When Bertrand bought a duplex to rent out, he drew up a full renovation list on his first walk-through: new kitchen, updated bathrooms, refinished floors, fresh paint throughout, new windows, a landscaped front yard. The contractor told him the number and he nearly backed out. Then she said something useful. Make the kitchen livable first. Rent it. Use that rent to fund the bathroom. Bertrand listened. Eighteen months later both units were updated, fully rented, and he had not taken on a dollar more in debt than he planned. The same sequencing logic applies exactly to bringing AI into a small business.

A roadmap is what bridges strategy and execution. It shows what you are building, when you are building it, why you are sequencing it that way, and what success looks like when you get there. Without a roadmap, strategy stays abstract and expensive, a set of intentions you restate every quarter. With one, it becomes operational reality: a specific thing you are doing this month, for a reason you can articulate, with the next thing already lined up behind it.

A Roadmap Is Not a Wish List

A wish list is every AI tool or capability you would love to have, arranged in no particular order, with no budget attached and no timeline. Most small business AI plans are wish lists dressed up as strategy, and they are recognisable by the fact that nothing on them ever moves. A roadmap is different because it answers four questions for every single item on it.

1. What exactly will this do for my business?
2. What does it cost in money and in time?
3. When do I need it to produce results?
4. What has to be true before I add this, and what therefore comes first?

The fourth question is the most important one. It is what turns a list into a sequence. You do not refinish the floors before the plumbing is done, and you do not buy an AI-powered customer analytics platform before your basic customer data is organised. Anything on your list that cannot answer all four questions is a wish list item, and it belongs in a separate note rather than on the roadmap.

Three Views of the Same Roadmap

Larger organisations maintain the same roadmap in three views, because different audiences need different levels of detail from it. The distinction is worth understanding even in a very small business, because the three views correspond to three genuinely different questions, and confusing them is why so many plans satisfy nobody.

At small business scale you are all three audiences, so you do not need three documents. You do need to notice which question you are asking. The strategic view asks whether this is the right thing to do at all. The operational view asks what has to happen next week for it to progress. The technical view asks what data and tooling must exist first, and it is the one owners skip most often, which is why so many AI purchases stall on the discovery that the underlying information was never in a usable state.

Sequencing Around Dependencies

The key to a realistic roadmap is identifying dependencies and sequencing around them rather than around enthusiasm. Dependencies come in five kinds, and most stalled plans turn out to have missed one of the last three.

Technical dependencies. One initiative needs a capability built by another. A recommendation engine depends on the data infrastructure existing first. Capability dependencies. One initiative needs skills acquired during another. You cannot do advanced work without someone who understands it, and that understanding is itself a deliverable of the earlier phase. Data dependencies. One initiative needs data collected or cleaned during another. Prediction of any kind needs a history of outcomes to learn from, and that history has to have been recorded.

Change dependencies. One initiative needs the working habits established by another. New decision-making processes built around your first AI tool have to be settled before a second tool relies on them. Business context dependencies. One initiative only makes sense once another has succeeded, which is the logic behind testing with a small group before rolling out broadly. Mapping these explicitly is what stops a roadmap from being an optimistic ordering of things you want.

The artefact this produces is simple: a list of initiatives, the dependencies each one carries, and the earliest date it could therefore start. The example below is drawn at organisational scale, but the shape is the same whatever your size, and the useful discipline is filling in the middle column honestly before you fill in the third.

The Three-Wave Model for a Small Business

The most practical structure for a small business AI roadmap has three waves, and each one funds and enables the next.

Wave 1: quick wins, days 1 to 60

This wave is about getting one or two AI tools working well enough to save measurable time within 60 days. The criteria are deliberately strict. The tool must be low cost, under 50 dollars a month. It must require minimal setup, under four hours in total. And it must produce a time saving you can measure in hours per week. Good candidates are AI writing assistance for emails, social posts or product descriptions; an AI scheduling tool that replaces manual back-and-forth; or a chatbot handling after-hours website questions.

Wave 1 is not about transforming your business. It is about making the kitchen livable. You are proving to yourself, and to anyone who works with you, that AI actually works in your specific context before committing to anything larger. That proof is the actual deliverable, and the time saved is a bonus.

Wave 2: core processes, months 2 to 6

Once you have documented savings from Wave 1, use part of them to fund Wave 2. This wave applies AI to a core business process, something you do repeatedly that touches customers or revenue directly. Common examples include automating customer follow-up after a purchase or appointment, using AI for first-pass responses to enquiries before a human reviews them, or integrating AI into inventory and ordering so restocking is flagged rather than remembered.

