The Delivery Model Ladder

Your team can use AI every day and still run the same business, so buying the tools does not tell you how far you have come. The Delivery Model Ladder places your firm by what changed in how you deliver the work. It has four stages: Stage 0 Assisted, Stage 1 Enhanced, Stage 2 Augmented, and Stage 3 AI-native.

A team can use AI every day without changing how the firm runs. The Ladder looks for changes in cost to acquire, cost to deliver, retention, or price. Margin and revenue per person show whether those changes reached the business.

In a franchise network's delivery data, the largest group of firms I've reviewed sits at Stage 0 or Stage 1 as of August 2026. Moving from Stage 1 to Stage 2 is the hard part because the firm has to redesign the work, not add another tool.

The Ladder is for owners and operational leaders of established US small and medium-sized businesses with teams and repeatable processes. Apply it to repeatable customer work where its delivery measures fit. Use what happened in your numbers, not what is planned. The AI Readiness Assessment can place your firm in sixteen questions.

What stage of AI adoption is my firm actually in?

Start with four numbers: cost to acquire, cost to deliver, retention, and price. I call them The Four Numbers. Do not start with subscription counts or a survey of who uses AI.

Margin and revenue per person are the results. They show whether a change in the four numbers made a difference to the firm. Ask which number moved, by how much, and what caused it.

If none moved, you are at Stage 0. If both cost numbers fell but retention, price, and margin stayed flat, you are at Stage 1. Faster work inside the same delivery model is still Stage 1.

Why doesn't everyone using AI show up on the financials?

Stage 0 Assisted is individual use. People use AI on their own, in their own way, for their own tasks.

The firm still sells, staffs, and delivers work as it did before. Each person keeps their own tools, prompts, and knowledge. Someone doing the same job elsewhere in the firm may start from scratch.

That can still be useful. People learn what AI handles well and where it fails. But if none of the Four Numbers moved, the firm itself has not changed yet.

Why do most service firms stall at Stage 1?

Stage 1 Enhanced puts AI inside an existing workflow. A proposal that took four hours may take minutes. A first draft that took a day may take an hour. Cost to acquire or cost to deliver falls.

The problem is what happens next. In the firms I've observed through a franchise network's delivery data, the saved time often disappears before it reaches margin. The same people deliver the same scope for the same price, so the firm has free capacity but no plan for it.

That is the Stage 1 pattern, a cost falls but retention, price, and margin stay flat. You can point to a faster task, but you still need to decide how that helps the business.

This is also a common pattern in The Production Gap. The gain exists in a sentence like “we save fifteen hours a week,” but nobody decided what those hours should produce.

What actually changes at Stage 2?

Stage 2 Augmented begins when the firm redesigns delivery around AI. Someone decides what will no longer be done by hand, what must be checked, and who owns each part of the work.

Now the time freed at Stage 1 has a job, improving proposals, onboarding, reports, testing, or turnaround time. The change should reach the customer, not just the person doing the work.

There are two ways to see that change. Better delivery can improve retention. Or the firm can change what it sells, such as moving from hours to a finished outcome or from a project to a subscription. That changes price.

And either route reaches Stage 2 only when margin follows, because the firm needs to keep some of the value it created.

Here is where I would start with that change.

Give one person ownership. That person tracks which experiments worked, decides what becomes standard, and shares the new process with everyone doing the same job. In a $1–20M firm, that can be one clear owner rather than a task force.

Write a short boundary. Say which tools people may use, where they may use them, and which data must stay inside the firm. A rule people can find works better than silence or a long policy nobody reads.

Put learning time on the calendar. People cannot redesign delivery while every hour is already promised to client work. If the firm does not fund the learning time, the old workflow wins.

This is the Unfunded Hour in The Production Gap. Start with one workflow. Give it an owner, finish the change, and then move to the next one.

What does Stage 3 look like?

Stage 3 AI-native is still early. The firm has rebuilt delivery far enough that growth no longer has to follow hiring.

The firm keeps its knowledge where people and agents can use it. Decisions, client context, processes, and past work are written down. Agents do repeatable production work, while people direct it, check it, and handle the judgment calls.

The offer changes too. The firm sells outcomes or capacity instead of hours. It measures output per person, quality, and cost to deliver rather than celebrating time saved.

The firms exploring this stage are not typical. Many were built in the AI era, employ people who know how to build these systems, and work in areas where output is easy to check.

