The AI automation roadmap: timeline, costs, risks and ROI

This is what a typical first year looks like when a UK business automates properly: five phases, each with a fixed scope, a UK-market cost, and a decision point where you can stop. No six-figure first invoice, no lock-in, and nothing automated that a human should still approve.

The five phases

Each phase stands alone. Foundations are skipped when your systems are already sound, and the bigger builds only happen after the smaller ones have paid back.

Phase 0 · Weeks 1–2

£499(credited against any build)

AI Ops Audit

A fixed-price audit of one part of your business. We map how the work actually flows today, count the hours it costs, and hand you a written plan with confirmed build costs for the first project.

  • Process map of the area you chose
  • Hours and cost of running it today
  • Recommended first build with a fixed price
  • What to leave human and why

Decision point: You decide whether to build. The £499 is credited against any build that follows.

Phase 1 · Weeks 3–6

£4,000–£12,000

Foundations

Only needed if the audit finds your data lives in inboxes and spreadsheets. We connect the systems the process touches — email, accounts, CRM, job management — so information flows into one place before any AI touches it.

  • Existing systems connected (keep the software you already pay for)
  • One shared home for the process data
  • Re-keying between systems removed

Decision point: Skipped entirely if your foundations are already sound — the audit tells us.

Phase 2 · Weeks 5–12

£6,000–£15,000

First automation

One well-scoped automation, built end to end: reading enquiries, preparing quotes, chasing paperwork, or weekly reporting. A human approval step is built in at every decision that carries consequence.

  • The working automation in your live systems
  • Human approval steps at every consequential decision
  • A 30-day pilot with your team using it for real

Decision point: Go / no-go on expanding, based on the pilot numbers.

Phase 3 · Months 3–6

£15,000–£35,000

Bespoke AI agent

For the biggest bottleneck: an agent that handles a whole workflow — qualifying enquiries, updating records, drafting quotes, chasing customers — escalating to your team whenever it isn't confident.

  • An agent scoped to one major bottleneck
  • Confidence-based escalation to a named human
  • Full audit trail of everything it does

Decision point: Only started once Phase 2 has proven the maths.

Phase 4 · Ongoing

£300–£1,500/month

Run & expand

Monitoring, tuning and support for what's live, plus the next process from the audit plan. Each build should pay for itself faster than the last because the foundations are already there.

  • Monitoring, maintenance and improvements
  • Quarterly review against hours actually saved
  • The next process from your audit plan

Decision point: Cancel any time. Everything built is yours.

The risks — and how they're designed out

Most failed AI projects fail in predictable ways. These are the five we see most, the warning sign for each, and where in the roadmap it's handled.

Automating before the data is ready

Warning sign: AI gives confident but wrong answers because the inputs live in someone's inbox.

How it's handled: Phase 1 exists precisely for this — no AI touches a process until its data lands in one system.

Scoping too wide

Warning sign: "Automate our operations" instead of "automate enquiry follow-up". Projects die of breadth, not depth.

How it's handled: Every phase builds one process. The audit picks the one with the clearest payback.

No human approval step

Warning sign: A wrong quote or a compliance miss goes straight to a customer.

How it's handled: Anything carrying spend, safety or compliance consequence keeps a named human approver. Non-negotiable.

The team doesn't use it

Warning sign: The system works but everyone keeps doing it the old way.

How it's handled: The 30-day pilot runs with your team before any expansion, and the audit interviews the people who do the work today.

Replacing working software mid-project

Warning sign: The automation project turns into a software-migration project and stalls.

How it's handled: We connect what you already use rather than replacing it. Migration is only ever a separate, deliberate decision.

How the ROI maths works

Payback comes from four places. The first is the easiest to count; the other three are usually worth more. A well-chosen first process typically pays back in 6–12 months.

Hours recovered

10 hrs/week at £25/hr loaded cost

≈ £13,000/yr

Faster response

Quotes out in hours, not days — higher win rate

Varies by pipeline

Errors removed

No re-keying between systems, fewer credit notes

Often 5–15% of admin cost

Capacity freed

Same team handles more volume without hiring

One hire deferred ≈ £30k+

Run your own numbers in the AI Automation ROI Calculator — it uses the same maths with figures from real Forth platforms.

Common questions

How long does a typical AI automation project take?

From first audit to a live first automation, typically 8–12 weeks. The audit itself is two weeks, foundations (when needed) run weeks 3–6, and the first automation is built and piloted by weeks 5–12. Larger bespoke agents follow in months 3–6, once the first project has proven its numbers.

How much does the whole roadmap cost?

£499 for the audit (credited against any build), £4,000–£12,000 for foundations if needed, £6,000–£15,000 for the first automation, and £15,000–£35,000 for a bespoke agent tackling a major bottleneck. Most businesses spend under £25,000 in year one and recover it in hours saved within 6–12 months.

What is the payback period on AI automation?

For a well-chosen first process — high volume, low judgement — payback is typically 6–12 months. Ten hours a week recovered at a £25 loaded hourly rate is roughly £13,000 a year, which covers a £6,000–£15,000 first build within the year, before counting faster quotes or avoided hires.

What is the biggest risk in an AI automation project?

Automating before the data is ready — AI working from inputs that live in inboxes and spreadsheets produces confident nonsense. The second biggest is scoping too wide. Both are designed out here: foundations come first, and every phase builds exactly one process.

Do I have to commit to the whole roadmap?

No. Every phase ends with a decision point. The audit stands alone, the £499 is credited if you build, and you can stop after any phase with everything built so far being yours. Nothing is a lock-in.

Where does my readiness score fit in?

The AI Readiness Score tells you which phase you'd actually start at. A high score means straight to the audit and a first automation; a lower score means the first project is foundations. It's a two-minute way to locate yourself on this roadmap.

Find out where you'd start

The two-minute AI Readiness Score tells you which phase your business is actually ready for — then the Process Automation Workshop turns your weakest process into a costed plan.

LIVE PROJECT

See this roadmap run on a real job

Kerbco went from an eleven-step manual permit chain to one enquiry form. Walk the same five phases with the real work inside each one.

Related: 2-Minute AI Opportunity Scan · ROI Calculator · What does AI automation cost in the UK?