Wave 2 tools generally cost more and take longer to adopt, in the range of 100 to 300 dollars a month and two to four weeks to fully settle in. The Wave 1 savings justify the investment and pay for the learning curve, which is the part people forget to budget for.

Wave 3: growth tools, months 7 to 12

Wave 3 covers tools that need more data, more integration and a longer horizon before they return anything: analytics that show customer patterns over time, AI-assisted pricing, multi-channel marketing automation. These are powerful and fragile if deployed early. They need clean data, established workflows and people already comfortable with AI, and Waves 1 and 2 are what build all three of those conditions.

You do not renovate the guest bathroom while the kitchen ceiling is still leaking. Fix what tenants need first, then fund what makes the property more valuable.

How Larger Organisations Phase the Same Work

Organisations with teams run the same logic over a longer clock, and it is useful to know the shape if you are growing into it. Phase one, months 1 to 6, runs foundational work such as data infrastructure, hiring and governance in parallel with quick wins such as content generation, document automation and simple chatbots. The foundation work is slow but necessary; the quick wins build momentum and demonstrate value while it happens. Phase two, months 7 to 18, leverages that foundation to launch two or three strategic initiatives at once, alongside a second round of quick wins. Phase three, months 19 to 36, deepens integration until the organisation moves from having AI to operating through it.

The pacing rule matters more than the timeline. Organisations that try to launch ten initiatives simultaneously run out of resources, scatter their attention and succeed at none of them. Two or three initiatives per quarter is a realistic starting velocity, increasing as capability builds, and even mature organisations rarely exceed five or six concurrent initiatives because coordination overhead grows faster than the initiative count does.

Making Each Phase Fund the Next

This only works if you actually track what Wave 1 saves. Before deploying your first AI tool, write down how long the manual version of that task takes each week. After 30 days, measure again. The difference is your return, and it is the only number that makes the next purchase an easy decision rather than an anxious one.

Worked through: if AI handles your social media drafts and saves three hours a week, and you value your time at 40 dollars an hour, that is 120 dollars a week, roughly 480 dollars a month, in recaptured time. That comfortably justifies a 150 dollar a month Wave 2 tool with room to spare. The arithmetic is trivial. What is not trivial is having written down the before number, because nobody can reconstruct it afterwards and everybody underestimates it in hindsight.

Bertrand tracked this by putting a sticky note inside his rental unit's electrical panel: kitchen done, three months, 4,200 dollars, monthly rent offset 800 dollars. He needed that note when the bathroom estimates came in and he almost got scared off. The maths was already there, written down when he was calm, waiting for the moment he was not.

Tracking What Is Actually Happening

For each item on the roadmap, track six things against the plan. Timeline, because an early warning that something is slipping is what makes course correction possible at all. Budget, because cost overruns force reallocation and it is far better to find that out early. People, because assigned time being quietly pulled onto other work is the most common real bottleneck. Milestones, and specifically what is causing any delay rather than merely that there is one. Assumptions, meaning whether the things you believed when you approved this are still true. And outcome, whether the results so far point toward the business result you bought it for.

Rhythm makes this sustainable. Run the first two months of a quarter as execution and the third as review and planning for the next one: what got done, what slipped and why, what you learned, whether your assumptions held, whether outcomes matched expectations. Then adjust. If something proved harder than expected, extend the timeline or add resource. If it proved easier, pull the next item forward. If your market changed, reassess priorities before continuing on momentum.

Out of this comes your velocity, which is simply how much you actually complete per quarter. It is the most useful planning number you will ever have, because it is measured rather than hoped for. Organisations operating at maturity tend to complete two to three major initiatives and two to four quick wins per quarter. Whatever your own figure turns out to be, build future roadmaps on it rather than on optimistic estimates, and multi-quarter forecasts stop being fiction.

When a Tool Does Not Work as Expected

Something in your roadmap will not work as planned. A tool will underperform, a vendor will change pricing, your team will resist adoption. Treat this as a renovation surprise, a pipe behind the wall you did not know about, rather than as evidence that the whole plan was wrong. Three moves handle almost every case.

1. Give it a defined trial extension. Do not decide in the moment of frustration. Set a two-week extension with one specific metric you are watching. If it does not hit that metric, cut it.
2. Do not let it block the next wave. If your Wave 1 chatbot is struggling, do not freeze Wave 2 indefinitely. Fix it, replace it, or deprioritise it and move to a different Wave 1 tool.
3. Document what failed and why. One paragraph in a notes file is enough, and it stops you buying the same thing again in six months when a new version arrives with a better demo.

