Consult · Executives
How to build an AI roadmap for your company
An AI roadmap is a sequenced plan that names which business processes a company will apply AI to, in what order, with what budget, owners, and success measures. A good AI roadmap starts from operational problems rather than tools, ranks candidates by value and readiness, and schedules small pilots before large rollouts.
- Reading time
- 9 min
- Chapters
- 6
- Last reviewed
- November 10, 2026
- Author
- Lingows AI Team
- Difficulty
- Intermediate
Before you start
- An executive willing to sponsor the roadmap and make stage decisions
- Access to department leads and frontline staff for short interviews
- A rough view of which systems hold the company's operational data
The steps at a glance
- Step 01
Interview the people who do the work
Gather pain points, volumes, and data locations from department leads and frontline staff.
- Step 02
Build a process inventory
Record every candidate process in one consistent format with owner, volume, systems, and risk.
- Step 03
Score each candidate
Rate business value, data readiness, feasibility, risk, and change effort on a simple scale.
- Step 04
Sequence the work in stages
Plan a foundation stage, one pilot, a review, and then a rollout or the next pilot.
- Step 05
Set governance and budget
Name owners, approve a use policy, and budget for running costs as well as the build.
- Step 06
Review the roadmap every quarter
Rescore open candidates and retire projects that no longer make sense.
Chapter 01
What is an AI roadmap and why does a company need one?
An AI roadmap is a written plan that turns scattered AI interest into an ordered list of projects with owners, budgets, and measures. Without a roadmap, companies tend to buy tools before defining problems, run disconnected pilots, and struggle to explain AI spending to leadership or to the teams expected to change how work gets done.
Most companies already have AI activity before they have an AI plan. Individual staff use chat assistants on their own, a vendor adds an AI feature to a product the company already pays for, and a department head asks for a pilot after a conference. None of that is wrong, but none of it adds up to a direction.
A roadmap gives leadership one place to decide. The roadmap lists the candidate processes, the order in which they will be tackled, the budget each stage needs, and the measure that decides whether the next stage happens. A good roadmap also records what the company has decided not to do yet, which matters as much as the list of projects.
Chapter 02
How do you find where AI could help?
Finding where AI could help starts with an inventory of repetitive, text-heavy, or decision-heavy processes, gathered from the people who do the work. Interview department leads and frontline staff, list each process with volume, time spent, error impact, and data location, and capture pain points in plain language rather than in tool names.
Skip the question about where the company can use AI. People answer with tools they have read about. Ask instead where work piles up, where errors are expensive, where someone retypes information from one system into another, and where customers wait for an answer.
Record each process in the same format so the list can be compared later. The template below is a workable starting point and fits in a shared spreadsheet.
Process inventory template
| Field | What to record | Example |
|---|---|---|
| Process | A short name for the work | Invoice data entry |
| Owner | The person accountable for the outcome | Accounts payable lead |
| Volume | How often the work happens | Daily, for every incoming invoice |
| Time spent | Staff hours per week, estimated by the team | Estimated by the team during the interview |
| Inputs | Where the information comes from | Emailed PDFs from suppliers |
| Systems | Where the result is stored | Accounting system |
| Error impact | What goes wrong when a mistake happens | Late or duplicate payments |
| Data sensitivity | Whether personal, financial, or regulated data is involved | Supplier bank details |
Chapter 03
How do you decide which AI projects come first?
Deciding which AI projects come first means scoring each candidate on business value, data readiness, technical feasibility, risk, and change effort, then favoring projects that score well on value and readiness at the same time. A high-value process with messy, scattered data usually belongs later on the roadmap, after foundational data work is complete.
Use a simple one to five score for each factor and keep the process owner in the room while scoring. The numbers matter less than the arguments the scoring forces. A team that cannot agree on the value of a process has not yet defined the problem, and that is worth discovering before money is spent.
Scoring factors
| Factor | Question to ask | A low score means |
|---|---|---|
| Business value | What changes for customers, revenue, cost, or risk if the process improves? | Improvement would be nice, but nobody would notice |
| Data readiness | Is the information the process needs digital, accessible, and reasonably clean? | Data lives in paper files, inboxes, or someone's memory |
| Feasibility | Can current models and tools handle the task reliably? | The task needs judgment the business cannot yet describe |
| Risk | What happens when the system is wrong? | Errors reach customers or regulators without review |
| Change effort | How much do people and processes need to change? | Several teams must change how they work at once |
Chapter 04
How should an AI roadmap sequence pilots and rollouts?
An AI roadmap should sequence work in small, measured stages: a foundation stage for data access and policy, a first pilot on one high-readiness process, a review against agreed measures, and only then a wider rollout or the next pilot. Each stage ends with a go, adjust, or stop decision recorded by an owner.
