The Short Answer
What you pay an automation agency is decided by three things: how many processes you are automating, how messy the inputs are, and how many systems each one has to touch. One well-defined process using clean data from two mainstream apps is the cheapest thing we build. A connected set sharing one data model costs more but less per workflow. A staged rollout across a whole team, with approvals and an audit trail, is a different order of work again.
The awkward part is that the software underneath costs almost nothing. Zapier and Make both publish a free tier and paid plans that a small business can afford without thinking about it. So the honest question is not what automation costs. It is what the difference between a builder subscription and an agency fee actually buys.
This guide answers that, and tells you when the answer is "not enough, do it yourself". We do not publish prices, because a number set without seeing your processes is a guess. We quote a fixed price after a free discovery call.
What the Tools Actually Cost
Worth knowing before any agency quotes you, because it is the floor under every proposal.
Zapier meters by task, where a task is counted when it successfully completes a unit of work and failed actions are not counted. Its free plan covers 100 tasks a month, and paid plans start at 19.99 dollars a month billed annually for 750 tasks, scaling by task tier from there. Make meters by credit, where each module action such as adding a spreadsheet row counts as one, with 1,000 credits free and Core at 9 dollars a month for 10,000. Both publish in dollars, so treat sterling as approximate.
For a small business running a handful of simple, reliable automations, that genuinely may be all you need. Nobody should pay an agency to connect a form to a spreadsheet.
So What Is the Fee For
Five things, and they are worth different amounts to different businesses.
Working out what to automate. This is the part most people underestimate and it is where the money is actually saved. The process you think is worth automating is frequently not the one that is. Volume, how rule-bound it is, and how much it costs when it goes wrong matter more than how annoying it feels.
The parts that are not drag-and-drop. Connecting two mainstream apps is easy. Reading a supplier's inconsistent PDF invoice, matching a name that is spelled three ways across two systems, or talking to a platform whose interface is barely documented is not. That is where days go.
Deciding what happens when it goes wrong. A workflow that runs 500 times a month will fail. The engineering is in what happens then: does it retry, park the item for a human, or alert someone. Automations built without this quietly lose things, and nobody notices for weeks.
The human checkpoint. Anything touching money, a filing, or a customer commitment needs a person approving before it completes. Designing where that checkpoint sits, so it catches errors without recreating the manual work, is judgement rather than configuration.
Somebody owning it afterwards. Tools change their interfaces, your prices change, staff leave. Ongoing support is not a subscription for nothing; it is the reason the thing still works in a year.
The Comparison
| Do it yourself | Freelancer | Agency | |
|---|---|---|---|
| What you pay for | Tool subscription only | Someone's time to build it | A built process plus its upkeep |
| Time to first result | Days, if you have the time | 1–2 weeks | 1–3 weeks |
| Handles messy inputs | Rarely | Sometimes | Yes |
| Error handling designed | Almost never | Varies | Yes |
| Still working in a year | If you maintain it | If they are still around | Under a support plan |
| Best for | Simple links between mainstream apps | One well-defined job | Several processes, or anything touching money |
When to Skip the Agency
We build these for a living and there are plenty of cases where we would tell you not to bother.
The workflow is two mainstream apps and a rule. Form to spreadsheet, order to Slack, booking to calendar. Both Zapier and Make do that on their free tiers.
You have somebody internally who enjoys this. Every business has one person who likes tinkering with systems. Give them a few hours a month and a paid plan, and let them do it.
The process is about to change. Automating something you are already planning to rework is paying twice.
You cannot say what "working" looks like. If nobody can state what the automation should produce or how you would know it succeeded, the specification does not exist yet and no amount of money fixes that.
When It Is Worth Paying For
The input is messy. Documents, emails, scanned paperwork, anything a human currently reads and interprets. This is where AI earns its place over plain automation, and where a builder tool alone will not get you there.
Getting it wrong costs real money. Anything touching invoices, payroll, filings or client commitments. The value is in the controls, not the speed.
It crosses several systems. Two apps is a Zap. Five systems with data that has to agree across all of them is a project.