Four situations deserve a deliberate decision rather than an attempt to absorb them: two initiatives competing for the same person's time, one technical blocker holding up several things at once, pressure to add scope or compress a timeline beyond what is realistic, and a change in your market that suggests the priority order is now wrong. In a larger organisation each of these escalates to someone. In a small business you are that someone, and the discipline is to make the trade-off explicitly instead of quietly trying to make everything work at any cost.

Keeping the Roadmap Visible and Alive

A roadmap that lives only in your head, or in a file you never open, is not a roadmap. It is a memory, and memories fade. Make yours visible in two ways. Keep a one-page summary showing your three waves, what is in each, and what is currently active. Print it and put it somewhere you look at weekly: a wall, a desk, the inside of a cabinet door. Then hold a monthly review of about fifteen minutes. First Tuesday of the month, look at the page and ask whether last month's tool did what you expected, what is next in the queue, and whether anything should move earlier or later.

Larger organisations carry more artefacts than this, typically a short summary for whoever funds the work, a detailed working document for whoever delivers it, and a quarterly narrative explaining what is happening now and why. A small business needs only the one-pager and the narrative, and the narrative can be a paragraph you write for yourself. What matters is not the format but that a version exists which someone other than you could read and understand.

When the roadmap changes, and it will, explain why. This sounds like a large-organisation concern, but it applies to any business with staff: people who see priorities change without explanation lose confidence in the plan and start hedging. The pattern to copy sounds like this. We planned to launch the recommendation work this quarter and we have pushed it to next quarter, because it depends on groundwork that turned out to be more complex than we expected. We have accelerated that groundwork and we now know what it needs. In the meantime we are launching a simpler version to deliver value while the foundation stabilises.

Bertrand has a handwritten index card on his refrigerator with four lines: kitchen done, bathroom in unit one done, bathroom in unit two in progress, windows next year. That is his roadmap. It has been there for eighteen months. He crosses things off, and he is ahead of schedule. A wish list grows. A roadmap moves.

Anti-Patterns

• Sequencing by excitement. Ordering initiatives by how interesting they sound rather than by what has to be true first is the defining feature of a wish list.
• Skipping the data question. Buying a tool that needs clean historical data before checking whether you have any is the most common way an AI plan stalls.
• Running everything at once. Launching too many initiatives simultaneously scatters attention and resources, and typically means nothing finishes well.
• Not writing down the before number. Without the manual baseline captured in advance, you can never prove a saving, and the next purchase becomes a matter of faith.
• Letting one failure freeze the roadmap. A struggling Wave 1 tool is a reason to fix, replace or drop that tool, not a reason to suspend everything behind it.
• Planning on optimistic velocity. Building future phases on what you hope to complete rather than what you have historically completed produces a schedule that is wrong from the first week.
• Changing the plan silently. Reordering priorities without explaining why teaches everyone around you that the plan is decoration.

Practice Prompts

• Take your current AI wish list and try to answer all four roadmap questions for each item. Move anything that fails into a separate parked list.
• For your top three items, write out the dependencies in each of the five categories, then set an earliest possible start date from them.
• Pick your Wave 1 candidate and check it against all three criteria: under 50 dollars a month, under four hours of setup, measurable weekly time saving.
• Before deploying anything, time the manual version of the task for one week and write the number down where you will find it again.
• Draft the one-page roadmap summary, print it, and put it somewhere you physically pass every week.
• Schedule the monthly fifteen-minute review in your calendar as a recurring event you do not cancel.
• Write the paragraph you would use to explain a delay to your team, for an initiative that has not slipped yet.

Reflection

Look at whatever currently stands in for your AI plan and ask which of the four questions it can answer for each item. Most owners find they can answer the first two easily, the third vaguely, and the fourth not at all, which is exactly why the plan has not moved. Then ask a harder question about your own behaviour. When a tool disappoints you, is your instinct to cut it immediately, to keep paying for it out of sunk-cost discomfort, or to set a defined trial with a metric and decide on evidence? The answer says more about whether your roadmap will survive contact with reality than the roadmap itself does.

Glossary

• Roadmap. A sequenced plan where every item states its purpose, cost, deadline and prerequisites, as distinct from a list of things you want.
• Wave or phase. A grouping of initiatives that share a timeframe and a purpose, where each group builds the conditions the next one needs.
• Dependency. Something that must be true before an initiative can start, whether technical, capability, data, change or business context.
• Velocity. How much work you actually complete per quarter, measured from history and used to forecast rather than estimated optimistically.
• Quick win. A low-cost, low-setup initiative chosen to produce measurable savings fast and prove the approach works in your context.
• Baseline. The recorded duration or cost of the manual process, captured before deployment so the saving can be proved afterwards.
• Trial extension. A defined additional period with one named metric, used to decide on an underperforming tool by evidence rather than by mood.

