Automation
What an AI automation agency actually builds, and what it costs.
By Felix Grant4 min read
Most AI automation work is less dramatic than the term suggests. It is moving data between systems without retyping it, drafting routine documents for a person to check, chasing things that need chasing, and producing reports nobody has to build by hand. UK agencies that publish prices typically start at around £750 for an audit and £1,500 for a first workflow. The value is in choosing the right jobs to automate and making sure they keep working.
Builds
Typical builds
The work splits into a few recognisable types, and most useful automations mix two or three. Data moving copies records between systems so nobody retypes them: a new website enquiry goes into the CRM, the right person is assigned and a reply is drafted. Chasing sends reminders on a schedule until something happens, such as overdue invoices chased politely, then escalated to a person.
Drafting uses AI to write a first version for someone to check: proposals, product descriptions, replies to common questions. Sorting and summarising uses AI to read and classify incoming material, such as tagging support emails by topic and urgency. Reporting pulls figures from several systems into one report on a schedule, like a Monday sales and marketing summary that matches Finance. Publishing schedules content with checks and a veto window.
AI agents combine several of these, with the AI deciding some of the steps, for example an assistant that looks up an order, checks stock and drafts a reply for approval. Agents get the most attention. The simpler types usually save the most time.
Example
A worked example: a CRM switch
For one client, a CRM switch had left 2 property listings live. After we worked through it, 29 were live.
This wasn't an AI project, which is the point. The listings depended on data flowing from the CRM to the sites that published them, and that flow broke in the switch. The fix was the unglamorous core of automation work: map every field, check every feed, test with real records, and add an alert so the next break is spotted in hours rather than weeks.
Many AI automation projects start like this. Before anything clever can be added, the basic data has to move correctly between systems.
Audit
What an audit should produce
If you pay for an automation audit, you should come away with more than a list of tools. You should get a list of candidate tasks, each with how often it happens, how long it takes and who does it now, and an estimate of time saved per task in hours per month, based on your numbers rather than an industry average.
You should get a risk note for each task, covering what happens if it goes wrong and whether it touches money, customers or personal data. You should get a recommended order, usually starting with frequent, low-risk tasks, a platform recommendation that fits who will maintain it, and a rough cost for building the first two or three.
If an audit ends with you should use AI agents and no numbers, it wasn't an audit.
An audit without numbers isn't an audit.
Prices
UK price ranges, and ours
A study of 277 UK AI automation agencies by TrustedAIAgencies, reported by IT Brief, found 59% publish no prices. Among those that do, median starting prices were £750 for an audit, £875 for an n8n project, £950 for Make, £975 for Zapier, £1,500 for a workflow automation build, £4,000 for an AI agent or custom build, and £250 for a monthly plan. Starting prices describe the simplest job. Costs rise with the number of systems, the error handling needed, whether personal or financial data is involved, and whether someone has to maintain it.
We publish ours. A Discover workshop is £950: a working session to find the tasks worth automating, with a written plan. A Discover audit is £1,950: a full audit as described above, credited against a build. Build starts at £2,500 for one scoped workflow connecting up to three systems, tested, with failure alerts and a written handover. Partner starts at £1,500 a month for ongoing builds, maintenance and monitoring.
Our Build price sits above the single-workflow medians on purpose, because it includes the testing, alerts and handover that keep an automation working after launch. If you need one simple Zapier workflow, a cheaper provider may suit you better, and we'll say so.
Results
How to judge results
Agree how you will measure an automation before it is built. Time saved is the hours per month the task used to take, minus the time spent checking the automation's output. Error rate is how often that output needs correcting; for AI drafting, track how much a person edits before sending. Speed is how quickly something now happens, such as response time to an enquiry. Reliability is how often it fails and how quickly failures are noticed. Cost is platform fees plus maintenance, set against the time saved.
Review after the first month and again after three, and turn off anything that isn't earning its keep.
Use AI where a task needs reading, writing or judgement, and plain automation for everything else, which is cheaper and more predictable. In our experience, automation mostly removes the repetitive part of people's jobs, so they spend more time on the parts that need them. If you want to know which tasks are worth automating, start with a Discover workshop.
Key takeaways
- Most automation is data moving, chasing, drafting and reporting
- Agents get attention; simple automations usually save the most time
- A useful audit gives tasks, hours saved, risks, order and cost
- UK medians start around £750 for an audit and £1,500 for a build
- Measure time saved, errors, speed, reliability and cost from day one
Written by
Felix Grant
Founder of Kitebox, a business technology and AI consultancy that builds what it recommends. Based in Mallorca, working with UK businesses.
About Felix →