That matters because the wider picture is very different. The Production Gap cites an economy-wide baseline in which 89% of executives reported no labor-productivity impact and 66% of firms using AI were only adding it to existing tasks. Stage 3 firms show what may be possible, not what the average firm will look like next year.

I've seen only a couple of companies begin to explore this stage, and both were young firms born in the AI era. For an established service firm, getting there will take more than copying their tools.

What does it take to get to Stage 3?

Five requirements stand out when I compare early Stage 3 firms with the service firms in my data. None starts with a purchase.

1. Make a delivery decision. Decide how agents may reach the systems where work lives, what they may do, and how their work will be checked. Then rebuild the workflow so agents execute while people direct and review.

Buying an “agent tool” without those decisions usually creates another pilot. The tool is present, but the delivery model stays the same.

2. Split work into parts someone can check. Agents work best on tasks that repeat and have a clear result. Research memos, first drafts, quality checks, proposal assembly, and status reports are common examples.

Before giving an agent the work, name who checks it and what passing means. That is where much of professional services gets harder, because when the work depends on judgment, you cannot always write down what a correct result looks like.

3. Let senior judgment reach more work. In many small service firms, senior people become the bottleneck. Their judgment reaches only the work they can personally touch.

At Stage 3, one senior person can direct several agent loops. Agents handle the repeatable work. The senior person makes the calls that need context, trust, and taste.

This works only when the firm's knowledge is not trapped in individual people. The system needs enough context to route work without adding another layer of managers.

4. Share context before you automate delivery. Start by making decisions and documents easy for agents to read. Let people ask questions against the firm's own knowledge before an agent touches client work.

A useful test is simple: could someone rebuild last month's main decisions from what the firm wrote down? If not, the first step is better records, not more autonomy.

You cannot automate a process that lives only in someone's head. There is nothing for an agent to follow and nothing for a reviewer to check.

Documenting that process no longer needs a large project. Someone can record the screen while doing a weekly task, then use AI to turn the recording into a first draft of the process. The experienced person still checks it, but they no longer start with a blank page.

So begin inside the firm with an ordinary task, where you can check whether the answer is useful before asking a client to depend on a new AI service.

5. Measure output and quality together. More output is not enough. Pair it with a quality measure, such as rework rate, turnaround time, or how often work needs to be escalated.

The Stage 3 pattern is revenue per person rising while retention and margin hold or improve over several quarters. If volume rises while quality falls, the firm has not reached Stage 3.

The order matters. First, write down decisions and define work that can be checked. Then make the delivery decision. Wider reach and better numbers come after the new process works.

A Stage 1 firm does not need to aim straight at Stage 3. It needs to make the next Stage 2 decision and keep building from there.

How do I use the Ladder without buying anything?

Compare the four stages with this quarter's numbers, not next quarter's plans. Ask, “Which of the four numbers moved, by how much, and why?”

Stage	How work runs	What changes	How to tell
Stage 0 Assisted	People use AI on their own	No clear change in the Four Numbers, margin, or revenue per person	People talk about using AI, but it does not appear in a monthly report
Stage 1 Enhanced	AI speeds up parts of the existing workflow	Cost to acquire or cost to deliver falls. Retention, price, and margin stay flat	You can name hours saved, but not a business result
Stage 2 Augmented	The firm rebuilds workflows around AI	Retention improves or the firm changes what it sells, and margin follows	Customers notice better delivery, or the firm sells an outcome where it once sold hours
Stage 3 AI-native	Agents use shared company knowledge inside a rebuilt delivery model	Revenue per person no longer follows headcount. Speed and quality no longer follow hours worked	The firm can grow without adding people at the same rate

Is the Ladder an AI maturity model?

Yes, but it measures what changed in the business. Most maturity models score tools, training, policies, and pilots. A firm can have all four while its delivery model stays the same.

The Ladder counts a stage only when the matching pattern appears in the work and the numbers. Buying another tool cannot move the firm up.

That makes the next step easier to see. The firm needs a decision about delivery, pricing, staffing, or scope, not another item on a technology checklist.

How do the Four Numbers and the Production Gap fit?

The Ladder tells you where you are, and The Four Numbers tell you whether the business changed. If you are not moving, The Production Gap helps you look for what is getting in the way.

At Stage 1, the usual problem is not that AI failed. The work got faster, but nobody decided what the free capacity should do next.

---

I built the Ladder after watching hundreds of service firms adopt AI through a franchise network's delivery data.

I am not judging adoption from a list of tools. I am looking at how firms deliver work and the numbers they report each month. I write about those patterns in the newsletter.