Direct answer
In the UK in 2026, published prices put an AI audit at about £750–£8,000 depending on firm size (Lilach Bullock), a first single-workflow automation at £1,500–£5,000, and multi-system builds with AI agents at £20,000–£50,000 or more (both AI Agency Plus, a vendor's estimate of the market). Ongoing support runs from about £200 a month (Generativ) to £3,000 a month (AI Agency Plus, vendor data). AutomateFlow charges £3,500 for the Audit, £15,000 for a Pilot, £35,000 for a Build and £2,000 a month for Run, all excluding VAT.
Where your project lands within those bands depends on how many paths, systems and exceptions the work has, and on who looks after it afterwards. Below are the published prices with their sources, what an AI agent costs to build and to run, what bespoke software costs, and a worksheet for comparing proposals.
Who it is for and when it does not apply
Use this guide to budget for AI implementation, prepare a scoping conversation and compare proposals. It is not an ROI calculator, a quote, a guaranteed timeline or a promise of savings. If the process has no owner, verifiable outcome, or lawful system access, clarification is the first stage rather than estimating a build.
What one workflow costs
Most firms start with one workflow, and UK providers price that first step in one of three ways: a short discovery, a fixed-scope automation, or a proof of concept that runs in production. The figures below are the providers' own; where a figure is a provider's estimate of the market rather than its own price list, the source column says so.
Swipe or scroll to compare the columns.
| What you're buying | Published UK range | Typical time | Source |
|---|---|---|---|
| AI audit, firm under 10 staff | £750–£2,500 | — | Lilach Bullock, 2026 |
| AI audit, 10–50 staff | £2,500–£8,000 | — | Lilach Bullock, 2026 |
| Discovery, often credited against the build | £500–£2,000 | 1–2 weeks | AI Agency Plus, 2026 (market estimate, vendor data) |
| First single-workflow automation | £1,500–£5,000 | 1–3 weeks | AI Agency Plus, 2026 (market estimate, vendor data) |
| London SME automation build, after a £4,000–£10,000 audit | £12,000–£40,000 | — | Kolonell, 2026 |
| Multi-system programme with AI agents | £20,000–£50,000+ | 3–6 months | AI Agency Plus, 2026 (market estimate, vendor data) |
| AutomateFlow Pilot: one workflow in production, up to three systems | £15,000 excl. VAT | Typically 4–6 weeks | Solutions |
Our Pilot sits above the single-workflow band because it includes exception handling, an audit trail, documentation, a handover and a 30-day defect period across up to three systems. If your workflow is simple enough for a £1,500 setup, or for a setting in a tool you already pay for, that is the better buy, and we'll tell you so on the call.
Published audit, pilot and build prices in the UK (October 2026)
We checked these prices on providers' own pages on 23 October 2026. They are "from" prices or stated ranges, not quotes, and providers define scope differently, so compare what each one includes. AutomateFlow is listed with the same treatment.
Swipe or scroll to compare the columns.
| Provider | Offer | Published price | VAT | Source |
|---|---|---|---|---|
| M22 | AI audit | From £1,500 | Not stated | m22.group |
| AutomateFlow | Audit (one workflow, up to five systems) | £3,500 | Excluded | Solutions |
| Generativ | Two-week discovery | £3,500 | Excluded | generativ.co.uk |
| The AI Consultancy | Readiness sprint | From £3,500 | Excluded | theaiconsultancy.ai |
| SpotDev | AI and data readiness assessment | From £5,000 (20% credited) | Not stated | spotdev.co.uk |
| OpenKit | AI audit | £10,000 | Excluded | openkit.co.uk |
| Generativ | AI builds | £3,000–£15,000+ | Excluded | generativ.co.uk |
| SpotDev | AI foundations | £8,000 | Not stated | spotdev.co.uk |
| AutomateFlow | Pilot (one workflow in production) | £15,000 | Excluded | Solutions |
| The AI Consultancy | Discovery and pilot | From £15,000 | Excluded | theaiconsultancy.ai |
| SpotDev | Custom agents | £20,000 | Not stated | spotdev.co.uk |
| AutomateFlow | Build (one operational app) | £35,000 | Excluded | Solutions |
| The AI Consultancy | Build and embed | From £40,000 | Excluded | theaiconsultancy.ai |
| Generativ | Maintenance | £200–£500 per month | Excluded | generativ.co.uk |
| AutomateFlow | Run (up to 12 hours) | £2,000 per month | Excluded | Solutions |
| The AI Consultancy | Fractional AI officer | From £3,000 per month | Excluded | theaiconsultancy.ai |
| OpenKit | Implementation | From £5,000 per month | Excluded | openkit.co.uk |
Read the table as a map of entry points, not a ranking. A £3,500 audit that covers one workflow and a £10,000 audit that covers a whole department are different products. Before comparing numbers, check how many workflows, systems and user roles are included, whether discovery and support are separate, and whether VAT is added. Prices change; check the source page before you rely on a figure.
