Artifacts Playbook

Buying an AI Tool Playbook

A four-gate decision protocol for adding an AI tool to an operating business. Built for one decision meeting, durable across tool cycles.

PDF 19 pages v1.0 Updated Apr 2026

Download PDF buying-an-ai-tool-playbook.pdf · 165 KB Download →

A decision protocol for adding an AI tool to an operating business.

Built for one decision meeting. Durable across tool cycles.

Start here

Who this is for. Chief executives, chief operating officers, chiefs of staff, heads of function, and operators accountable for a workflow an AI tool is proposing to change. Whoever will be answerable for this workflow one year from now should be in the room.

When to use it. When a tool is on the table and a purchase decision is active. Use it before the negotiation call, before a pilot, and again at renewal.

What decision this helps make. Whether this tool earns a place in the operating system of the business.

How to run a 45 to 60 minute decision review. Block a meeting with the workflow owner, the executive accountable for the function, and two people who can speak for cost and for data access. Walk the four gates in sequence and mark each one Clear, Concern, or Blocker. Leave the meeting with a written decision, an owner, and the next review point.

Decision outcomes

Four defensible outcomes. Each is operational.

Buy now. The workflow is named, owned, and understood. The data is reachable. Review and approval logic is clear. The net gain, after setup, review, training, and coordination, is material. The team can name the work that stops elsewhere. If the vendor were gone in twelve months, the business would feel inconvenience rather than damage.

Run a bounded pilot. The workflow is believable. One or two assumptions about operating fit or real gain remain unproven. A narrow, dated test is the cheapest way to retire those unknowns.

Wait. The workflow is not ready, the data is not reachable without remediation, or the accountable owner is missing. The right next step is operating work, not procurement.

Reject. The purchase is being driven by pressure, narrative, or a polished demo, and the workflow will not survive contact with the tool. Rejection is a legitimate outcome, and often the cheapest one.

Decision standards

Each gate takes one status at the end of the conversation, not field by field.

Clear. No open questions that would change the decision. What remains is execution detail.

Concern. Open questions that can be retired within a pilot, a short diagnostic, or a focused piece of operating work. A Concern shapes the decision, it does not stop it.

Blocker. An open question that, left unanswered, would make any decision other than Wait or Reject indefensible.

The four gates

The body of the playbook walks each gate with prompts, evidence boxes, and the scoring conversation:

  1. Workflow. Is the workflow real, named, owned, and understood as it runs today?
  2. Data and access. Is the data the tool needs reachable, in the right shape, and governable?
  3. Operating fit. Will the tool actually live inside the team’s review, approval, and handoff flow?
  4. Cost and exit. Does the net economic case hold up across setup, review, training, and coordination, and can the business survive the vendor disappearing?

Download the playbook to run the full review.

Subscribe

A roughly monthly dispatch from karlbaz.com. Links to the month's memos, plus a short note on what matters now.

Roughly monthly · No promo · I do not sell or share your address · Unsubscribe anytime