How long does AI implementation take for a small business?

AI implementation for a small business is measured in weeks, and a well-scoped first build is measured in days. A single workflow goes live inside 7 to 10 working days, a set of up to three connected workflows takes 4 to 6 weeks, and working software should be running on your own stack inside the first month of an engagement. A supplier quoting quarters for a first workflow is describing its own process rather than the work.

What are the standard timelines for each build?

At the smallest end, an AI Front Door, the enquiry-handling layer for a sole trader or small local firm, installs in 5 to 10 working days. A Quick Win Install, which automates one costed workflow end to end, has that workflow live inside 7 days. Both timelines are possible because the scopes are deliberately tight, and tight scope is where the speed comes from.

A Core Workflow Setup covers up to three connected workflows, with CRM or case-management integration where it applies, and runs 4 to 6 weeks with the first workflow live inside 2 weeks. Managed AI Operations, built for larger or regulated organisations, runs as a continuing weekly rhythm and begins with an onboarding pilot inside 30 days.

What happens week by week after the first call?

From a standing start, the sequence is short. A 30-minute call, then a written audit of your highest-payback workflows within three working days. Week 1 to 2 produces a working prototype, built under NDA and your data-handling policy on real data or redacted samples, so the system is visible before any commitment to the full build. Week 3 to 6 is build and deploy: production code, the governance and monitoring around it, and training for your team.

The detail worth noticing is that the prototype arrives before the commitment. When a supplier's plan schedules months of discovery before anything runs, the calendar is protecting the supplier rather than the client.

What do real builds show?

The published timelines hold up against delivered work. A Midlands home-services business had its missed-call recovery workflow live in 7 working days. A UK SaaS founder team had its first workflow, drafting investor updates from approved inputs, live in 10 working days. A Manchester bookkeeping firm's document-intake workflow, the build that removed 84% of invoice and receipt re-keying, was live in 2 weeks.

Three weeks is the recurring figure for fuller workflows. A South East lettings firm's complete maintenance-triage workflow, a UK insurance broker's renewal follow-up and a regional law firm's matter-intake pilot each went live in 3 weeks, the last with human approval in place from day one.

What stretches an implementation timeline?

Integration access is the most common drag: the build is ready before the permissions are. Every system a workflow touches needs credentials, API access and a person authorised to grant them, and a week spent waiting for an external IT contractor to approve a connection is a week on the calendar. Lining up access before the build starts is the cheapest acceleration available.

Document variety is the second. A workflow that reads one consistent format is quick to make reliable; a workflow that reads dozens of supplier layouts and scan qualities needs heavier extraction logic and a longer tuning period. Approval cycles are the third: each person who must sign off wording, and each committee that meets fortnightly, adds elapsed time that no amount of engineering removes.

Regulated firms add a fourth, and it is legitimate. Compliance review of customer-facing wording and a data protection impact assessment signed off before go-live belong in the plan rather than around it, and they are compatible with speed: a UK mortgage and regulated advice firm had its compliance wording approved before release and its intake workflow live in 2 weeks.

Why do the weeks after go-live count as implementation?

Go-live is the midpoint of an implementation rather than the finish. The first weeks of live running are when exception handling is tuned against real traffic and when the error-rate record that justifies wider autonomy is built up. The home-services build above was monitored for its first 30 days before follow-up automation was added, and the bookkeeping firm's exception categories were tuned over the first month of live running.

A plan that includes those monitored weeks is an honest plan. A plan that ends at go-live has moved the same work outside the quote, and the tuning still has to happen; the only question is whether it was priced and owned from the start.

Hold any implementation plan to two dates: the day you first see the system run on your own data, which should sit inside the first fortnight, and the day the first workflow goes live, which for a small business should sit inside the first month. Everything else in the plan exists to serve those two dates.