Where the money goes
A quote should show the work behind the number. For one clearly bounded workflow, I would normally break the scope into these parts.
| Work | Typical scope | Days | Cost at £450 |
|---|---|---|---|
| Process mapping | Inputs, decisions, exceptions, owners and acceptance tests. | 1 to 2 | £450 to £900 |
| Build and integration | Prompts, rules, connections, logging and failure handling. | 3 to 8 | £1,350 to £3,600 |
| Handover | Testing record, operating guide and team training. | 1 to 2 | £450 to £900 |
| Total planning range | One useful workflow with a named owner. | 5 to 12 | £2,250 to £5,400 |
The range assumes the process already exists and somebody can explain it. If nobody owns the process, discovery becomes a separate job. My guide to which business processes AI can automate includes the four tests I use before pricing.
The six things that change the price
Process clarity
A written process with known exceptions is cheaper to automate than work held in three people’s heads.
Data condition
Clean tables and consistent documents are easier than scanned PDFs, missing fields and conflicting records.
Integrations
Each CRM, inbox, spreadsheet, database or private API adds access, testing and failure cases.
Risk
A private internal draft needs fewer controls than an action that changes customer data or sends a public answer.
Exceptions
The normal path is usually easy. The price sits in the unusual cases and what the system must do when confidence is low.
Support
A clean handover costs less than a managed service with monitoring, updates and response cover.
Costs that sit outside the build
The consultancy fee is only one line. Ask for the operating costs before you approve the project.
- Model usage. The bill depends on the model, prompt size, output size and number of runs.
- Automation software. n8n, Make, Zapier or another orchestration tool may carry a monthly charge.
- Hosting and storage. Custom code, logs and document stores need somewhere to run.
- Monitoring. Someone must know when an API changes, a run fails or costs jump.
- Internal review. Human approval is still work. Budget the owner’s time honestly.
A workflow that touches personal data needs more than a clever prompt. The ICO AI risk toolkit is a useful starting point. The NCSC secure AI guidance also puts security across design, development, deployment and operation.
Day rate or fixed price?
Use a day rate while the problem is still being defined. Use a fixed price once the workflow is clear.
A fixed quote should name the input, output, systems, review point, failure behaviour and acceptance test. Without those six details, the fixed price only hides the uncertainty.
Hourly rates, retainers and ROI
Hourly fees are hard to compare when discovery and implementation overlap. A day rate makes the working block visible. A fixed fee makes sense after the scope is stable. A retainer is better reserved for ongoing support, monitoring and a planned queue of improvements.
Calculate likely ROI before approving the build. Use handling hours saved, avoidable errors and faster response times. Do not count vague productivity gains that nobody will measure.
Three quote warnings
- No discovery. The supplier prices before seeing the real process or source data.
- No owner. Nobody on your team is responsible for checking outputs and approving changes.
- No handover. The quote ends at deployment and leaves documentation, access and support unclear.
If your scope is still unclear, start with the consultant versus agency guide. It explains which delivery model fits a small build, a multi-team programme or a permanent internal capability.
How these figures were calculated
The figures are arithmetic based on my published £450 day rate. They describe a bounded single-workflow scope. They are not claimed as a UK market average.
