Arkwright explainer

What does AI automation cost in the UK?

UK AI automation is priced by workflow risk, integration depth, data readiness, review requirements, and the maintenance model. The model bill is usually the smallest line.

Audience
Operators sizing a first build
Price range
£3k pilot to £100k+ operating system
Last reviewed
May 2026

Decision guide

Budget planning

Use this when you need a realistic cost range before asking suppliers for proposals.

The practical question

What should an AI automation project cost once reliability, integrations, handover, and support are included?

Use this when

  • You need a board-safe range before committing to discovery.
  • You have seen wildly different quotes and need to know what is missing.
  • You want to separate prototype cost from production cost.

How to read it

  1. 01Start with the budget bands.
  2. 02Check which assumptions move the range.
  3. 03Use the final section to decide whether a costed workflow scope is worth it.

What to weigh up

You will leave with

  • A realistic budget range for common UK SME automation work.
  • The difference between a prototype, a usable workflow, and a production system.
  • The quote drivers that matter more than model usage fees.

Pricing method

  • Use directional UK SME bands, not a supplier-wide market survey.
  • Price around workflow scope, integrations, data readiness, governance, testing, handover, and maintenance.

Cautions

  • Proof-of-concept pricing is not production pricing.
  • Maintenance and exception handling are often omitted from cheap quotes.

Three honest tiers

Where a UK SME build typically lands.

The bands below assume one or two integrations, defensible governance, and a real team that has to live with the result.

£3k - £10k

Focused pilot

Prove operational value on one contained workflow.

What is included

  • One workflow, one or two systems
  • Limited data cleanup
  • Human review step retained
  • Decision-grade evidence pack

Where this stops working

You expect it to scale across teams, write back to systems autonomously, or replace headcount.

Most common

£10k - £35k

Production workflow

A working automation a real team uses every day.

What is included

  • Integrations with permissions
  • Error handling and logging
  • Accuracy checks and monitoring
  • Handover documentation

Where this stops working

You need to govern multiple workflows together, run sensitive permissions, or coordinate across departments.

£35k - £100k+

Operating system

Multiple workflows joined up into one system.

What is included

  • Deeper CRM / ERP integration
  • Automated accuracy checks and monitoring
  • Permission and approval policies
  • Governance and audit posture

Where this stops working

You have no internal owner ready to take handover, or no appetite for a continuous review cadence.

Worked example

A £14k document-intake automation

A mortgage broker receiving 200+ supporting documents a week across email and a vendor portal. The team was re-keying into the CRM and flagging exceptions by hand.

Input

Mixed-format documents (PDF, image, email body) and a CRM with structured deal records.

Process

  1. 01Discovery (2 weeks). Mapped 8 document types, captured 60 real samples, defined acceptance criteria with the operations lead.
  2. 02Build (3 weeks). Extraction step, exception queue, approved updates written back to the CRM, audit log on every action.
  3. 03Testing + handover (1 week). A set of accuracy checks, manager-review training, and a runbook for the support team.
  4. 04Year-1 maintenance. ~£2.5k annual budget for prompt updates, new document type support, monitoring review.

Output

Manager sees a daily exception queue and a clean stream of approved updates into the CRM. Team time on intake drops materially.

What to watch

A quote that omits the exception queue, the audit log, or the maintenance line — those are the parts that decide whether the system lasts.

Price drivers

What changes the number up or down

The line items inside a quote rarely move uniformly. These are the levers we see most often when scope shifts during discovery.

  • Factor

    Workflow complexity

    Pushes up

    Many branching paths, regulated outputs, irreversible actions.

    Pulls down

    A single happy path with a human approval gate.

  • Factor

    Integration depth

    Pushes up

    Bespoke connections, legacy systems, writing sensitive data back into core systems.

    Pulls down

    Common SaaS that is straightforward to connect to.

  • Factor

    Data quality

    Pushes up

    Fragmented sources, inconsistent labelling, manual exports.

    Pulls down

    One clean source of truth and recent housekeeping.

  • Factor

    Governance posture

    Pushes up

    Regulated industry, audit requirements, multi-stakeholder sign-off.

    Pulls down

    A single accountable owner with delegated authority.

  • Factor

    Change rate

    Pushes up

    Rules that change monthly or seasonally.

    Pulls down

    A stable process with clear ownership.

Read before deciding

Caveats

  • Ranges vary by supplier, risk posture, urgency, and the level of operational support required.
  • Do not treat a proof-of-concept quote as a production quote unless support, monitoring, and handover are included.
  • Cheap automation becomes expensive when no one owns it after launch.
  • Add a maintenance line to every quote you compare — it is the line most quotes leave off.

FAQ

Common questions

Can you quote before discovery?
Only as a range. A useful quote needs workflow scope, system access, data quality, review requirements, and risk boundaries. Anything tighter is a guess.
Is SaaS cheaper than custom?
Often, for standard workflows. Custom is cheaper over a multi-year horizon when your process, data, or governance needs do not fit the SaaS shape.
What is a realistic ongoing maintenance figure?
For a production workflow, plan 15-25% of build cost as an annual running budget — covering monitoring, accuracy checks, prompt or model updates, changes when a supplier updates their tools, and small adjustments. Critical workflows sit higher.
Why is "AI cost" mostly not the model bill?
Model spend is usually a small line in the total. Discovery, integration, testing, governance, change management, and ongoing ownership dominate the bill for any workflow that has to hold up in operations.

Arkwright next step

Get a defensible range, not a number plucked from the air.

Bring one workflow you would like priced. We will give you a band, the assumptions behind it, and the questions that move the number up or down.