Chapter Three

From Pilot to Operating Leverage

Five steps. Each one removes a specific cost or risk. Skip one and the investment stays where it is: a demo that never reaches the work that costs money.

A discipline, not a product.

That sequence has a name only because the industry needs a standard for it. Five steps, each closing a failure mode that costs real money, in an order that cannot be rearranged because each step depends on the one before it.

It works with any model, any technology stack, any team, and it applies whether you build the AI yourself or someone builds it with you. We are sharing it because a standard kept private is not a standard. If your firm proves its output and the firm next door does not, you are not in the same competitive position, whatever model you both run.

Any model, any team

The discipline is the standard. The technology underneath is a choice.

Cumulative

Each step depends on the one before it. Skip one and the return from every step after it drops.

Figure 3.1 / The five steps

How an AI investment becomes operating leverage.

IN · AI INVESTMENTCost centre. Unproven.

Decide

Agree what the AI should do, in plain language, signed by whoever owns the process.

SKIP IT: REWORK COSTS MORE THAN THE BUILD

Ground

Connect it to your firm’s live data, documents and policies, with a named source of truth.

SKIP IT: ERRORS REACH CLIENTS AS INVOICES

Calibrate

Run it against your hardest real scenarios and compare it to what your team actually did.

SKIP IT: THE FIRST FAILURE IS IN FRONT OF A CLIENT

Prove

Check every output against your rules. A separate system. Pass or fail, with the rule named.

SKIP IT: THE INVESTMENT CARRIES LIABILITY, NOT RETURN

Operate

Your team approves what executes. The record assembles itself as a byproduct.

SKIP IT: NO RETURN ON THE TIME SAVED

OUT · OPERATING LEVERAGEProven. Creating Value.

Figure 3.2 / Value created, risk removed

What each step is worth, and what it takes off the table.

	Step	Value to the firm	Risk eliminated
01	Decide	A written agreement that survives a personnel change. No rework.	No one can say afterwards what the process was meant to do.
02	Ground	Answers drawn from your own records, at the moment they are asked for.	Confident output invented from material that was never yours.
03	Calibrate	A measured rate of agreement with the people who do the work today.	The first hard case arrives in front of a client instead of in a test.
04	Prove	A formal check that your rules held on this specific output.	Liability for output no one can account for. The cost of being wrong.
05	Operate	Proof produced while the work happens. Capacity grows without hiring.	Senior people rebuilding six months of history out of email.

Why the order holds

The steps are cumulative. Proving an output against rules nobody agreed proves nothing. Skip one and every step after it returns less.

Every step is set out the same way.

Each of the next five chapters opens with what it costs to skip the step, then shows what the step is and what good looks like.

What the step isWhat it actually is, in one paragraph, with no technology in it.

What it costs to skipThe dollar cost, the time cost, or the risk you carry without this step.

What good looks likeThe document or behaviour you should be able to point at when the step is done.

At your firmA situation you have lived through. The cost is named.

Skip this stepWhat breaks, and where it shows up.