What is an AI readiness assessment?
An AI readiness assessment is a scored diagnostic of whether a company can adopt a specific workflow — data in a known place, a named owner, refusal rules, and a measurement plan — before it buys tools or funds a build. McKinsey's Superagency in the Workplace survey of 118 US CxOs (October–November 2024) found that only 1% of leaders call their companies "mature" on gen AI deployment, even as 92% plan to increase AI investments over the next three years. It is a stop/go on one workflow, not a badge. This post sits in the GTM Teardown hub because readiness is a go-to-market constraint, not an IT maturity pageant.
That 1% figure is a C-suite, larger-organization reading. Owner-led companies should not treat enterprise "mature" as the bar. The useful question is narrower: can this team run this workflow without inventing data, owners, or rules after the invoice?
McKinsey's later State of AI (November 2025) makes the same split in different numbers: 88% of organizations use AI, 39% report any EBIT impact. Buying a tool is easy. Capturing value is a workflow-redesign problem. The March 2025 companion found workflow redesign had the biggest EBIT effect of 25 attributes tested, while only about 21% of organizations had redesigned even some workflows.
Why run an AI readiness assessment before buying tools?
An AI readiness assessment exists because "we use AI" and "AI moved a number we care about" are different states. Teams feel a lag — quotes too slow, an inbox that never clears, no one able to find last quarter's numbers — and reach for a chatbot, a copilot seat, or a custom agent. That is the same symptom-chasing pattern What is a GTM teardown is built to interrupt: a loud pain, a purchased tactic, no diagnosis of the leak.
As McKinsey's Superagency research put it:
while nearly all companies are investing in AI, only 1 percent of leaders call their companies “mature” on the deployment spectrum
The mix behind that headline is not a rounding error: nascent 8%, emerging 39%, developing 31%, expanding 22%, mature 1%. Superagency also argues the biggest barrier to scaling is not employees but leaders. For a founder-led company that is not an insult. It is the diagnostic: if no one owns the workflow, the model will not invent the owner.
An AI readiness assessment that ends in a software shortlist has failed. The output is a scored map: adopt this workflow, fix this go-to-market leak first, or do not spend yet. If you are already gathering quotes, read how much AI consulting costs after you know which constraint you are paying to remove.
How does an AI readiness assessment map onto GTM teardown dimensions?
An AI readiness assessment inherits the teardown's chain: revenue (or delivery) moves only as fast as the weakest link, so you score the chain rather than the tool. Overlay a proposed AI workflow on each of the six dimensions and ask whether the workflow actually touches the leak — or just decorates a dimension that is already fine.
| GTM dimension | Readiness question for a proposed AI workflow |
|---|---|
| Positioning & message-market fit | Can you say, in one sentence, who this workflow is for and what it must never claim? |
| Demand generation | Will this change how you earn attention, or are you automating a channel you do not control? |
| Capture, qualification & routing | Where does the input land, who qualifies it, and what happens when the model is unsure? |
| Sales process & conversion | Does the workflow change a measured conversion step, or only the prose around it? |
| Activation & retention | Does the customer (or internal user) reach first value faster, and can you see drop-off? |
| Tracking & reporting | Can you name the baseline, the event, and the review cadence before anyone logs in? |
The sixth dimension is still the one people skip. If tracking is a guess, every other readiness score is a guess. An AI readiness assessment that cannot name a baseline is a vendor questionnaire with extra columns.
Do not bolt on Cisco's Infrastructure pillar and call the result "enterprise-grade." Cisco's AI Readiness Index 2025 surveyed 8,000 senior IT and business leaders at organizations with 500+ employees across 30 markets and 26 industries. Use that research as evidence that even well-staffed enterprises fail the "ready" bar — Pacesetters have sat near 13% for the last three years — not as a scorecard a 12-person company should copy. Skip GPU capacity and network-readiness pillars entirely at that size. Data location, owners, and measurement transfer. Fabric interconnects do not.
Gartner's AI Maturity Model is the other enterprise frame you will be sold: five stages (Awareness → Active → Operational → Scaled → Transformational) across seven pillars (strategy, data, governance, engineering, operating model, culture, AI product/value). Useful language for a CIO program. A vanity stage label is not a buying decision for an owner-led team.
Which AI-specific checks belong on an AI readiness assessment?
An AI readiness assessment that only scores "do we have a strategy slide" will pass companies that still cannot locate the spreadsheet. Keep the extra checks to four, and make them binary enough to argue about.
1. Data location. Where does the source of truth live for this workflow — CRM, inbox, shared drive, someone's laptop? Cisco reports 64% of its 500+ employee respondents struggle to centralize data. That is an enterprise sample, and it still fails. A 12-person company does not need a data lake. It does need one named system of record for the workflow under review. If the answer is "it depends who you ask," you are not ready to automate.
