The Four Numbers

You can save hours with AI and still finish the quarter with the same costs and revenue, so the hours alone cannot tell you whether it worked for the firm. The Four Numbers tracks cost to acquire, cost to deliver, retention, and price.

These numbers cover how much it costs to win a customer, how much it costs to serve them, how long they stay, and what they pay. Margin and revenue per person then show whether the change reached the firm.

There is one check between the four numbers and those results: where did the free capacity go? A firm can save hours every week and still keep the same costs, offer, and revenue. If nobody gives those hours a new job, the business does not change.

The framework is for owners and operational leaders of established US small and medium-sized businesses with teams and repeatable processes. Its measures fit hourly, project, retainer, and subscription work. Apply it where your business uses those models; it does not assess every kind of operation. It also connects each result to a stage of The Delivery Model Ladder.

What should a service firm measure when it adopts AI?

Do not use hours saved as the main measure. Hours are easy to estimate, but the estimate may never appear in the firm's financials.

Track these four numbers instead:

• Cost to acquire: everything the firm spends to win one customer. That includes marketing, sales, tools, and the time senior people spend on pitches.
• Cost to deliver: everything the firm spends to produce one unit of service. Include labor, tools, and rework.
• Retention: how many units a customer buys before leaving. This may mean renewal months, repeat projects, or repeat engagements.
• Price: what the firm charges for one unit of service. The unit may be an hour, a project, a month, or an outcome.

The table shows how they fit together. Two numbers describe the customer. Two describe the work.

	What it costs you	What it earns you
Per customer	Cost to acquire	Retention
Per unit of work	Cost to deliver	Price

The unit changes with the billing model, but the four numbers do not.

Billing model	One unit of work	How to read retention
Subscription	One service month	Months retained
Retainer	One retainer month	Renewals
Project	One project	Repeat projects and referrals
Hourly	One hour	Repeat engagements

The Delivery Model Ladder shows how far AI has changed the delivery model. The Four Numbers show whether the change actually happened.

Why are there four numbers and not five?

The four describe the basic math of a service business:

( price − cost to deliver ) × retention − cost to acquire = contribution per customer

Price and retention determine what a customer brings in. Cost to acquire and cost to deliver determine what the firm spends. Add the result across customers and you get the margin the business keeps.

A fifth number would either sit inside one of these four or describe a result they produce. The billing unit can change without changing the structure.

Why isn't margin one of the four numbers?

Margin is the result those four numbers produce, so I use it to check what happened rather than count it as another thing to change.

Cost to deliver is already part of margin. Putting both in the same list would count some of the same money twice.

Use margin as the check. If the four numbers improved but margin stayed flat, the gain went somewhere else.

Look at the margin left after both delivery and acquisition costs. Gross margin includes delivery cost but can hide a rise in the cost of winning customers.

Why isn't revenue per person one of the four numbers?

Revenue per person is another result, it changes after you change how the work gets done.

I kept it among the four for a long time because it matters. But a firm cannot change the ratio directly. It can change price, delivery, staffing, or the number of units sold. Revenue per person follows later.

So the number still matters, it helps confirm Stage 3 of the Ladder, when you can grow without hiring at the same rate.

Why is price one of the four?

Price shows who keeps the gain when delivery gets cheaper.

This is easy to miss in an hourly firm. If AI cuts the time needed for the work, the firm may sell fewer hours and charge less. The client keeps the saving even though the firm created it.

Changing the unit can break that link. The firm might sell a finished outcome, a subscription, or a block of capacity instead of the hours used to produce it. Margin then shows whether the new offer worked.

Price is also the only one of the four that needs a direct business decision. AI cannot change what the firm sells on its own.

How do you know if AI savings are disappearing?

Ask one question before the quarter closes:

Of the hours AI freed this quarter, what share now has a named use?

I call this the gate. It does not measure the hours again. It checks whether the firm decided what those hours should do next.

Where the capacity went	What should change
More delivery with the same team	Cost to deliver falls
Better delivery, onboarding, speed, or communication	Retention rises
A repackaged offer	Price changes
Nothing named	The saving disappears

Now, the four numbers can stay flat because the rollout is small, it is still early, or sales had a bad quarter. Asking where the time went gives you something to check before those results arrive.

I've watched hundreds of service firms adopt AI through a franchise network's delivery data. A common pattern is that the tool works, but nobody gives the saved time a new job. A senior person finishes a draft in 40 minutes instead of three hours. If nothing else changes, the firm still has the same payroll, offer, and revenue.

The Production Gap covers this failure in more detail. The gain is real, but it never reaches the business.

Which number moves at each stage of the Ladder?

Each stage of the Ladder has a different pattern in the four numbers and the two results.

Ladder stage	Four Numbers	Use of free capacity	Result
Stage 0: Assisted	No clear movement	Nothing to assign yet	Margin and revenue per person stay flat
Stage 1: Enhanced	One or both cost numbers fall	Nothing named	Margin stays flat
Stage 2: Augmented	Retention improves or price changes	Capacity has a clear use	Margin improves
Stage 3: AI-native	The four hold while the firm grows	Reassignment is part of normal delivery	Revenue per person no longer follows headcount

Stage 1 is common. An April 2026 Census Bureau working paper found that 66% of firms using AI only added it to existing tasks, while 64% made no organizational changes.