The volume is genuinely high. A process running hundreds of times a month justifies engineering that one running twice does not.
How These Projects Actually Go Wrong
Almost never through the technology. The three failures we see repeatedly are all decisions made before anyone writes anything.
Automating the loudest process instead of the biggest. The task everyone complains about is often low volume and full of exceptions. The task nobody mentions, because it has always been done that way, runs four hundred times a month to a fixed rule. Ask what people do most, not what they hate most.
Nobody owning it after launch. An automation is not furniture. Prices change, a supplier alters their invoice layout, a tool updates its interface. Without someone whose job it is to notice, it degrades quietly, and the business concludes automation does not work when what happened is that nobody maintained it.
Removing the human too early. The temptation after a good month is to switch off the approval step, because it feels like the last bit of manual work. That step is what catches the case the rules never anticipated. Keep it wherever money or a commitment to a customer is involved, and make it fast rather than absent.
None of these cost extra to avoid. They are decisions, and getting them right is most of what separates automation that lasts from automation that gets quietly turned off.
Reading a Quote Properly
Ask these and the differences between proposals become obvious.
- 1. Which specific processes, named, and why those first?
- 2. What happens when a step fails, and who finds out?
- 3. Where does a human approve, and what does that person see?
- 4. Which tool is this built on, and who pays that bill?
- 5. What are the tool costs at our actual volume, not the starting tier?
- 6. Who owns the workflows if we part ways?
- 7. What is included in the monthly fee, and what is billed extra?
- 8. What does month two look like if nothing breaks?
Question four matters more than it looks. Some agencies build on their own account, so leaving means losing the automation entirely. Ours are built on yours.
Payback, Honestly
The usual sales line is hours saved times an hourly rate, which overstates it, because the hours saved are rarely a whole person's worth and rarely convert into either fewer salaries or more sales on their own.
The better test is narrower. Take one process, count how many times it runs a month and how long each takes, and be honest about whether the time freed goes into something that earns money or just evaporates. A first automation needs to save a few hours a week to make sense within a year, which is achievable for a genuinely repetitive process and not achievable for a nice-to-have. Bring that number to the call and we can tell you quickly whether the build will clear it.
Sector detail for practices where this arithmetic works well is at /industries/accountants, and the same reasoning applied to specific workflows is at /blog/ai-automation-for-accountants-uk.
How We Price It
Fixed price after a free discovery call, never a day rate and no published figure to anchor you against someone else's situation. The scope levels are on /services/ai-automation: a single workflow end to end, a bundle of three to five sharing one data model, or a staged rollout across teams. Support is a separate monthly fee. Tool fees are yours and paid directly to Zapier, Make or whatever it runs on, so you can see them rather than find them marked up inside a retainer.
On the call we will tell you when a builder tool on your own account does the job. For how automation fits alongside the rest, see /blog/ai-development-cost-uk-2025.
Want a fixed number for your own processes? Book a free discovery call at /contact.
Frequently Asked Questions
Why pay an agency when Zapier costs a few pounds a month?
For a simple link between two mainstream apps, you should not, and we will say so. The fee covers messy inputs a builder cannot parse, error handling for when a workflow running hundreds of times a month fails, a human approval step where money is involved, and somebody keeping it working when tools and prices change.
What are the ongoing costs after the build?
Two separate things. Support from us is a monthly fee agreed with the build, scaled to how much we are watching. Tool fees are paid by you directly to Zapier, Make or whatever the workflow runs on, which keeps them visible rather than marked up inside a retainer.
Do we own the automations?
Yes. They are built on your accounts, not ours, so if we part ways everything keeps running and you keep the logins. Worth asking any agency this before signing, because it is not universal.
How do we know which process to automate first?
Look for high volume, clear rules, and a low cost of being occasionally wrong. Chasing documents and rekeying data usually qualify; anything needing judgement usually does not. Picking the wrong first process is the most common reason automation disappoints.
How long before it pays for itself?
A first workflow needs to save a few hours a week to make sense inside a year. That is realistic for a genuinely repetitive process and not realistic for a nice-to-have, and the honest check is whether the freed time goes into work that earns money rather than simply evaporating.