Enterprise AI readiness assessment: the 5 checkpoints you must pass before going live (with self-evaluation criteria)
Should we actually start doing AI? Not based on gut feeling or competitor anxiety. This breaks the decision into five checkpoints you must pass before launch, each a yes-or-no self-evaluation question. It transforms unclear worry into a concrete checklist covering data, workflow, people, and governance. The sequence can't be changed, and this is where most organizations stumble.
By
Tenten AI FDE 團隊
導入方法論
Published
September 18, 2025
Read time
6 分鐘
Six months ago, a manufacturing client asked us a straightforward question: should they start using AI?
Behind that question was anxiety. They'd watched competitors announce their AI initiatives, read piles of white papers, and their CEO had been asked about AI progress three times in board meetings. But nobody could articulate one central concern: if we start now, will this become one more system that nobody uses after three months?
We didn't give them a proposal. We gave them a checklist.
What is an AI readiness assessment
An AI readiness assessment is a way to figure out whether your organization's data, processes, people, and governance can support a working AI system before you commit budget and staff to it. It doesn't ask what AI can do. It asks whether you're actually ready for it.
We break it into five checkpoints. Each one is a yes-or-no question, not multiple choice. If you can't pass even one, the problem isn't your model. It's the checkpoint. More than half the companies we work with get blocked somewhere that has nothing to do with AI technology.
The 5 checkpoints and self-evaluation criteria
| Checkpoint | Core Question | Passes If (check off only if true) | Cost of Failing |
|---|---|---|---|
| 1. Problem Definition | What's the specific, measurable pain point you're solving? | You can say in one sentence: who, in what scenario, saves how much time or eliminates which error, and that scenario happens dozens of times per week | You build a feature nobody has reason to use |
| 2. Data Readiness | Does the data exist, is it accessible, and is it clean enough? | Critical data has a clear source and owner, permissions are secured, it's been updated in the last three months, and the format is consistent | RAG retrieves outdated or contradictory answers |
| 3. Workflow Integration | Where does the AI output actually plug in? | You can point to exactly which system and which step users use today, and they don't need a new habit to use it | Single-digit adoption rates |
| 4. People and Adoption | Is there someone on the ground who'll own adoption? | A designated business-side owner (not just IT) commits to watching usage data and feeding back for two months after launch | Peak on day one, then slow death |
| 5. Governance and Risk | Can you afford a mistake, and can you trace what happened? | You've clearly defined acceptable error ranges, manual review points, and who's accountable for the output | One hallucination shuts down the whole project |
The table looks simple. Be honest when you fill it out. When we work through this with clients, we almost always hit a wall at checkpoint one: what they call a pain point is actually three separate problems bundled together, and none of them hurt enough to make people change their behavior.
Why you can't skip ahead
You go through these checkpoints in order, front to back. No jumping around.
Years ago, we had a project with excellent data and a model that performed well. We got excited and skipped past checkpoint two. Checkpoint one, which was problem definition, was never really clear. Make the sales team more efficient. After launch, we discovered that each salesperson meant something different by efficient. No matter how accurate the system was, nobody felt it was built for them. We ended up with single-digit adoption.
Each checkpoint is the foundation for the next. If you don't nail the problem, you won't know which data you actually need. If the data is poor, talking about workflow integration is empty. If you haven't worked out the workflow, whoever you assign to drive adoption will be the wrong person. Projects that skip checkpoints don't fail faster. They fail more expensively. You're building on unstable ground.
Do all five need to pass before you launch
No. This is the biggest misunderstanding we see.
If you wait for all five to be completely ready, you'll wait forever. Most companies don't have checkpoint two, data, fully handled. Readiness assessment doesn't issue a pass-fail grade. It tells you which checkpoint to tackle first and whether it's worth tackling now.
Checkpoints one and four are absolute stops. If you're stuck there, don't start engineering. Lock down your problem and your people first. Checkpoints two, three, and five can have gaps when you launch, as long as you put those gaps on your to-do list before launch, not after. A project with 70 percent clean data, a clearly defined problem, and one person on the team who desperately wants it will ship faster and go further than a project where all five checkpoints are fully checked off.
From anxiety to checklist
That manufacturing client got stopped at checkpoint three. They couldn't find a natural spot for it in their daily work. We scaled back the scope. We found something that happens every single day, where the output goes straight back into their existing work order system. That smaller version hit over 60 percent actual use by month two after launch.
This is what we do before we start any project at Tenten: engineers stay out at first. We sit with the client and work through all five checkpoints together. A polished demo doesn't mean anything. Real usage, every day, is what counts. That gets decided the moment you're honest about these boxes.
One stuck workflow
is enough to begin
Tell us what the team does today, where it breaks down, and what a better working day should look like.