That is what Stage 1 looks like. Tasks get faster, but the firm still sells, staffs, and delivers the same way.

Stage 2 begins when the free capacity improves retention or supports a new offer and price. Stage 3 appears later, when the four numbers hold while revenue grows without matching headcount.

How does AI move cost to acquire?

AI can reduce the work needed to win a customer. It can help with prospect research, proposal drafts, case write-ups, content, qualification, and follow-up.

Measure the total cost per customer won. Cost per lead, content volume, and outreach volume can improve while the total cost of winning a customer gets worse.

More cheap content or outreach is not always better. If response rates fall, the firm may spend more to win the same work. Poor-fit customers can also cost more to serve and leave sooner.

So read the cost of winning a customer beside retention and cost to deliver, because one can improve while the other two get worse.

How does AI move cost to deliver?

AI can reduce the labor inside each unit of service. Common examples include research, first drafts, formatting, status reports, routine analysis, and internal handoffs.

Track the actual cost of one engagement, project, or month of service. Do not rely on a survey or a vendor estimate of hours saved.

Pick the unit, add its labor, tools, and rework, and compare the same unit each quarter. If you cannot say what the work cost before and after, you cannot tell whether delivery improved. I learned this the hard way: a 23× cut in model cost that I could not call a win.

How does AI move retention?

Retention moves when customers notice better delivery. Work may be more complete, turnaround may be faster, onboarding may be smoother, or communication may improve.

Retention can also fall. AI output that looks finished but is wrong creates rework and weakens trust. Usage may rise while customers become less satisfied.

Sales also affects retention. A quarter with more poor-fit customers can lower retention even when delivery stayed the same. Read the number by service line or customer group where possible.

And retention takes time to show up, usually several quarters of renewals or repeat purchases before you can see a clear pattern.

How does AI move price?

AI does not move price by itself. The firm must decide to change what it sells or how it charges.

A higher hourly rate is still an hourly rate. A larger change is to sell the outcome instead of the hours, capacity instead of a fixed project, or a subscription instead of separate jobs.

This lets price follow the value of the work rather than the labor used to produce it. But the firm still has to deliver the promised quality. A new price without the matching delivery change can hurt retention later.

What confirms the delivery model changed?

Revenue per person is total revenue divided by total headcount. It shows whether growth still requires hiring at the same rate.

The number begins to separate from headcount when agents carry a real share of production and the firm's knowledge is easy for people and agents to use. Work no longer stops with the one person who holds the context.

Margin and revenue per person answer different questions. Margin shows whether the firm kept the gain. Revenue per person shows whether the firm changed how work gets made.

I've seen only a couple of young firms where revenue per person is starting to separate from headcount. Both were born in the AI era. An established firm should treat the number as a long-term measure, not proof that a single rollout worked.

What do firms measure instead?

Common measure	What it tells you	What to watch instead
Leads or content produced	Volume	Cost to acquire per customer won
Estimated hours saved	An estimate of an input	Cost to deliver one unit of service
AI logins or adoption rate	Usage	Retention and the quality customers experience
Discounting	A price cut	Price per unit of work
Free hours	Capacity without a plan	What share has a named use
Revenue and headcount as separate totals	Growth without its staffing cost	Revenue per person

Activity measures can still help run a project. They just cannot prove that the business changed.

How do you measure the ROI of AI in a service firm?

Measure the business before and after the change. Compare cost to acquire, cost to deliver, retention, and price. Check where the free capacity went, then use margin and revenue per person to see whether the gain lasted.

This is harder than measuring one purchase because AI spreads across many tasks. The place where the value appears may not be the place where the firm spent the money.

A 2026 NBER survey of nearly 6,000 executives found that 89% reported no labor-productivity impact from AI over the previous three years. The same group reported a realized gain of 0.29% and expected 1.4% over the next three years.

You do not need a separate AI dashboard for this. Start with what the firm already tracks, what it spends to win customers, what delivery costs, what it charges, and whether customers return.

These numbers can move for reasons unrelated to AI. A strong sales quarter, a lost customer, a price change, or a new hire can all affect them. Read them by service line or customer group where possible, and record what else changed during the period.

Which number should a firm move first?

Start with cost to acquire or cost to deliver. AI can affect either within a quarter, which makes them practical first measures.

Before the cost falls, decide what the free capacity will do. Will it improve delivery and retention, or support a new offer and price?

In a franchise network's delivery data covering 150+ units and growing, firms that improved margin gave the free capacity a job within two or three quarters. They used it for better delivery and onboarding, or they changed the offer.

So start with one cost number and decide where any freed time will go, then watch whether retention or price changes next.

---

The Four Numbers is the measurement layer in Proofwork. Use The Delivery Model Ladder to place the firm, The AI Readiness Assessment to find the next move, and The Production Gap to see where the gain can disappear. I write about these patterns in the newsletter.