AI Readiness Audit · One page

Five questions before you fund the AI engagement.

Read this in 6 minutes. Answer each question yes / not yet. Score below. Recommended door at the bottom. The same questions we'd ask in a 30-minute call, written down so you can answer them yourself first.

  1. Q01

    Is the workflow named?

    AI work that lands is almost always tied to a specific workflow with a known owner. "AI strategy" without a named workflow tends to die in pilot. "Triage every inbound support ticket" or "draft the close-week status post" is what we'd build.

    ✓ What good looks like

    You can write the workflow on a sticky note: who does it, what tool they use, what triggers it, what the output is. Bonus if you can name the person who'd be most relieved to stop doing it.

    Not yet: start here

    If the answer is "AI for productivity" or "AI across the org," name a sub-workflow first. Pick the one a single person feels most. Build there.

  2. Q02

    Is the data accessible?

    AI systems are only as good as what they can read. If the data lives in three SaaS tools with no API, a shared drive nobody owns, and a Slack history nobody searches, the AI part is the easy half. The other half is data plumbing.

    ✓ What good looks like

    The system that holds the data has an API or an export, and someone has admin access. Knowledge bases are at least 60% accurate. PDFs, transcripts, and tickets are findable by date or ID.

    Not yet: start here

    If the data hygiene story is "we'd need to clean it up first," cleaning it up is the project. AI on a messy KB drafts a messy answer faster.

  3. Q03

    Is there a human in the loop?

    Most production AI work in small businesses runs as a draft-and-review pattern, not full autonomy. The system proposes; a person reviews. Determining where the human checkpoint goes is a Day 1 design choice, not a Phase 2 question.

    ✓ What good looks like

    You can name who reviews the output, when, and how long it takes. "The senior accountant signs off before invoices post" or "the support rep edits the draft before sending": concrete, named, time-boxed.

    Not yet: start here

    If the answer is "the AI just handles it," you're describing autonomy, which has a higher bar for accuracy and observability than most teams want to underwrite. Pick a draft-and-review version first.

  4. Q04

    What's the governance posture?

    Three questions the legal, security, or compliance lead will ask in week two: where does the data go? Which models are you using? What's the rollback plan if it produces something bad? Having opinions on these now is faster than getting them later.

    ✓ What good looks like

    You can answer: "Data stays in our infrastructure or a vendor we've vetted." "We're using a named model from a named provider." "We have a kill switch and an eval queue." Specifics, not vibes.

    Not yet: start here

    If governance comes up only as a blocker, a 1-week governance scoping pass costs less than a 6-week post-launch audit. Do it before Phase 1, not after.

  5. Q05

    Have you tried this yourself yet?

    The fastest credibility test is whether someone on your team has tried the workflow with ChatGPT, Claude, or a vendor demo for 20 minutes. Just to feel where the model is good and where it isn't.

    ✓ What good looks like

    You've prompted the workflow yourself. You can describe what the model gets right ("the tone is fine") and what it misses ("it doesn't know our refund policy"). That gap is the engagement.

    Not yet: start here

    If the team hasn't poked at it directly, do that this week. Even 30 minutes shrinks the design space. You'll know whether the gap is small or large before you scope the build.

Score yourself

Count yeses. Find your band. The recommended door is where teams in similar shape have had the cleanest first engagement.

5 of 5 yes

→ Custom Development engagement

You have a named workflow, accessible data, a clear human-in-the-loop, governance opinions, and direct experience with the model. This is a build-grade setup. Talk to us about a Custom Development engagement (typically 8–12 weeks, fixed-fee).

3–4 yes

→ Advisory engagement first

You're close, but one or two foundations need work, most often data access or governance. A 2–3 week Advisory engagement scopes the right first system, makes a buy/build/wait call, and produces a one-page architecture sketch you can act on.

1–2 yes

→ Workplace Agent

The workflow is fuzzy or the data isn't ready. That's a different first door. Workplace Agent in your Slack lets you build the AI muscle on starter channels (digest, research desk, internal Q&A) while you figure out the bigger build. Coming soon.

0 of 5

→ Don't start yet

Honest answer: don't underwrite an AI engagement until at least one of these is yes. The real risk is that you'll spend on something that won't land. Spend a week getting question 1 to yes, then revisit.

Want a second pair of eyes on your answers?

Reply to the email this came from, or use the contact form. 30-minute first call, no slides, honest read on whether we're a fit.

Talk to us See the Services process