is a judgment call, and if the customer can't make that call, you're both flying blind. So early on, get them to define success in concrete terms. Can they show you ten examples of a great output and ten examples of a bad one? Can they tell you why one is better? Customers who can do this are ready to build an eval loop with you. Customers who say "we'll know it when we see it" are telling you the project has no steering wheel.
3. Where does the output go, and who trusts it?
An AI output that lands in a dashboard nobody checks is worth nothing. Trace the whole path: the model produces something, and then what happens? Who reads it, who acts on it, and critically - do they believe it? This is where AI readiness diverges sharply from regular software. People have to build calibrated trust in a system that's occasionally, confidently wrong. If the customer's plan is "the AI decides and humans stay out of it" on day one, they're not ready. Mature customers design a human-in-the-loop first and earn their way to more autonomy.
4. Is there someone who can integrate?
You can do a lot, but you're not there forever, and you can't own their auth, their infra, or their internal systems permanently. Look for the counterpart - one engineer or technical person on their side who's genuinely engaged, not just cc'd. If you're doing an integration and nobody on the customer side can tell you how their SSO works or who owns the API gateway, that's a readiness gap dressed up as a scheduling problem.
5. Who's the sponsor, and what happens when the first output is wrong?
Every AI deployment produces embarrassing outputs early. That's not a bug, it's the nature of the technology - you calibrate over time. The question is whether the organization has the stomach for it. Is there an executive sponsor who framed this internally as "we're going to iterate toward something great," or did someone promise the board a finished miracle by Q3? The first environment survives the inevitable ugly demo. The second one gets cancelled the moment the model hallucinates in front of the wrong person.
The tells that don't show up on a scorecard
Beyond those five, there are softer signals you should learn to read.
The way a customer talks about failure is enormously revealing. Teams that are ready talk about the AI making mistakes as a normal, expected thing to design around. Teams that aren't ready either pretend it won't happen or treat every error as a catastrophe. Neither denial nor panic is a good sign; you want calm realism.
Watch how they respond to a rough first version, too. When you ship something early and imperfect - which you should do on purpose - do they engage with it and start giving you specific, useful feedback? Or do they go quiet and wait for "the real thing"? A customer who rolls up their sleeves on version zero is a customer who'll get to production. A customer who needs it polished before they'll touch it will never accumulate the feedback loops that make AI products actually good.
And pay attention to how many people are in the room versus how many are doing the work. Big steering committees with no hands on keyboards are a classic not-ready pattern. One motivated champion who can make decisions beats a committee of twelve every time.
Final Thoughts
If you're on the buying side and reading this nervously - good. The best customers are the ones who assessed their own readiness honestly before the FDE ever showed up. What would one notice about your organization in the first two weeks?