The Baseline Is the Strategy

A lot of AI pilots try to prove value after the fact.

The team builds something. People try it. A few examples look good. Someone says it saved time. Then leadership asks the obvious question: what did it change?

That is when the conversation gets messy.

Nobody knows exactly how long the workflow took before. Nobody knows how much rework was normal. Nobody knows which steps were the real bottleneck. Nobody knows whether the team saved time, shifted work, reduced risk, or just created a cleaner first draft.

The pilot may still be useful. But the value case is now harder than it needed to be.

The baseline should not come after the pilot. The baseline is how you decide what to build.

If you do not know what the workflow looks like today, you cannot prove what changed tomorrow.

A baseline is not a finance exercise

When people hear “baseline,” they often think spreadsheet. That is part of it, but it is not the whole point.

A good baseline is a map of how the work actually happens. Not how the process is supposed to happen. Not how it was documented three years ago. The real version.

Who starts the work? What triggers it? What systems do people open? What documents do they trust? Where do they copy and paste? Who gets asked for context? Where does work wait? What gets reviewed? What comes back for rework? What breaks when the volume goes up?

That is the baseline. It turns vague complaints into something the business can act on.

The real workflow is usually messier than the official one

The official process says proposals come from a template. The real workflow says the account executive pulls discovery notes, searches old decks, asks delivery for similar work, checks pricing, finds the right case study, writes the first draft, sends it to three people, waits, revises, and then tries to get it out before the deal cools.

The official process says support triages tickets. The real workflow says a junior person reads the ticket, searches the knowledge base, checks Slack, asks the same senior person, waits for product context, drafts a response, and sends it only when someone is confident enough.

The official process says finance produces a weekly report. The real workflow says numbers move through exports, spreadsheet tabs, emailed comments, manual checks, and last-minute edits from people who know where the bodies are buried.

That difference matters. AI cannot improve the process you wish you had. It can only improve the workflow you can actually describe.

What to baseline

You do not need a perfect model before you start. You need enough truth to make a decision.

For most workflows, that means capturing a few basics:

• volume: how often the work happens
• cycle time: how long it takes from start to finish
• effort: how much human time goes into each unit
• handoffs: how many people or teams touch it
• rework: how often it comes back
• source of truth: where the trusted context lives
• review path: who approves, edits, or rejects the output
• business lever: revenue, cost, speed, quality, risk, or capacity

This is where the value case starts to become real. Not because the numbers are perfect, but because the team can finally see the shape of the work.

The baseline tells you whether AI is ready

Sometimes the baseline makes the AI case stronger. You find a repeated workflow with clear volume, visible pain, scattered context, a known review path, and a metric worth moving. That is a good candidate.

Sometimes the baseline tells you to slow down. The source of truth is unclear. The rules are undocumented. The examples are inconsistent. The approval path depends on who is asking. Nobody owns the workflow. The team cannot agree on what good output looks like.

That does not mean AI will never work there. It means the workflow is not ready for production AI yet. And that is useful to know before you spend money building the wrong thing.

What the baseline shows	What it usually means
High volume, repeated work	Good AI candidate
Lots of copy-paste between systems	Context or integration problem
Same expert asked every week	Hidden knowledge workflow
Heavy review and rework	Quality bar needs to be explicit
No clear source of truth	Fix context before building
No owner	Hard to change the workflow
No measurable lever	Weak investment case

A baseline is not just measurement. It is a filter.

The baseline also changes the design

Once you understand the current workflow, the AI system becomes easier to design.

If the bottleneck is context gathering, the system needs better source-of-truth access. If the bottleneck is inconsistent output, the system needs examples, rules, and review criteria. If the bottleneck is approval, the system needs a decision packet, not just a draft. If the bottleneck is risk, the system needs citations, confidence signals, and a human review path. If the bottleneck is capacity, the system needs to show where time comes back and what work replaces it.

This is why baseline work is strategic. It does not just support the ROI model. It tells you what the production system has to do.

The baseline is not paperwork. It is product strategy for the workflow.

Measure before, then measure after

The simplest AI measurement plan is still the best one.

Before AI
How does the workflow work today?
What does it cost?
Where does it wait?
Where does it break?

After AI
What changed?
What got faster?
What got better?
What risk dropped?
Where did capacity go?

That is enough to keep the work honest. You do not need to turn every AI project into a six-month measurement study. But you do need a before picture. Otherwise, the after picture is mostly vibes.

A good first baseline can be rough:

• interview the people closest to the work
• sample ten recent examples
• time a few real tasks
• look at system timestamps
• count review loops
• count handoffs
• log one week of actual effort

That will tell you more than another AI brainstorm.

The shift

The baseline is not the thing you do once the pilot is over. It is the thing you do before deciding what deserves to be built.

It helps you see which workflow matters, which metric can move, what context AI needs, what might break in production, and how you will know whether anything improved.

Without a baseline, every AI result is easier to argue about. With a baseline, the conversation gets simpler:

This is how the work used to happen. This is what changed. This is the metric that moved.

That is the strategy.

It starts one step earlier, with the reminder that AI usage is not AI value.

---

More in this series, Proving AI Value:

• AI Usage Is Not AI Value
• What Metric Can This Workflow Move?
• The Baseline Is the Strategy (you are here)