Sequencing protects budget. A pilot that runs for a fixed period against a defined measure produces a clear decision. A pilot without an end date tends to linger, half adopted, and blocks the next project on the list.
A typical sequence
- 01FoundationApprove an AI use policy, confirm which tools staff may use, and fix the data access the first pilot depends on.
- 02First pilotRun one process with a small group, a human review step, and a baseline measured before the pilot starts.
- 03ReviewCompare results with the baseline, gather feedback from the people using the system, and decide go, adjust, or stop.
- 04Rollout or next pilotExtend a successful pilot to more users, or start the next candidate from the scored list.
The guide on picking a first AI use case covers the first pilot in more detail, including the warning signs that suggest a candidate belongs later on the roadmap.
Chapter 05
What governance and budget does an AI roadmap need?
An AI roadmap needs a named executive sponsor, an owner for each project, a written AI use policy, a review process for data and security, and a budget that covers running costs as well as build costs. Model usage, monitoring, maintenance, and staff training continue after launch and belong in the plan from the start.
Running costs surprise companies more often than build costs. Every AI system that calls a model pays for usage, needs monitoring for errors and drift, and needs updates when the underlying models or connected software change. Budget for the run, not only the launch. The answer on measuring AI ROI covers how to set the measures that justify that spending.
Governance does not need a committee for every decision. A short policy that says which tools are approved, which data may and may not be used, when a person must review output, and whom to ask when unsure covers most situations a team will meet.
Checklist
Roadmap governance checklist
Chapter 06
How often should an AI roadmap be updated?
An AI roadmap should be reviewed at least every quarter and after every pilot decision, because model capabilities, prices, and vendor offerings change quickly. A quarterly review rescores open candidates, retires projects that no longer make sense, and adds new processes surfaced by teams who have started using AI in daily work.
The model landscape shifts often. Providers such as OpenAI, Anthropic, Google, xAI, and Meta release new models and change prices on their own schedules, and capabilities that were unreliable a year ago may be dependable now. A roadmap written once and filed away steers the company with old assumptions.
Keep the roadmap vendor neutral where possible. Write each project in terms of the outcome and the process, then choose the model or product at build time. A change in the market then updates one decision instead of forcing a rewrite of the whole plan.
Lingows AI builds roadmaps this way in AI consulting engagements. TitanWave, a Lingows group company, produces a structured AI Blueprint that names where AI will work for a company and where AI will not.
01Summary
Key takeaways
What to remember
- Start an AI roadmap from operational problems, not from tools or vendors.
- Inventory processes in one consistent format so candidates can be compared.
- Favor projects that score well on both business value and data readiness.
- Run small pilots with baselines and a recorded go, adjust, or stop decision.
- Budget for model usage, monitoring, maintenance, and training after launch.
- Review the roadmap every quarter, because the model market changes quickly.
02FAQ
Questions about this guide
What takes the most effort when building an AI roadmap?
The interviews and the process inventory take the most effort when building an AI roadmap, because the work depends on access to busy department leads and frontline staff. Scoring, sequencing, and writing the plan move faster once the inventory is complete and the process owners agree that the list is accurate.
Who should own the AI roadmap?
An executive sponsor should own the AI roadmap, with a named owner for each project on the roadmap. The sponsor makes stage decisions and protects budget. Project owners are usually the leaders of the processes being changed, not the IT team alone, because the business outcome is the measure of success.
Should an AI roadmap name specific AI vendors?
An AI roadmap should describe projects by outcome and process rather than by vendor wherever possible. Choosing the model or product at build time keeps the plan stable while the market changes. A roadmap can still note an existing contract or platform when a project depends on software the company already runs.
What is the difference between an AI roadmap and an AI strategy?
An AI strategy states why and where a company intends to use AI and which principles guide the work. An AI roadmap is the operational plan underneath the strategy, listing specific projects, order, owners, budgets, and measures. Many mid-sized companies combine both into one short document that leadership reviews every quarter.
Can a small company build an AI roadmap without a consultant?
A small company can build a useful AI roadmap without a consultant by following the core steps: interview staff, inventory processes, score candidates, and run one small pilot. Outside help adds value when the company lacks time, technical judgment about feasibility, or a neutral voice to settle disagreements about priorities.
Written by
Lingows AI Team
Engineering and strategy team, Lingows AI
The Lingows AI team has built AI systems since 2017 and software since 2012. Collective posts cover model landscape changes, tooling selection, and the direction of applied AI.
Related software
TitanWave
AI onboarding that shows where AI works for a company and where it does not.
Put this guide to work with AI Consulting.
AI consulting turns a guide like this into a scoped plan for your company: where AI fits, where it does not, and what to build first.