---
name: analytics-pitfalls-before-the-ai-strategy
description: "Before the AI strategy: 18 cards from Analytics Pitfalls & Principles, by Brian Cariveau, to check an AI strategy, a pilot, an agent or a rollout."
license: CC-BY-4.0
---

# Before the AI strategy

From Analytics Pitfalls & Principles, by Brian Cariveau: 18 cards and 3 frameworks for an AI strategy, a pilot, an agent or a rollout. Every card, free: https://www.analyticspitfalls.com/library

## How to use it

Use it when someone shares or describes an AI strategy, a pilot, an agent or a rollout, before it goes any further.

1. Take Brian Cariveau's questions below in order, against their actual work, and ask them in his words rather than your own. For each, open the files it names and answer from them.
2. Skip a question the work already answers well. Say so in a line and move on.
3. Then pick the one or two cards that matter most for this work, say why in a sentence each, and give their principles in the cards' own words. Never hand someone the whole list.
4. Say the work looks like a pitfall; never tell someone they have it. If nothing fits, say so.
5. Then put it on the canvas, and offer the board (below).
6. Finish with the links of the cards you used, and say which skill comes next (Where this fits, below).

## Brian Cariveau's questions

1. How many times a day does this problem cost us money? Compare it to what people do today, not to nothing. A great demo doesn’t count. From: The Demo Trap (`cards/demo-trap.md`), The AI Hype Cycle Trap (`frameworks/ai-hype-cycle-trap.md`).
2. Show me the receipt. Not licenses, not logins, not hours saved on a survey. What do people actually do differently now? From: Licenses and Lip Service (`cards/licenses-and-lip-service.md`), The 41,000 Hours (`cards/41000-hours.md`), Usage Is the Receipt (`frameworks/usage-is-the-receipt.md`).
3. Are we asking the genie for a cup of coffee? Aim it at something big enough to matter. From: The Cup of Coffee Wish (`cards/cup-of-coffee-wish.md`).
4. What data does it read, and has anyone actually defined it? Planned isn’t in stock. From: Planned Isn't In Stock (`cards/planned-isnt-in-stock.md`), Four Hundred and Twelve Boxes (`cards/four-hundred-and-twelve-boxes.md`).
5. Who checks its answers, and against what tolerance? And who’s watching the weekend it runs on its own? From: Is It Right? (`cards/is-it-right.md`), Legal Reads Everything (`cards/legal-reads-everything.md`), The Runaway Weekend (`cards/runaway-weekend.md`), The Factory With the Lights Off (`cards/factory-with-the-lights-off.md`).
6. Will a person actually read what it writes before it goes out the door? Confident and nicely formatted doesn’t mean anyone checked it. From: Confident, With Formatting (`cards/confident-with-formatting.md`).
7. Does it speed up one step while the customer still waits? Or is it paving the cow path, the old way only faster? From: Faster Saws (`cards/faster-saws.md`), The Porsche at the Front (`cards/porsche-at-the-front.md`), Faster at the Old Way (`cards/faster-at-the-old-way.md`).
8. Who does it lean on, and who goes first? One expert with a line outside their door is a bottleneck. Start with a respected skeptic, not the fan club. And if the boss still asks for the old report, nothing changes. From: The Line Outside Elmer's Door (`cards/line-outside-elmers-door.md`), Everything Runs Through You (`cards/everything-runs-through-you.md`), The Back Row (`cards/back-row.md`), The Drain at the Top (`cards/drain-at-the-top.md`).
9. Now get past the if-then part. Whose job changes, what do they have to trust, and what are they giving up? The agent is the easy part. From: It Was Never About the Tech (`frameworks/it-was-never-about-the-tech.md`).
10. Who’s on your board? Before you build, convene a few people who’d poke holes in it, and hear them out now. It’s a lot cheaper than hearing it later.

