AI Readiness Assessment Checklist: What to Measure Before You Commit to a Platform

Most AI readiness content hands you a checklist for your own operation. Is your data clean? Is your infrastructure on the cloud? Have you named an owner? Those are fair questions, and we work through them in our guide to the AI readiness assessment for operations. But there is a second checklist almost nobody writes, and it is the one that protects your budget once you start spending. Before you sign with any platform or consultant, you need an AI readiness assessment checklist for the vendor itself: what to measure about the tool you are about to trust with a decision worth 10 to 40% of revenue.

This is that checklist. It does not tell you whether your operation is ready. It tells you whether the platform in the demo will give you decision-grade data, or just another dashboard.

The short version

• The readiness checklist most people run is about your operation. This one is about the platform you are evaluating to inform the decision.
• Ten criteria separate a tool that produces a defensible answer from one that produces a confident-looking chart.
• The check buyers skip most often is the one that matters most: does it measure what people actually do, or only which app was open?

Before the checklist, one question: what decision is this platform informing?

You are not buying software. You are buying the data behind a decision. Restructure too aggressively and you lose institutional knowledge you cannot rehire. Move too slowly and a competitor automates first. 55% of companies that made AI-driven workforce changes say they regret it (the common thread was “we acted too quickly”). Those were not bad leaders. They were working from data that could not carry the weight of the call.

So evaluate every platform against the decision, not against its feature list. A tool can be genuinely impressive and still be the wrong instrument for a 10 to 40% revenue question. The right test is simple to state and hard for most vendors to pass: at the end of the engagement, will I be able to sit in front of my board and defend a specific recommendation with evidence I can trace? If the answer is a dashboard and a maturity score, you have bought oversight, not a decision. The checklist below exists to surface that gap before you sign, not after.

The 10-point platform-evaluation checklist

Run every vendor, consulting firm, and analytics tool through these ten checks. Each one names the check, why it matters, and the evidence to demand on the call. Do not accept a verbal “yes.” Ask for the artifact.

A platform does not need a perfect score on all ten. A free infrastructure framework will fail the ownership and decision-grade checks on purpose, because that is not its job. What matters is that you know which checks the tool passes before you decide what you are asking it to do. The failure mode is not picking the wrong category. It is committing headcount to an answer from a tool that was never built to produce it.

The one check buyers skip: does it measure what people actually do?

Nine of these ten checks get at least a glance in most evaluations. Check one, the unit of analysis, is the one buyers wave through, and it is the one that quietly decides whether the whole engagement was worth it.

Here is why it slips past. Almost every platform can show you activity. It will tell you your team spent six hours in Salesforce and two in Excel. That looks like insight. It is not the input an automation decision needs. You cannot automate “six hours in Salesforce.” You can automate “47 minutes a day copying values from an email into an account form.” One is an application label. The other is the work. This is the difference between app-level data and ground truth data, and it is the line between a tool that describes your operation and a tool that can tell you what to change.

Ask any platform the plain version of check one: show me one real, redacted day for a single person, and tell me the tasks and the minutes on each. If the answer is application names and time totals, you have a productivity dashboard. If the answer is named tasks (“record lookup,” “data entry into the policy form,” “manual reconciliation”) with minutes and an automation score against each, you have something you can build a decision on. Our Capture, Classify, Insight approach exists to produce exactly that: not “in the claims system” but “entering customer data into an account form,” classified automatically, for every person, every day. The work has not disappeared. In most operations it is simply invisible at the app level, and invisible work cannot be prioritized.

How to run the checklist in a demo

A checklist is only as good as the questions it turns into. Here is how to run these ten checks live on a vendor call, in the words that make a polished demo tell you something real. Ask for the artifact every time.

