ai readiness model
most ai projects fail before the first prompt. not because the model is wrong.
each layer holds the one above. skip one and the model becomes a liability.
01 data is it captured, clean, and reachable? if not, the model invents the parts no one wrote down — confidently.
the model doesn't just consume numbers. it consumes definitions, policies, decisions, context — every block of knowledge the company carries. if a definition lives in someone's head, the model invents one. if it's inconsistent across teams, the model picks one and commits.
the smell test isn't volume. it's reconstructability. ask a senior person a non-trivial question about how the business works. if they rebuild the answer from memory, you don't have a data layer — you have tribal knowledge wearing a costume.
captured isn't stored. clean isn't tagged. reachable isn't "it's in there somewhere." the bar is a new hire — or a model — finding the right answer without translating, without asking, without guessing.
02 workflow is the work documented and stable? if not, ai makes a bad process faster — not better.
most companies don't document workflows. they document clicks. the loom shows where the buttons are, the sop tells you which fields to fill — none of it says what to decide. the judgment that makes the work valuable stays in someone's head.
the smell test is brutal. read your "documented" workflow. does it tell you when to discount and when to escalate? when to push back and when to concede? when a deal needs a vp and when it doesn't? if it only tells you which tool to open, you have a tutorial — not a workflow.
the bar isn't a process map. it's prediction. a new hire can read the doc and predict what a senior person would do — without watching anyone. if they can't, ai won't either.
03 ownership who owns the outcome and the failure mode? if nobody, it belongs to everyone — which means it belongs to no one.
the question isn't who owns the ai initiative. it's who owns the business outcome the ai is meant to move. those are different people. most companies have a head of ai. almost none have a clear answer to "who owned this customer onboarding number before — and will after?"
the smell test: name the metric the ai should improve. then ask who was accountable for that metric last year, before ai existed. if no one was, the ai doesn't have an owner. it has a sponsor. and sponsors don't survive bad quarters.
the bar isn't a committee. it's continuity. the owner was responsible for the outcome last year, will be next year, and is now — whether the ai works or not. the ai is a tool they're trying. they are the owner.
04 measurement is success defined in numbers, not vibes? if not, every ai investment becomes a faith-based bet — and faith doesn't survive a budget cycle.
the kpi written after the pilot ends isn't a kpi. it's defense. real measurement is the number you wrote down before launch — baselined before launch, with a target date attached. anything else is interpretation.
the trap is measuring the model instead of the outcome. accuracy, latency, cost per query — interesting. not the numbers that justify the spend. those live on the p&l, the customer calendar, or the team's clock. if the ai can't move a number that already mattered, it didn't matter.
the bar is one sentence: x to y by z. "time to close from 11 days to 6 days by q3." if you can't write that sentence before the project starts, you don't have a measurement problem. you have a project problem.
05 adoption will the team actually use it on monday? if not, you don't have adoption — you have access.
companies confuse access with adoption. they roll out the tool, send the email, count the logins, and call it deployed. none of that means a single person changed how they work. most enterprise ai lives in this gap — fully provisioned, mostly unused.
the smell test is one question: name a workflow that stopped being done manually last month because of ai. not "people are exploring it." not "engagement is up." a deck, a report, a task — something someone used to do by hand, and now doesn't. if you can't name it, you don't have adoption. you have a license.
the bar is one sentence, with a name and a number in it: "the qbr deck used to take maria 11 hours. last month: 90 minutes." that sentence is the bar. everything else is implementation theater.