## Put it on the canvas

When the questions are done, shuffle the strategy onto one page: four boxes. This isn't the business lean canvas. It's People, Process, Technology and Agentic.

- **People**: Who has to change, who leads it, who pushes back, and what they have to learn or give up.
- **Process**: How the work flows, where it waits, and what stops.
- **Technology**: The platforms, the data and the definitions it runs on.
- **Agentic**: Anything in the new agentic world: which agents, what they do on their own, the tolerance they work within, and who watches them.

Three to five short lines a box, in the person's own words where you can. Under each box, one line for the adaptive part: what has to change in people for that box to work.

Draw it as a visual. Where you can make an artifact or a web page, fill in `canvas.html`: replace every {{…}} (each _LINES with `<li>` items) and keep the credit line. Anywhere else, draw it as a two-by-two table.

## Convene the board

Offer it, don't spring it. If they say yes, play each member in turn: each asks their hardest question about the canvas and says what would change their mind. Then say what you'd change on the canvas, and draw it again.

The board, until they swap anyone (ask):

- The CFO
- Legal
- The most skeptical leader in the building
- Someone whose job it changes
- Whoever gets the call when it breaks on a Sunday

## Where this fits

These run in order: the strategy, then the work, then the results.

1. **The strategy**: Before the AI strategy (`analytics-pitfalls-before-the-ai-strategy`), Before the data strategy (`analytics-pitfalls-before-the-data-strategy`), Before the BI strategy (`analytics-pitfalls-before-the-bi-strategy`). Pick the one you’re writing. They sit side by side; there’s no order between them. This one is here.
2. **The work**: Before the dashboard (`analytics-pitfalls-before-the-dashboard`). Once for each dashboard the strategy calls for.
3. **The results**: Before the readout (`analytics-pitfalls-before-the-readout`). When the first numbers go to leadership.

When you're done, say what comes next, by name: Before the dashboard (`analytics-pitfalls-before-the-dashboard`). Once for each dashboard the strategy calls for.

