Starting with AI: A Practical Guide for Business Leaders

TL;DR

If you're a leader thinking about "doing AI" this year, the biggest risk is starting in the wrong place. This guide gives you a 12-question readiness assessment grouped into four dimensions — data, workflow, people, and decision authority — and a simple scoring rubric to tell you whether to build, buy, or wait.

---

Why most AI starts wrong

I've sat in ~40 kick-off meetings in the last two years. The pattern is almost always identical: a leader read something about GPT, a peer's company shipped a chatbot, the board is asking, and now there's a mandate to "do AI by Q3."

The team then picks a use case that sounds impressive, hires a vendor or a consultant, and six months later has a demo that no one uses. Sound familiar?

The failure mode has nothing to do with the model. It has to do with skipping the readiness step. AI is not a technology you deploy; it's a workflow change you enable. If the workflow isn't ready, no amount of GPT-4 will save you.

The four dimensions of AI readiness

Before you scope any AI project, honestly answer 12 questions across these four dimensions. If you score below 60% in any single dimension, that's where you start — not with the AI itself.

1. Data readiness (3 questions)

• Is the data the AI will use findable? (Not "does it exist somewhere on SharePoint" — findable in seconds via a documented source.)
• Is it clean enough that a new hire could use it without asking three people what a column means?
• Do you have permission to use it for AI? (This is where GDPR, DPA, and customer contracts come in — and where 30% of pilots get killed at legal review.)

2. Workflow readiness (3 questions)

• Can you draw the current workflow on a whiteboard in under 5 minutes, step by step, with the human decision points marked?
• Do you know which step in that workflow actually costs time or money? (Most teams pick the wrong step — the one that's annoying, not the one that's expensive.)
• Will the humans doing that step want the AI to help, or will they route around it?

3. People readiness (3 questions)

• Is there a named owner on the business side (not IT) who wakes up worrying about this outcome?
• Do the end users have 10 minutes a week to give feedback during the first three months? (No feedback = no adoption.)
• Does someone technical already understand your data? (Bringing in a vendor who has to learn your data adds 3 months. Every time.)

4. Decision-authority readiness (3 questions)

• Is the person sponsoring this allowed to change the workflow if the AI works, or do they need six sign-offs?
• Is there a budget line for month 7–12 (production), not just months 1–6 (pilot)?
• Is there a clear kill criteria — the numeric bar the pilot has to hit to move forward, agreed upfront?

Scoring: what to do with the answers

Count how many questions you can answer "yes" to, per dimension:

• 3/3 in every dimension → Green light. Scope the project.
• 2/3 in 1–2 dimensions → Yellow light. Fix the gap first (usually 4–8 weeks), then start.
• 0–1 in any dimension → Red light. Do NOT start an AI project here. Fix the underlying dimension first, or pick a different use case where the score is higher.

The reason this matters: you can't fix data readiness with a better model. You can't fix decision authority with a better prompt. And you definitely can't fix people readiness with a slicker UI.

The trap: picking the impressive use case over the ready one

Here's the pattern I see most often. The leader lists five potential use cases. The one that scores highest on the readiness assessment is boring — say, automating an internal report that three analysts spend Fridays on. The one that scores lowest is exciting — say, a customer-facing agent that handles support.

Guess which one gets picked?

The boring one would ship in 8 weeks, save 400 analyst-hours a year, and build organizational muscle for the next project. The exciting one takes 9 months, gets stuck on data access, and quietly gets shelved after the pilot demo.

Start with the boring one. You get a win, you learn how your organization actually implements AI, and you earn the right to try the ambitious one next.

A concrete first project template

If you're at green-light readiness and just need a shape for your first project:

1. Weeks 1–2: Formalize the workflow map + measure the current cost (time, money, error rate).
2. Weeks 3–4: Build the smallest possible AI prototype — internal, no polish, one user.
3. Weeks 5–6: Have that one user work with it daily. Instrument every interaction.
4. Weeks 7–8: Decide: expand, adjust, or kill. Do not extend the pilot without a decision.

Eight weeks. One user. One workflow. That's your starting point — not a company-wide transformation.

What to do this week

If you take one thing away: run the 12 questions. Write your answers down. Show them to your team. If two of you disagree on any answer, that's the actual first thing to fix — before anything AI-related.

Most companies skip this and go straight to picking a vendor. That's why most AI projects fail. Yours doesn't have to.

---

If you want help running this assessment with your team, book a call — we do this in a 90-minute workshop.