Baseline Before You Touch the Tool

Why this matters

Most people start with the AI and then try to decide if it helped. That is backwards. First establish the current reality: how long does the task actually take? What does "good enough" look like? Where do errors or rework usually appear? What is the real cost of being wrong? Without a baseline, every AI run is just a feeling.

What you will be able to decide after this

• Whether a task changed for the better — with a number, not a vibe.
• What "good enough" means before you start, instead of discovering it after.
• Whether the AI actually beat the current way of doing the work.

Core lesson

The baseline is four measurements on the current way of doing a task, taken before any AI is involved:

• Time: how long does the task actually take, end to end — not the ideal time, the real time.
• Quality standard: what does "good enough" look like in terms you could defend to someone who knows the work?
• Failure points: where do errors or rework usually appear?
• Cost of being wrong: what actually happens when the output is wrong?

The baseline does not need to be fancy. It needs to be real. A stopwatch and three honest answers beat a dashboard and a hope.

Without a baseline, every AI run is just a feeling.

Worked example

A program manager wants AI to draft status updates. Before touching the tool, they time their current process: 22 minutes per update, including gathering notes from three sources and reconciling two conflicting numbers. The quality standard: "a stakeholder can act on it without asking follow-up questions." The failure points: the conflicting numbers, and the occasional invented date. The cost of being wrong: a stakeholder plans around a date that does not exist.

Now the AI run can be compared against something. Without the baseline, the AI's 4-minute draft looks like a miracle — until the invented date ships.

Practice

Pick one recurring task you care about. Before reading the Apply step, estimate: how long does it take now, what does good enough look like, where do errors appear, what is the cost of being wrong? Write your guesses. Then measure for real in the Apply step.

Apply — produce the artifact

Choose one recurring task you actually care about. Document the current process: time, quality standard, common failure points, and what "done" looks like. Do this without AI.

Verify

Show the baseline to someone who knows the work. If they say your time or quality standard is unrealistic, revise it before you proceed.

Sources

This module is original practice guidance based on the authoring standard and does not depend on a specific external factual claim. Editorial review is still required.