The UK SME AI Automation Roadmap: a 30/60/90-Day Template That Survives Your Next Board Meeting

Most CEOs we meet have an AI strategy. Very few have a roadmap that survives a board meeting.

The strategy deck explains why AI matters. The roadmap explains what gets built, in what order, by when, and how anyone will know whether it worked. The difference is operational, and it is the difference between a programme that compounds and a programme that quietly stalls after the first pilot.

This guide is the 30/60/90-day template we use with UK SMEs in the £4M to £160M revenue band. It assumes you have already decided that automation matters. The question it answers is the next one, which is: now what, exactly, in what order, against which KPIs.

What goes in an AI automation roadmap?

A useful AI automation roadmap names six things: the business outcomes it serves, the processes you plan to automate in each phase, the order you build them in, the KPIs you will measure each phase against, who owns delivery, and the decision points that decide whether you continue. For a UK SME, it should fit on a single board page.

Below is the structure we hand to every CEO before the first build phase.

Roadmap component	What it answers	What it is not
Outcome anchor	The business result the roadmap exists to produce (hours saved, capacity unlocked, cycle time reduced)	A list of tools or vendors
Process queue	Which workflows get automated, in what order, across the first three phases	A wishlist of every manual task
Phase boundaries	What "done" looks like at 30, 60, and 90 days	A continuous, undated to-do list
KPI targets	The numbers that determine whether each phase succeeded	A list of activities completed
Named owners	Who is accountable for delivery and for measurement	A vendor relationship
Decision points	Where the board can choose to continue, pause, or change direction	An assumption that everything goes to plan

Six components, one page, three time horizons. That is the format. The rest of this post is how to fill it in honestly.

Gartner has reported that around 30% of generative AI projects are abandoned after proof of concept. PwC's 2026 Global CEO Survey found that 56% of CEOs have seen neither revenue gains nor cost savings from their AI investments to date. The pattern is almost never about model choice. It is about the absence of a roadmap that answers these six questions in order.

The 30/60/90-day template

The phased model below is deliberately conservative. It is designed to make the first board update boring in the best possible way: planned, on time, on target. Most UK SMEs we work with want exactly that, not a moonshot.

Phase	Time horizon	Outcome	Lead KPI	Decision point
Phase 1: Scope and validate	Days 0 to 30	One automation candidate fully scoped, mapped, and signed off	Process documented; baseline hours captured	Green-light the build, or pick a different first workflow
Phase 2: Build and prove	Days 31 to 60	First automation running in production with one team	4 to 8 hours saved per week, per user, on the automated task	Continue to expansion, or harden the build before connecting
Phase 3: Connect and compound	Days 61 to 90	First automation linked into adjacent workflow; second candidate scoped	Cycle time reduced on the connected process; capacity unlocked across the team	Commit to the next 90 days, or close the programme cleanly

Days 0 to 30: scope and validate

The first phase is the one most teams skip and pay for later. The goal is not to build anything. The goal is to be certain you are building the right thing.

In a typical Phase 1 we run a kick-off with the founder or sponsor, then three or four working sessions with the people who actually do the work today. We map the current process end to end, including the handoffs that nobody documented. We capture baseline numbers: hours per week, errors per month, cycle time, current cost-to-serve. Without those baselines, no later phase can prove value.

The output is a one-page scope: the workflow, the systems it touches, the proposed automated version, the KPI targets, the owner, and the risks. The board can read it in five minutes. If the working team cannot sign off on the scope at the end of week four, you have learned something useful before any build budget was spent.

The most common Phase 1 outcome we see is a swap. The workflow the CEO wanted to automate first is not actually the highest-leverage candidate. The people closer to the work usually know this, but rarely get asked. The same pattern shows up in revenue operations specifically: the team doing the work knows where the manual cost is, but the cost is invisible from above.

Days 31 to 60: build and prove

Phase 2 builds the first automation and proves it in production. One workflow, one team, measured against the baselines captured in Phase 1.

The target is concrete: 4 to 8 hours saved per week per user on the automated task, with at least two weeks of stable operation before the phase closes. In our own client work, the foundation phase consistently lands inside 6 to 10 weeks from scoping to stable production, with the build itself typically taking 3 to 5 weeks once the scope is locked.

Phase 2 also establishes the human-in-the-loop pattern that will apply to everything that follows. Anything customer-facing, anything regulated, and anything irreversible keeps a human in the decision. The principle that the best AI keeps you in control is what allows a UK SME board to approve the next phase without a long compliance debate.

What goes wrong in Phase 2 is almost always integration friction, not AI. The CRM API behaves unexpectedly. The data quality in one system is worse than anyone admitted. A field that everyone assumed existed turns out to be a free text comment box. Plan for this. Most of the Phase 2 budget covers integration and human-in-the-loop design, not model usage.

Days 61 to 90: connect and compound

Phase 3 is where most programmes stall, and where the compounding actually starts.

The job is to take the working automation from Phase 2 and connect it to the workflow on either side of it. A research automation feeds the CRM, which triggers the nurture sequence, which logs activity to the weekly review. Now a single change ripples through a connected system rather than living in isolation.

Phase 3 also scopes the second automation candidate. By day 90, the board should see two things: the first automation running with measurable hours saved, and the next candidate scoped to the same Phase 1 standard. That is what unlocks the next 90-day cycle.