What an AI agent costs to build and to run
An AI agent has two costs: building it once, and running it every month. The build is priced like any other workflow above; for one agent on one task in production, our Pilot is £15,000 excl. VAT. Running costs are separate, depend on volume, and are often missing from quotes.
Swipe or scroll to compare the columns.
| Running cost | What drives it | Who you pay |
|---|---|---|
| Model usage | How many messages or documents, how long they are, and which model | The AI provider, usually billed by usage |
| Channel fees | Messages on the WhatsApp Business Platform, charged per message since July 2025 (from October 2026 service replies are free only up to a monthly allowance, according to Meta's pricing page); call minutes for voice | Meta, or your phone provider |
| Hosting and workflow tools | Servers, databases, n8n or similar | Your hosting provider |
| Monitoring and fixes | Someone reviewing corrections, failures and drift | Your team, or a support plan (our Run: £2,000 a month, up to 12 hours) |
Some vendors now price customer-service agents per outcome: Intercom lists its Fin agent at $0.99 per outcome, such as a resolved conversation, and Salesforce lists Agentforce at $2 per conversation (vendor list prices, 2026). Whichever model you choose, ask the supplier to estimate monthly running costs on your real volumes before you sign.
What an AI receptionist costs
On 24 October 2026, Fasthosts listed its AI Receptionist at £39 a month for 30 calls (then 49p a call), £69 for 100 calls (then 39p, with appointment scheduling) and £99 for unlimited calls, all excluding VAT. A human answering service started at £14.99 a month plus VAT with Connect Communications, and MBE charged £47.94 a month plus VAT for 20 calls. None of them books the job unless it's connected to your calendar and CRM. Answering service, AI receptionist or missed-call text-back compares the options at your call volume and covers what happens after the call.
What bespoke software costs
A focused operational app is a bigger job than an automation, because people work in it every day. London agencies quote SME builds at £12,000–£40,000 after an audit (Kolonell, 2026), and agencies pricing in US dollars put an AI-enabled first version at $30,000–$60,000 over 8–12 weeks (Empat) or $35,000–$80,000 over 8–14 weeks (Bitsens), both vendor data. Our Build is £35,000 excl. VAT for one responsive web app with up to three user roles, one core workflow, up to two integrations, basic reporting and an audit trail, typically over eight to twelve weeks.
What moves the price: the number of user roles, each integration, approval and audit requirements, data you need to migrate, and native mobile apps, which our Build doesn't include.
What shapes scope
Swipe or scroll to compare the columns.
| Dimension | Scoping question | Why it can change effort |
|---|---|---|
| Process | How many paths, roles, and exceptions exist? | More branches require design and testing |
| Data | Which fields, documents, and history are needed? | Quality and transformation can dominate the work |
| Integrations | Which systems read or write, and under what contract? | APIs, limits, and conflicts need validation |
| Interface | Is a job or form enough, or is a queue needed? | State and user control require a work surface |
| Authority | What can run automatically, and what does a person approve? | Boundaries and escalation change architecture |
| Safety | What access, retention, audit, and recovery are required? | Risk requires controls and review |
| Operations | Who monitors and changes it later? | Handover and support continue after release |
AutomateFlow estimation model
The model below shows what needs to be estimated. It does not produce a price or timeline on its own.
