AI Roadmap Template: Horizons, Swimlanes and an 18-Month Example

An AI roadmap is a management document that shows how an organisation will introduce and scale AI over the next one to two years. Its columns are quarters. Its rows are the kinds of work that must move together: the use cases that deliver value, and the data, platform, governance, people and measurement work each of them needs. A steering group approves it, budgets are set against it, and every AI implementation plan in the organisation should trace back to one of its items. When the emphasis is on delivery, the same document is often called an AI implementation roadmap.

It sits between two other documents. The AI strategy states the goals and limits, the roadmap turns them into a sequence, and each item that reaches its delivery gate gets an implementation plan with owners, dates and acceptance criteria. The implementation playbook describes the programme phases that the roadmap schedules.

Who owns the roadmap

One person owns it and one body approves it. The owner is usually the chief AI officer, head of AI or CIO. The approver is the AI steering committee, or the executive committee where there is no steering committee. Each swimlane has a lead who keeps that row accurate. Where an AI centre of excellence runs intake, it maintains the register on the owner’s behalf: setting up an AI CoE covers that role, and enterprise AI team roles covers the people.

Strategy roadmap, adoption roadmap or transformation roadmap?

The labels describe emphasis, not different documents. An AI strategy roadmap starts from strategic goals and maps initiatives to each one. An AI adoption roadmap concentrates on people using AI well in current processes, so its people and skills lane carries the most items. An AI transformation roadmap redesigns processes and operating models around AI, which adds process-redesign and data work and usually lengthens the horizon. A generative AI implementation roadmap limits the scope to generative systems and needs extra evaluation and guardrail items in the governance lane.

Process redesign is where results show most clearly. McKinsey’s State of AI report, published in August 2026, found that nearly three-quarters of its AI high performers had fundamentally redesigned workflows because of AI, compared with about a quarter of other respondents. A roadmap without process-redesign items is an adoption roadmap, whatever its title says.

Horizons

Plan in quarters over 18 months. Confidence falls the further out you look, so print it on the roadmap and nobody will read an exploratory item as a promise. Ideas more than about two years away are rarely scoped; leave them in the strategy until they are.

Look beyond the first launch as well. In a Gartner survey of 432 respondents, run in late 2024 and published in June 2025, 45% of leaders at high-maturity organisations said their AI initiatives stayed in production for three years or more, against 20% at low-maturity organisations. Live systems need run, review and improvement items on the roadmap, not only launches.

Swimlanes

The six lanes cover the same ground as Gartner’s AI roadmap guidance, which names strategy, value management, organisation, talent, governance, engineering and data as the foundations, and judges progress by adoption goals, business value delivered and the maturity of each foundation. Here strategy sits above the roadmap, and organisation and talent share the people lane.

Check the balance between lanes too. BCG’s rule of thumb, restated in a January 2026 article on scaling AI, puts 10% of the effort into algorithms, 20% into technology and data and 70% into people and processes. A nearly empty people lane means the roadmap plans the smallest part of the work; a change management approach for AI helps fill it. The data lane deserves the same scrutiny. In April 2026, Gartner reported from a survey of 353 data, analytics and AI leaders that organisations with successful AI initiatives invested up to four times more of their revenue in foundations such as data quality, governance, AI-ready staff and change management than organisations with poor results. For the value lane, pick measures from the AI KPI guide.

Where the platform lane includes running models in your own data centre, the infrastructure track for on-premises AI breaks that row into phases, and the procurement guide covers choosing a vendor.

The roadmap has two parts. The register is the source of truth, with one row per item. The quarter grid is the picture people look at, and it should be produced from the register rather than edited by hand.

Roadmap register

Quarter grid

Keep the grid to IDs and short names; the detail lives in the register.

Items earn a slot through one intake route, not through seniority. There are five steps.

1. Intake. Capture every idea on one form with the problem, users, data and expected value. The use case intake form is a ready-made version.
2. Scoring. Score candidates on value, risk and effort. This template does not define that method; it takes the output of scoring use cases by value, risk and effort, or of the interactive use case scoring tool.
3. Foundation check. List what each high scorer needs in the other five lanes. Missing foundations become items with their own owners.
4. Capacity check. Commit no more use cases per quarter than you have delivery teams to build them and business capacity to absorb the change. One committed use case per delivery team per quarter is a sensible starting rule.
5. Placement. Committed items need an approved business case and a named owner. Items with open feasibility questions enter as exploratory, with a proof of concept as their next gate.

A readiness baseline helps with steps 3 and 4. One pass through the readiness and maturity assessment shows which foundation lanes need the most items.

Dependencies

Mark critical dependencies on the grid with arrows or ID references. When a dependency slips, the item that depends on it moves too, and showing that honestly is most of the point of a roadmap.