The clearest proof point we have for the compounding effect comes from one of our clients in commercial flooring. The first automation removed roughly 25 hours of manual project paperwork every month, and once that was running, the second and third automations were faster to scope because the systems were already connected. The full case study walks through how that 25 hours got reclaimed.

The KPIs that survive a board meeting

A roadmap KPI has to do two jobs at once. It has to be measurable inside 90 days, and it has to mean something to a non-technical board. Most "AI KPIs" fail one or both tests.

The four we recommend leading with:

1. Hours saved per week, per user, on the automated task. Direct, baseline-comparable, and the easiest number to explain. Captured before build, then re-measured at the end of Phase 2 and the end of Phase 3. This is the metric that decides whether to continue.
2. Cycle time on the connected process. Lead-to-quote, order-to-delivery, brief-to-publish. A connected workflow should show a measurable cycle time reduction, not just per-task time savings.
3. Capacity unlocked, measured in FTE-equivalent. Hours saved across the team, converted into the equivalent number of full-time roles redeployed onto higher-value work. A board can compare this directly to a hiring plan. Harvard Business Review's 2025 analysis of AI adoption found that most firms struggle to capture real value from AI not because the technology fails, but because their people, processes, and governance do, which is why this KPI sits above any tooling metric on the roadmap.
4. Cost-to-serve per workflow. If you can baseline it, this is the metric that shows up in margin. It is the slowest to move, but the most defensible at the year-end review.

What does not belong on the roadmap: number of automations live, number of tools integrated, number of users trained, model accuracy in isolation. Each is a useful operational metric. None of them are board KPIs. The board needs to see whether the business can do more without growing headcount, not how many integrations the IT team configured.

What a UK SME roadmap is not

The format above is deliberately narrow. A few things that frequently appear on "AI roadmaps" we have reviewed, and that we would push back on every time:

A list of tools to deploy. Tools are an implementation detail. A roadmap that opens with "deploy n8n, deploy HubSpot AI, deploy Gemini" has skipped the question of what it is for. The choice of workflow tool follows the workflow, not the other way around.

A transformation programme that needs a steering committee. UK SMEs in the £4M to £160M band rarely need this layer of governance, and adding it usually adds three months of meetings before anything ships. A named owner and a clear decision point at each phase is enough.

A vendor-led document. If your roadmap was written by the vendor who is also going to deliver it, you have a proposal, not a roadmap. The two documents look similar from the outside, and the difference is whether the assumptions and the KPIs serve the buyer or the seller.

A perpetually-rolling 18-month plan. Anything beyond 90 days on an AI automation roadmap should be a sketch, not a commitment. The technology, the team's confidence, and the use cases all move faster than that. The 90-day cycle exists because it is the longest horizon a UK SME can sensibly commit to before the next decision point.

A plan that assumes the first guess is right. The whole point of Phase 1 is to validate or swap the first candidate. The plan that survives the board meeting is the one that names where it might be wrong, and what would change if it were.

For a parallel example of what this looks like on the strategy side rather than the roadmap side, our piece on building an AI automation strategy walks through the upstream framework that this roadmap sits inside. For the question of when to start at all, where to start with AI automation covers the pre-Phase 1 readiness check. And if the board reaction to your draft roadmap is "we are not ready for this yet", the angel investment community case study shows how a small founder-led team built confidence with a focused first automation before committing to a wider programme.

Frequently asked questions

What goes in an AI automation roadmap?

A useful AI automation roadmap names six things: the business outcomes the roadmap serves, the processes you plan to automate in each phase, the order you build them in, the KPIs you will measure each phase against, who owns delivery, and the decision points that decide whether you continue. For a UK SME, it should fit on a single board page. Anything longer is usually a vendor proposal, not a roadmap.

What is the typical AI implementation timeline for a UK SME?

The standard phased model is 30 days to scope and validate, 60 days to build and prove the first automation, and 90 days to connect that first automation into a wider workflow. Most UK SMEs in the £4M to £160M revenue band see measurable hours saved within 6 to 10 weeks of the build phase starting. Anything promising results inside the first 30 days is usually skipping the validation work that protects the investment.

How do you phase an AI automation roadmap?

Phase by confidence, not by tooling. Phase 1 (foundation) proves one high-frequency, cross-system task can be automated reliably. Phase 2 (expansion) connects that automation into the wider workflow around it. Phase 3 (optimisation) refines what is already running and adds the next adjacent process. Each phase should answer one question: does this work, does it connect, does it compound. Skipping straight to Phase 3 is the most common cause of failed AI programmes.

What KPIs belong on an AI automation roadmap?

Lead with hours saved per week, cycle time reduction on the automated process, and capacity unlocked measured in FTE-equivalent. Add cost-to-serve per workflow if you can baseline it. Avoid vanity metrics like "number of automations live" or "tools deployed". The board cares about whether the team can do more without growing headcount, not how many integrations the IT team configured.

What does a board-ready AI automation roadmap look like?

It is one page, with three time horizons across the top (30, 60, 90 days), the named workflows in each, the KPI target for each, and the owner. The supporting detail lives in an appendix. A board does not need the tooling diagram. It needs to see the path from today's manual work to a quantified outcome, and the decision point where you would stop or accelerate.

How much should a UK SME budget for the first 90 days?

Foundation budgets for UK SMEs in the £4M to £160M range typically sit between £6,000 and £25,000 for the first phase, depending on how many systems need to connect and how much process documentation already exists. The variation is mostly integration complexity, not the AI itself. Most of the cost is the discovery, mapping, and human-in-the-loop design, not the model usage.