2. Named owner. Who can change the prompt, the routing, or the stop rule without a meeting? Superagency's finding that leaders — not employees — are the scaling bottleneck is the owner check in different clothes. If the founder is the only person who can say yes, the workflow is not adopted. It is a demo.
3. Human-in-the-loop. Who reviews exceptions, and on what cadence? "The team will watch it" is not a loop. Name the role, the queue, and what happens when that person is out.
4. Refusal path. What must the system refuse, and where does the refusal go? Agents do not usually fail because the model is dull. They fail because nobody wrote a rule for the conversation they entered. That is the argument in agents fail on scope: write the refusal rule before the capability list, name the human who receives every escalation, and log the action. An AI readiness assessment that skips refusal is scoring a chatbot you cannot defend.
Score each check present / partial / missing. Rank the missings by downstream leverage — the same teardown rule. You cannot fix six things. You can refuse to buy until the missing owner or the missing baseline is real.
Which AI readiness assessment model should you use?
An AI readiness assessment comes in three common models. Only one of them is sized for an owner-led company about to spend.
| Model | Who it's for | What it scores | What to skip |
|---|---|---|---|
| DIY checklist | Owner-led teams scoring one workflow before a tool or a vendor conversation | Data location, named owner, human-in-the-loop, refusal path, measurement, and which GTM dimension the workflow actually touches | Cisco's six pillars as a self-score; GPU and network "readiness"; Gartner stage labels as a vanity badge |
| Consultant diagnostic | Teams about to fund a build or a retainer who cannot read their own label | The DIY checks plus an outside ranking of the constraint, with a stop/go on spend | Open-ended "AI strategy" decks; copying Cisco Pacesetter share (~13%) as a target |
| Cisco / Gartner enterprise model | IT and business leaders at 500+ employee organizations (Cisco Index) or CIO maturity programs (Gartner toolkit) | Cisco: Strategy, Infrastructure, Data, Governance, Talent, Culture. Gartner: five stages across seven pillars | Do not copy this scorecard onto a 12-person company. Skip Infrastructure (networks, GPUs) entirely at that size |
Cisco's AI Readiness Index is still worth reading as a warning, not a template. Its six pillars are Strategy, Infrastructure, Data, Governance, Talent, and Culture. Only 15% of those large-organization respondents said their networks were fully ready for AI, versus 71% of Pacesetters; 34% felt IT infrastructure was fully adaptable and scalable; 26% reported robust GPU capacity. Those numbers explain why a well-staffed enterprise can still fail "ready." They are not homework for a shop that has not named a CRM owner.
How to run the DIY version in one sitting
- Name the workflow in one sentence — who it serves, what it does, what it must never do.
- Name the GTM dimension you think is leaking — positioning, demand, capture, sales, activation, or tracking. If you cannot pick one, tracking is the working hypothesis.
- Locate the data — one system of record, or an explicit "missing."
- Name the owner — the person who can change the rule without a committee.
- Write the human-in-the-loop and the refusal path — role, queue, what must be refused, where refusals go.
- Write the baseline — the number you can see today, the event you will log, when you will look.
- Band each check present / partial / missing and rank missings by leverage.
- Write the spend decision — buy this, fix this first, or do not buy.
A DIY pass is better than flying blind, with two honest limits. You cannot read your own positioning, and you have no outside cohort to tell you whether "we feel messy" is normal or a hard stop. That is the gap a consultant diagnostic is for — not a forty-page maturity novel. If the quote you are holding is priced like a build, check how much AI consulting costs against a named workflow, not against "AI" as a category. If you want an outside scored pass, isonew's AI Opportunity Diagnostic is built as that stop/go on one workflow, not as a Cisco replica.
What should an AI readiness assessment produce?
An AI readiness assessment should produce a short scored map, not a transformation narrative. Minimum output:
- The workflow in one sentence.
- Four AI checks banded present / partial / missing — data, owner, human-in-the-loop, refusal.
- One GTM dimension named as the real constraint — or an explicit finding that the proposed workflow does not touch the leak you think you have.
- A baseline and a review event.
- A spend decision — buy this, fix this first, or do not buy.
If the document does not end with a spend decision, it is a brochure. McKinsey's 88% / 39% split is usage without EBIT; the March 2025 rewiring work is the corrective — value tracks workflow redesign, which only about 21% had done. Skip GPU scorecards, network-readiness quizzes, and Gartner stage labels you cannot act on this quarter. Keep the GTM Teardown hub nearby so the AI checks stay attached to the funnel they are supposed to change.