Gates

Treat cancellation as a normal outcome. Gartner predicted in June 2025 that over 40% of agentic AI projects will be cancelled by the end of 2027 because of escalating costs, unclear business value or inadequate risk controls. A proof gate and a retire gate bring those decisions forward, when they cost less. Gartner’s October 2026 CIO survey points the same way: organisations with guidelines for abandoning low-value AI initiatives already in production were 2.6 times more likely to be among its vanguards, a group defined by better business outcomes and high AI value in the eyes of the CEO. Run the go-live gate exactly as your governance policy’s approval gates define it, so the roadmap and the governance committee hold one decision rather than two.

Regulatory dates as fixed markers

Put legal deadlines on the grid as fixed markers. Under the EU AI Act as amended by the Digital Omnibus on AI, obligations for stand-alone high-risk systems listed in Annex III apply from 2 December 2027, and those for high-risk AI in products covered by Annex I from 2 August 2028. Any planned use that could be high-risk, such as AI that filters job applications, needs its classification and deployer controls scheduled before it goes live. The EU AI Act timeline for deployers lists every milestone.

Everything in this example is invented. The organisation is an engineering and construction group with 6,500 employees in three EU countries, working for infrastructure and energy clients whose contracts restrict where project documents may be processed. Its AI strategy, approved in late 2026, sets three goals: more bids from the same bid team, less time spent searching project archives, and fewer delays in subcontract review. The roadmap was baselined in January 2027, so Q1 and Q2 2027 are committed, Q3 and Q4 2027 planned, and 2028 exploratory. If you are drafting an AI roadmap for 2027, the same calendar applies.

What the example shows:

• Foundations come first. D-01 and P-01 open the first quarter ahead of the UC-01 pilot, and D-02 and P-02 close in Q2, before the tender agent goes live in Q3.
• The high-risk candidate waits. CV screening support would fall under Annex III point 4(a), so UC-05 stays exploratory until the governance lane has screened every planned use (G-03) and put deployer controls in place (G-04) around the 2 December 2027 date.
• Certification is a decision, not an assumption. G-05 is a gap assessment against ISO/IEC 42001:2023, the December 2023 standard that sets requirements for establishing, implementing, maintaining and continually improving an AI management system. Certification is voluntary and is decided in Q2 2028; the ISO 42001 explainer gives the background.
• Value is tracked from the first quarter. Baselines are captured before each pilot, so the Q4 2027 board report compares like with like.

Boards hear more AI ambition than results. In Gartner’s survey of 2,002 CIOs and technology executives, published on 19 October 2026, only 13% said their enterprise had realised significant or transformational value from AI. McKinsey’s August 2026 survey found 37% of respondents reporting a positive EBIT contribution from AI, essentially unchanged from 2025. Present the roadmap as evidence and decisions, not as a list of activity, on one page with five blocks:

1. Goals and results. Each strategy goal, with value delivered so far against its baseline.
2. The next four to six quarters. At most eight headline items, each with its horizon label.
3. Risk position. The top risks from the register and the regulatory dates that shape the sequence.
4. Spend. Actual against plan, with run costs shown apart from build costs.
5. Decisions requested. Funding, policy approvals and items to stop.

Keep the full register and the dependency map in an appendix. The financial case for each committed item comes from its business case, and framing private AI for the board shows one way to set out the investment.

Publish each approved version with a number and a date, and keep a short change log of what moved and why. At every quarterly review the committed horizon rolls forward: planned items with approved business cases become committed, and exploratory items either move into the planned horizon or are dropped.

• A list of use cases with dates. Without data, platform, governance and people items, the dates are guesses.
• Equal certainty in every quarter. Drawing quarter six with the confidence of quarter one invites the board to hold you to it.
• Too much committed work. With more committed items than delivery teams, several slip at once.
• No way out. A roadmap without a retire gate only grows, and its run costs grow with it.
• No value lane. Without baselines, the annual report falls back on anecdotes.
• A vendor’s release plan pasted in. A supplier’s roadmap is an input to the platform lane, not the organisation’s plan.
• Regulatory dates kept elsewhere. Legal deadlines belong on the grid, where they constrain the sequence.

We reread every source listed here on 4 November 2026.

• Gartner: building and scaling an AI roadmap
• Gartner: AI maturity and three years in production (June 2025)
• Gartner: forecast of cancelled agentic AI projects (June 2025)
• Gartner: spending on data and analytics foundations (April 2026)
• Gartner: CIOs reporting significant AI value (October 2026)
• McKinsey: State of AI survey, 2026 edition
• BCG: scaling AI requires new processes (January 2026)
• European Commission: AI Act regulatory framework and timeline
• AI Act Service Desk: Annex III list of high-risk uses
• ISO: ISO/IEC 42001:2023 standard page