• “Show me one real, redacted day of output for a single employee. What are the tasks, and how many minutes each?” This tests unit of analysis and decision-grade output in one move. Slideware cannot fake a genuine day narrative.
• “Walk me through exactly what you capture on the machine, and what you never capture.” This tests capture method and privacy together. A confident answer names click-region over full-screen and a keystroke policy without being prompted. Our privacy architecture is built to answer this before HR asks.
• “Is every person in scope measured every day, or are you sampling and extrapolating?” This tests coverage. Sampling is not disqualifying, but you need to know you are getting an estimate, not a census.
• “How do you know your classifications are right? Show me your accuracy validation.” This tests proof. Any vendor can classify. Fewer can tell you how often the classification is correct.
• “When do I get first findings, and when do I get the final deliverable? Put dates on it.” This tests time to insight against your mandate clock. Vague timelines are a real answer, and the answer is no.
• “When the engagement ends, who owns the data, and how do I take it with me?” This tests ownership and exit. The reaction to this question tells you as much as the answer.
• “What do you need from my environment to start?” This tests integration fit early, before you have fallen for the output and discovered your stack cannot host it.

Write down which checks each vendor passes cleanly, which they dodge, and which they fail outright. By the third call the pattern is obvious, and you are no longer comparing marketing. You are comparing evidence.

From checklist to decision

The point of this AI readiness assessment checklist is not to disqualify tools. It is to match the instrument to the decision so you are not surprised six months in. A free framework is right for an infrastructure question. A consulting firm is right when the challenge is genuinely board-level strategy and you have the runway. A productivity platform is right for ongoing oversight. If you want the honest breakdown of which category fits which job, our guide to the best AI readiness assessment tools lays it out without the sales gloss.

But if what you actually have is a decision to defend, a mandate to answer, a number to challenge, then weight checks one, seven, and ten the heaviest: does it measure the real work, does it produce a defensible recommendation, and do you keep the data. That is the readiness that determines whether the rest matters. Measure twice, cut once.

If you want to pressure-test a platform against your specific mandate, book a 30-minute strategy call. We will walk through what a task-level baseline of your operation would look like, and what it would let you say in the room where the decision gets made.

FAQ: AI Readiness Assessment Checklist

What should an AI readiness assessment checklist actually measure? Two different things, and buyers conflate them. An operational readiness checklist measures whether your organization is ready (data, infrastructure, ownership, governance). A platform-evaluation checklist, the one in this article, measures the vendor you are about to trust with the decision: its unit of analysis, capture method, methodology transparency, privacy architecture, coverage, time to insight, whether the output is decision-grade or just a dashboard, its validation, its infrastructure fit, and who owns the data at the end. If you only run the first checklist, you can still commit budget to a tool that was never built to answer your question.

What is the difference between assessing readiness and evaluating a platform? Assessing readiness looks inward at your operation and asks whether you are prepared to act. Evaluating a platform looks outward at the vendor and asks whether the tool can produce an answer you can defend. Both matter, but they are separate exercises. A perfectly ready operation can still buy the wrong platform, and a strong platform cannot rescue a decision if you never checked what it actually measures. Run both checklists, in that order.

How do I know if a workforce platform gives decision-grade data? Ask for a redacted sample of one real day for one person, and look at the unit of analysis. If it reports application names and time totals (“six hours in the claims system”), it is a productivity dashboard. If it reports named tasks with minutes and an automation score against each (“47 minutes re-keying data between two systems”), it is decision-grade. Decision-grade data is traceable, task-level, and specific enough that you could put a single recommendation in front of your board and defend how you got there.

What privacy questions should I ask an AI readiness vendor? Ask exactly what is captured and what is never captured, and get it in writing. Confirm whether it captures the click region or the full screen, whether it logs keystrokes, how consent is handled, how data is encrypted, and who owns the data. The strongest answer is the least invasive one: click-region capture over full-screen recording, no keystroke logging, customer-owned data, and a clear consent workflow. This is workforce intelligence, not employee monitoring, and the vendor should be able to show you the difference.

How long should an AI readiness assessment take? Long enough to measure the real work, short enough to beat your mandate. Traditional consulting engagements typically run 12 to 18 months, which often lands the answer after the decision is already due. A modern activity-based assessment produces a full ground-up baseline in about 90 days, with first findings inside the first few weeks. Whatever the vendor claims, get the timeline to first insight and to final deliverable in writing, with dates, and hold it against the date your decision is actually due.