Related Lessons

This lesson turns planning into sequence, so it sits directly downstream of Strategic Foundations and AI Vision Setting, which establishes what you are trying to achieve before you order the steps. Resource Allocation and Budget Planning supplies the money side of the four questions, and the funding logic in the wave model assumes you have done that work. Identifying Your Next Wave of AI Use Cases is where the items on the roadmap come from once the first wave is complete.

For the execution end, Planning Your First AI Pilot Project expands what a Wave 1 initiative looks like in practice, and Communicating AI ROI to Leadership covers presenting the savings you have documented to anyone who needs convincing. The natural next step after roadmapping is Risk Assessment and Mitigation Planning, which addresses what could derail execution and how to plan for it before it happens rather than after.

Closing

A roadmap translates intention into a sequence you can actually execute. Build it around dependencies so the order reflects reality rather than enthusiasm. Run it in waves so each phase funds and enables the next. Track what you complete so your forecasts are based on measured velocity rather than hope. Review monthly, adjust quarterly, and explain the changes to anyone they affect. The best roadmap is not the most detailed one; it is the one that gets executed and adjusted as you learn. Bertrand's is an index card on a refrigerator with four lines on it, and it has kept a two-unit renovation on budget for eighteen months.

Key Takeaways

• A roadmap answers four questions per item: what it does, what it costs, when it must pay off, and what has to come first. Anything that cannot is a wish list item.
• Sequence around the five kinds of dependency, technical, capability, data, change and business context, rather than around what you are most excited about.
• Wave 1 is about proof, not transformation: quick wins inside 60 days using tools under 50 dollars a month and under four hours of setup.
• Wave 2 applies AI to a core process in months 2 to 6, funded by Wave 1's documented savings, typically at 100 to 300 dollars a month.
• Wave 3 holds the growth tools that need clean data and settled workflows, which is why they fail when deployed early.
• Record the manual baseline before you deploy anything, then measure again after 30 days, so each phase can fund the next on evidence.
• Track timeline, budget, people, milestones, assumptions and outcome, and build future plans on your measured velocity rather than optimistic estimates.
• When a tool disappoints, give it a defined trial extension against one metric, then cut it, and never let one failure freeze the rest of the roadmap.
• Keep a printed one-page summary and hold a fifteen-minute monthly review. A wish list grows; a roadmap moves.

Frequently Asked Questions

What is the difference between a roadmap and a project plan? A roadmap shows which initiatives you are pursuing and in what sequence, at a relatively high level, covering themes, dependencies and phasing over a long horizon. A project plan details exactly how one of those initiatives will be executed: tasks, milestones, resources and their interdependencies. You need both, and confusing them produces documents that are simultaneously too vague to act on and too detailed to read. The roadmap answers what you are doing and why. The project plan answers how you will do it.

How do I handle dependencies between initiatives? Map them explicitly before sequencing anything. For each initiative, write down what must be true before it can start: data available and clean, a skill acquired, a previous initiative proven, a working habit established. Use that map to inform the order. Where dependencies are soft, parallelise to compress the timeline. Where one initiative genuinely blocks another, sequence strictly and accept the delay rather than starting anyway and discovering the block later at higher cost. Balance dependency constraints against your resource constraints, since both are real.

What is a realistic timeline for AI adoption? Quick wins and proofs of concept run three to six months. Initial implementation of a real capability runs six to twelve months. Mature capability development runs eighteen to thirty six months, and full organisational transformation runs three to five years. Most people underestimate all of these. The bottleneck is usually organisational, meaning people adopting new ways of working, rather than technological. Plan conservatively on timeline, and treat any vendor estimate that assumes instant adoption as marketing.

How often should the roadmap be updated? Review the strategic roadmap annually and adjust it quarterly based on what you have learned, with a monthly check that takes only a few minutes. When market conditions shift significantly, or a major initiative hits an unexpected delay, assess the impact and adjust deliberately. The failure modes sit at both extremes: updating reactively after every disruption produces a plan nobody trusts, and ignoring genuine changes produces one nobody follows. Quarterly for adjustments and annually for a proper refresh is the sustainable rhythm.

How do I keep everyone aligned on the roadmap? Produce the version each audience needs rather than sending everyone the same document: a short summary for whoever funds the work, a detailed working version for whoever delivers it, and a regular narrative covering the next period in more detail. Share them broadly and on a schedule rather than only when something changes. When the roadmap does change, explain why, and communicate early successes and lessons as well as delays. Regular communication stops people guessing at priorities, and guessing is what produces work you did not ask for.