Capstone: A Company AI Roadmap

What The Capstone Is

Over eleven weeks you studied the AI landscape, found use cases, checked data readiness, weighed build, buy and partner, covered security and compliance, designed a pilot, mapped department workflows, planned change, built a cost case, shaped an operating model, and prepared for scale and risk. The capstone joins these into one document that a leadership team can act on: a company AI roadmap.

A roadmap is a time ordered plan that shows what you will do, in which order, with whom, at what cost and how you will know it works. It is not a list of every possible idea. It is a choice of a few things done well. The best roadmaps are short, concrete and honest about uncertainty.

The Parts Of The Roadmap

Write the roadmap in seven parts, each drawn from earlier weeks. Part one is the starting point: where the organisation stands on skills, data and tools, and where AI is already used, including unofficial use. Part two is the priority use cases from weeks 2 and 7, each with a process map and a baseline. Part three is the approach for each, whether build, buy or partner, with reasons from week 4. Part four is the risk and compliance plan from weeks 5 and 11, with risk tiers and named responsibilities.

Part five is the people plan from week 8: literacy levels, role changes and the champions network. Part six is the cost and benefit case from week 9, with ranges and payback. Part seven is the operating model from week 10, with the council, the CoE and the vendor register. Two pages for each part is plenty. Where you lack information, say so and add a step to find it.

• Starting point and current use
• Priority use cases with baselines
• Build, buy or partner choices
• Risk and compliance plan
• People plan
• Cost and benefit case
• Operating model

The 90 Day Plan

A 90 day plan is long enough to deliver something real and short enough to keep focus. Split it into three phases of about thirty days. In days one to thirty, set up: confirm the sponsor and the council, finish the risk review, choose one or two use cases, measure baselines, and prepare data access and the pilot environment. In days thirty-one to sixty, run the pilot with a small group, with monitoring and human review active, train the pilot users, and collect feedback weekly. In days sixty-one to ninety, evaluate against the success measures, compare cost with benefit, decide whether to stop, fix, extend or scale, and prepare the next quarter.

For each milestone write the owner, the date, the evidence that shows it is done, and the budget. Add stop rules agreed in advance, for example: stop if safety incidents pass a set level, or if adoption stays below a set share after training. Stop rules protect the budget, the staff and the reputation of the programme. A pilot that ends with a clear decision is a success, even when the decision is to stop.

The Steering Committee Deck

A steering committee is the group of senior people who guide and fund the programme. They have little time, so the deck must lead to a decision. Use ten to twelve slides and put the request on slide two. A good order is: the decision requested, the business problem, the proposal in one slide, the evidence from the pilot or baselines, the cost and benefit with ranges, the risks and how you manage them, the 90 day plan, the people and operating model, what could go wrong with the plan, and the next steps with owners.

Write each slide title as a full sentence that states the point, for example: The pilot cut handling time by a measured amount and quality held steady. Use one chart or table per slide, and put detail in an appendix. Show negative results as plainly as positive ones, since a committee that finds a hidden problem later will stop trusting the rest. End with the exact decision and the date by which you need it.

Be ready for the questions committees usually ask. What happens if the vendor changes terms? Who is responsible if the system is wrong? How do we know staff will use it? What is the smallest amount we can spend to learn the most? Prepare a one line answer for each, taken from the earlier weeks, and place the longer answers in the appendix. Practise the presentation aloud once, with a timer, and cut anything that does not serve the decision.

The Rubric

Your roadmap is scored on six criteria, 100 points in total. Problem and value count 20 points: is there a real business problem, a baseline and a clear benefit? Feasibility counts 15: are data, skills, tools and timing realistic? Risk and compliance count 20: are risks identified, tiered and owned, with KVKK, GDPR and EU AI Act duties checked with counsel? People and change count 15: is there a literacy plan, a role view and a champions network? Cost and measurement count 15: are costs complete, benefits measured honestly and presented as ranges? Operating model and clarity count 15: are owners, decisions and policies clear, and is the document easy to read?

Score each criterion from zero to its maximum. A high score needs evidence, not adjectives. For example, a baseline with numbers and a method earns more than the word significant. Use the rubric yourself before anyone else does and fix the lowest score first.

The Review Checklist

Before you present, ask a colleague from outside the project to review the roadmap using a short checklist. A fresh reader finds gaps that the authors cannot see. Ask the reviewer to mark each item as yes, partly or no, and to write one comment on the weakest item.

Check these points. Does the roadmap start with the decision wanted? Is every use case tied to a process, an owner and a baseline? Is the source of each number stated? Are costs complete, including people and review time? Is there a named accountable owner for each use case and for the programme? Do risks have an owner, a control and a monitor? Is there a pause switch and an incident playbook? Do stop rules exist? Is the legal review noted without invented specifics? Is the language plain, free of jargon and hype? Could a new reader explain the plan in two minutes?

• Decision stated up front
• Each use case has an owner, process map and baseline
• Numbers have sources and ranges
• Risks have owners, controls and monitors
• Stop rules, pause switch and playbook exist
• Plain language throughout

After The Roadmap

A roadmap is a living document. Review it every quarter with the council, retire items that no longer fit, and add what you have learned. Keep a short log of decisions and changes so that new members understand how you arrived here. Publish a short update for all staff, since people support what they understand.

Keep your claims modest and your evidence strong. The technology and the rules will continue to change, so build the habits that last: measure against a baseline, keep a human responsible, test before you trust, and learn in public inside the company. These habits will serve you with whichever tools arrive next.

Key Terms

Roadmap

A time ordered plan showing what will be done, in which order, by whom, at what cost and with what measures.

90 day plan

A three phase plan covering setup, pilot and evaluation within about three months.

Milestone

A dated point in a plan with an owner and evidence that it is done.

Stop rule

A condition agreed in advance that ends or pauses a pilot.

Steering committee

A group of senior leaders that guides and funds a programme.

Rubric

A scoring guide that lists criteria and the points for each.

Review checklist

A list of questions an outside reviewer answers before a document is presented.

Appendix

Supporting detail placed after the main deck so the main story stays short.

Decision request

The exact approval or funding that the presenter asks the committee to give.

Exercise: Build And Review Your Roadmap

1. Collect your outputs from weeks 6 to 11 in one folder and mark what is missing.
2. Write the seven part roadmap in about two pages per part.
3. Draft the 90 day plan with three phases, owners, dates, budget and stop rules.
4. Build the steering committee deck of ten to twelve slides with the decision on slide two.
5. Score your own work with the rubric and improve the lowest scoring criterion.
6. Ask a colleague to review it with the checklist and revise before presenting.

Further Reading

• NIST AI Risk Management Framework (AI RMF 1.0), National Institute of Standards and Technology, 2023
• Good Strategy Bad Strategy, Richard Rumelt, Crown Business
• The Pyramid Principle, Barbara Minto, Pearson
• Regulation (EU) 2024/1689, the Artificial Intelligence Act, Official Journal of the European Union