## The cards

- **The Demo Trap** (`cards/demo-trap.md`): pitfall, Adoption & Product. Demo success ≠ production value. The most impressive technology often solves problems that don't exist, while boring solutions fix problems that cost real money every day. At work: "The AI demo went great. Leadership loved it."
- **Licenses and Lip Service (When Everyone Logs In and Nobody Changes)** (`cards/licenses-and-lip-service.md`): pitfall, Adoption & Product. Usage is the receipt. Logins, licenses and course completions are promises. Daily use on real work is proof. At work: "Everyone’s got a license, and almost everyone finished the training."
- **The 41,000 Hours (Saved From What, Spent Where?)** (`cards/41000-hours.md`): pitfall, Metrics & Evidence. Never skip the middle. Engagement, experience, effectiveness, efficiency: in that order, every time. At work: "AI saved us 41,000 hours this year."
- **The Cup of Coffee Wish (Asking a Genie for Small Things)** (`cards/cup-of-coffee-wish.md`): pitfall, Adoption & Product. Ask for the big thing first. Tell it what you want at the end, not the steps to get there, and then check everything.
- **Planned Isn't In Stock (When an Agent Reads Data Nobody Defined)** (`cards/planned-isnt-in-stock.md`): pitfall, Data Foundations. Customer-facing agents read only from defined, trusted data. If a word matters to the business, write down what it means before an agent acts on it. At work: "Just point the agent at the shared drive."
- **Four Hundred and Twelve Boxes (Building the AI Ontology From the Top Down)** (`cards/four-hundred-and-twelve-boxes.md`): pitfall, Data Foundations. Simple for humans is simple for agents. Build the map one trusted domain at a time, in words a new hire would understand.
- **Is It Right? (The Wrong Question to Ask of AI)** (`cards/is-it-right.md`): pitfall, Metrics & Evidence. Ask whether it's within tolerance, and how you know. Nothing is right every time, in a factory or in AI.
- **Legal Reads Everything (When Checking Every Answer Catches Less)** (`cards/legal-reads-everything.md`): pitfall, Delivery & Execution. Decide what safe enough looks like before you need it. Check by risk, not by habit.
- **The Runaway Weekend (From Checking Every Email to Checking None)** (`cards/runaway-weekend.md`): pitfall, Delivery & Execution. In the loop, then on the loop, then learning loops. Never from a human approving every step to nobody at all. At work: "It’s been perfect for three weeks. We can stop checking it."
- **The Factory With the Lights Off (When Nobody Can See the Agents)** (`cards/factory-with-the-lights-off.md`): pitfall, Strategy & Focus. Turn the lights on. Every agent and shared skill goes on a screen, with an owner, a backup and a usage count. At work: "I built my own agent for that."
- **Confident, With Formatting (When Nobody Reads What the AI Wrote)** (`cards/confident-with-formatting.md`): pitfall, Metrics & Evidence. Your name is on it, so read it. Checking is faster than doing, and it is never zero.
- **Faster Saws (Every Step Got Faster, and the Customer Still Waits)** (`cards/faster-saws.md`): pitfall, Delivery & Execution. Faster stations aren't a faster factory. Walk one real order through the business, color the work blue and the waiting red, and the red wins by a mile almost every time.
- **The Porsche at the Front (When the Digital Side Outruns the Factory)** (`cards/porsche-at-the-front.md`): pitfall, Strategy & Focus. Put the slowest car at the front. The whole system goes as fast as its slowest step, so let that step set the pace for everything.
- **Faster at the Old Way (Paving the Cow Path With AI)** (`cards/faster-at-the-old-way.md`): pitfall, Dashboards & Visualization. Stop asking for the deck. Ask for the answer, and what the person would do about it. At work: "AI can build us a new report every day now."
- **The Line Outside Elmer's Door (When the Whole Plan Waits on One Expert)** (`cards/line-outside-elmers-door.md`): pitfall, Leadership & Teams. Codify your Brent. Write down what's in their head, in plain English, and give it to everyone. At work: "Ask Dave. He’s the only one who knows how it works."
- **Everything Runs Through You (When the AI Lead Becomes the Bottleneck)** (`cards/everything-runs-through-you.md`): pitfall, Leadership & Teams. One person is one tap. If every path to the new way of working runs through you, the first Brent to codify is you.
- **The Back Row (When Only the Enthusiasts Stand Up)** (`cards/back-row.md`): pitfall, Adoption & Product. It moves from the front row. When someone people watch stands up, the room follows.
- **The Drain at the Top (When the Boss Still Wants the Old Report)** (`cards/drain-at-the-top.md`): pitfall, Leadership & Teams. The drain goes all the way to the top. If you want to change what hundreds of people do, change what their leaders ask for.

## The frameworks

- **The AI Hype Cycle Trap** (`frameworks/ai-hype-cycle-trap.md`): The AI Hype Cycle Trap is building the impressive AI demo instead of the boring, high-value fix — because demo success and production value are not the same thing, and the gap between them is where budgets go to die.
- **Usage Is the Receipt** (`frameworks/usage-is-the-receipt.md`): Usage Is the Receipt means the only proof an analytics product worked is whether people actually use it — adoption is the receipt, everything else is a claim.
- **It Was Never About the Tech** (`frameworks/it-was-never-about-the-tech.md`): It Was Never About the Tech is the observation that analytics transformations fail on politics, people, and principles far more often than on tooling — the technology is rarely the binding constraint.

---

© 2026 Brian Cariveau, Analytics Pitfalls & Principles. Free to share and adapt under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/): keep this credit, "Brian Cariveau, Analytics Pitfalls & Principles (https://www.analyticspitfalls.com)", and say if you changed it.
The names Analytics Pitfalls & Principles, The Auction Block and The Second Factory are not part of the license. Don't use them for your own version, or to suggest it comes from us.
