A useful AI roadmap for a company of a few dozen to a few hundred people is two pages, not a slide deck. It lists who already uses AI and for what, one rule for company data, the things the company will never let a machine do alone, the first piece of work with a named owner, and the dates the board will hear about it. This article is for the owner, CEO or operating partner who has been told to produce an AI roadmap and wants to know what it should contain.
I was COO and then President of a public company that closed more than 40 acquisitions, and I ran the integration on most of them. Now I write fixed-fee AI plans for CEOs of private-equity-backed and owner-led mid-market companies through 013 Labs.
Why most AI roadmaps fail
Most fail for one of three reasons, and all three show on the first page.
It is a vendor's product plan with your logo on it. A software company's roadmap is a sequence of its own products. That answers the vendor's question, how to sell more, and not yours: what should the company do first, and who is accountable?
It is a maturity score. The company is rated from one level to the next, with a list of what it takes to climb. The result is a tidy chart and no decisions. A company with many tools and no owner can outscore one with a single tool and a clear rule, and the second company is the safer one.
It has no owner. Every line says "the team" or "IT" or "leadership." Where no name is written, nothing happens, and the board finds out two quarters later.
The two pages, section by section
The first page says where the company is and what the limits are. The second says what happens next and when the board hears.
Who already uses AI tools, and for what
What to write: A plain list of where staff already use AI tools, what they put into them, and who the user is. Say that nobody is in trouble. Include quiet uses: a customer quote pasted into a public chat tool, a contract summarized before a call.
Hypothesis, a composite example. Picture a distributor with a few hundred people. The sales manager uses a public chat tool to tidy customer emails, and a finance analyst pastes invoice details into another. The page records both, with names, and notes that customer pricing is leaving the building.
The data rule
What to write: One rule, in a sentence a new hire could repeat, and the name of the person who approves exceptions. A long policy is read once and ignored. A single rule gets followed.
Hypothesis, a composite example. Picture an owner-led services firm. Its page reads: "Nothing that identifies a customer or an employee goes into any tool the controller has not approved." The controller's name sits beside it.
The never-build list
What to write: Three to five lines naming what the company will not let a machine do alone. The first line is almost always the same: nothing that sends money, signs a contract or makes a promise to a customer without a named person's approval.
Hypothesis, a composite example. Picture a company with a busy collections team. Its list says no system changes a credit hold, writes off a balance or contacts a customer about money without a named person approving first.
The first piece of work and its owner
What to write: One piece of work, not a program. Name the owner, the problem it solves and the day you will look at the result. Choose something routine, where preparation takes the time and a person still decides. The best owner is often someone who already uses AI well, which is the subject of a free guide on spotting your AI champion.
Hypothesis, a composite example. Picture a company where an operations coordinator spends half of each Monday assembling the weekly numbers. The page makes her owner of one piece of work: a draft of those numbers prepared automatically, the source beside each figure, reviewed by her before it reaches the CFO.
The measure that says it worked
What to write: One measure the company already tracks, such as hours spent, errors, days to collect or time to answer a customer. Write the starting point down before work begins, and the date you will compare. Without a recorded starting point, every result is an opinion.
Hypothesis, a composite example. Picture the coordinator's Monday task. Before work starts, someone records how long the weekly numbers take and how often they need correcting. Several weeks later the same two figures are written beside them.
The board calendar
What to write: The dates the board will hear about the roadmap, who presents, and what they report: the first piece of work, its measure, any addition to the never-build list, and whether the data rule was broken.
Hypothesis, a composite example. Picture a company with quarterly board meetings. The page lists the next two, names the CEO as presenter at the first and the work's owner at the second, and says what each will show.
What is deliberately missing
Three things people expect on a roadmap are left off on purpose.
No platform choice. Choosing a platform before you know the first piece of work is how companies end up with software nobody owns. Tools are chosen after the work is defined.
No budget line until the first piece has run. A budget written before anything has been tried is a guess. Once the work has run and been measured, the next budget can be an informed one.
No headcount. The first piece of work is done by people already on staff, a few hours a week.
How the roadmap changes after a deal
An acquisition changes every line on both pages. The other company brings its own tools, habits and unwritten uses of AI, and its data rule may be different or absent. Run the same two pages on the new company early and merge them, rather than bolting one roadmap onto the other. The timing is covered in the first 100 days after the deal.
When the board asks
A board, lender or buyer rarely wants the whole roadmap. It wants three answers: what is already happening, who is accountable, and what will never be allowed. The first page gives them. The second shows that someone will report back. If you can hand over two pages with names on them, you have answered the question. If you can only hand over a deck, you have probably answered a different one. Investors who score this in a portfolio use the same pages as a readiness check, covered in the piece on an AI readiness assessment for a portfolio company. The agenda and the ten questions are on one page: AI board briefing template.
A template in words
Here is the outline to fill in, a few lines per blank.
Page one, where we are:
- Today, the people using AI tools are ___, for ___, and what they put in is ___.
- Our one data rule is ___, and ___ approves any exception.
- We will never let a machine, working alone, ___ (three to five lines, starting with money, signatures and promises to customers).
Page two, what happens next:
- The first piece of work is ___, owned by ___, to solve ___.
- We will know it worked if ___ moves from ___ to ___ by ___.
- The board hears about it on ___ and ___, presented by ___, and it covers ___.
If a blank cannot be filled with a name or a date, that is the real finding. Fix that before adding anything else.
Mistakes that repeat
- Starting with a tool. The roadmap becomes a purchase list.
- Giving the work to IT alone. IT keeps systems running and secure. Who may use AI, and for what, is a business decision.
- Naming a committee as owner. A committee cannot be asked a question on Monday morning.
- Leaving out the never-build list. The limits are what the board remembers.
Frequently asked questions
What should an AI roadmap include?
Who already uses AI and for what, one data rule, a short list of things the company will never let a machine do alone, the first piece of work with a named owner and a measure, and the board dates. Leave out platform choice, budget and headcount until the first piece has run.
How long should an AI roadmap be?
For a company of a few dozen to a few hundred people, two pages. The first describes where you are and the limits, and the second describes what happens next and when the board hears.
Do we need an AI roadmap template?
Only one that forces names and dates, like the fill-in outline above, where every blank needs a person, a rule or a day. Avoid templates that rate your maturity or list vendor options, because they produce a chart and no decisions.
Who should own the AI roadmap?
One named person inside the company with real time and the authority to say no, usually the CEO or a senior leader reporting to the CEO. Not a committee, and not an outside adviser. Each piece of work on it also has its own owner.
How often should it be updated?
Before each board meeting, and whenever a deal closes or a new use of AI turns up. Updating means adding a name, a date or a limit, not rewriting it.
Not sure where you stand? Take the 3-minute AI readiness scorecard.