Swipe or scroll to compare the columns.
| Delivery component | Expected output | What can increase effort |
|---|---|---|
| Discovery | Process map, owners, scope, and acceptance criteria | Unobserved work, conflicting rules, or undefined exceptions |
| Solution design | System of record, data contracts, states, permissions, and safe stops | Multiple source systems, roles, or audit requirements |
| Build and integrations | Automation, connectors, and the approved interface | Unstable APIs, complex transformations, or custom queues and approvals |
| Verification and release | Tested scenarios, reconciliation, alerts, and recovery | Hard-to-reproduce data, important consequences, or multiple environments |
| Operations and support | Handover, monitoring, maintenance, and a change procedure | Extended availability, multiple vendors, or frequent changes |
Complexity signals
Use these levels as a shared scoping language, not as an automatic price multiplier.
Swipe or scroll to compare the columns.
| Dimension | Low complexity | Medium complexity | High complexity |
|---|---|---|---|
| Process | One linear path, one owner, few exceptions | Several branches and roles | Many states, approvals, and context-dependent exceptions |
| Data and integrations | One defined source and controlled export | Two or more systems with available contracts | Competing sources, history to migrate, or unstable interfaces |
| Interface and authority | Job or form with reversible writes | Queue, correction, and basic approval | Custom roles, sensitive actions, or decisions that are hard to reverse |
| Operations | Limited schedule and one owner | Alerts, resumption, and planned maintenance | Critical continuity, multiple vendors, or special obligations |
A project can combine different levels. The process may be simple while data access requires additional controls. The estimate should explain those differences instead of assigning one overall label.
Method and definitions
- Discovery: observe the current path and separate repeatable steps from interpretation and approval.
- Narrow scope: choose one input, one output, and a bounded exception set for the first release.
- Blueprint: define the system of record, integrations, states, permissions, and failure behaviour.
- Build and test: check the normal path, missing values, duplicates, slow responses, stopping, and resumption.
- Controlled release: hand over access, logs, alerts, correction procedure, and expansion criteria.
“Cost” means the resources required for discovery, design, implementation, integration, validation, and operation, not only the time to write a script. “Implementation” means a usable and operable system, not a demo that works only on a chosen file.
Reusable scoping worksheet
Complete this worksheet before requesting a proposal, using synthetic examples:
- Accepted operational outcome: ____________________
- Owner and approval responsibility: ____________________
- Input event: ____________________
- System of record for each field: ____________________
- Systems to read/write and available contracts: ____________________
- Known deterministic rules: ____________________
- Interpretation steps that might use bounded AI: ____________________
- Actions forbidden to the agent or automation: ____________________
- Exceptions, duplicates, and missing data: ____________________
- Required permissions, retention, audit, and restoration: ____________________
- Acceptance criterion and test scenarios: ____________________
- Owner for alerts, maintenance, and stopping: ____________________
- Criterion for expanding the first release: ____________________
How to compare proposals
- Ask each provider to use the same definition of scope, deliverables, and exclusions.
- Check whether discovery, testing, release, handover, and support are included or estimated separately.
- Compare assumptions about data, access, APIs, volume, availability, and exception ownership.
- Ask for acceptance criteria, a change procedure, and conditions for expanding the first release.
- Separate initial delivery cost from operations, maintenance, and later changes.
The workflow should justify the type of build. Automation or operational application? provides the full framework for choosing among configuration, integration, automation, bounded AI, an application, and human control.
Evidence
A proposal can describe an architecture but cannot prove a result, ROI, duration, or savings. Compare any figure only together with its assumptions, definitions, period, and source. AutomateFlow’s public materials separate service descriptions, blueprints, and published implementations from measured results.
Limitations
This page is not a quote, guaranteed timeline or ROI calculation. Prices in the provider table were checked on the providers' own pages on 23 October 2026. The ranges in the workflow, AI agent and bespoke software sections were checked against the named sources on 24 October 2026; several are a provider's estimate of the market rather than its own price list and are marked as vendor data. We dropped Goodspeed's proof-of-concept, sprint and support figures because its page was no longer available on that date. Receptionist prices are from the vendors named in the answering-service guide, read the same day. Any of them may have changed since. Scope, data quality, integration availability, and operating ownership can materially change a proposal; these must be clarified for the real workflow.
Related AutomateFlow pages
- AI audit and AI strategy
- AI implementation prices and engagement models
- How to choose an AI consultancy in the UK
- Answering service, AI receptionist or missed-call text-back: what each costs
- Xero integrations: native app, Zapier, Make or a custom build?
- AI information
- How AutomateFlow works
- About AutomateFlow
