How do agencies justify their setup fees when many AI tools appear to be free or low-cost?

29 May 2026

Why can a setup fee be legitimate when the AI tool itself is cheap?

The honest answer is that the software price and the implementation price are buying different things. A low monthly AI subscription buys access. A setup fee should buy judgement, configuration, integration, risk control and adoption work. Those are not the same product.

Look at the current tool market. Microsoft says Microsoft 365 Copilot Chat is available at no additional cost for eligible Microsoft 365 users, while Microsoft 365 Copilot Business is listed from £13.80 per user per month when paid yearly, excluding VAT, on its UK pricing page. OpenAI says ChatGPT Business is priced at $25 per user per month monthly, or $20 annually, with a two-seat minimum, according to its Business plan help page. Zapier has a free tier and paid automation plans starting from $19.99 per month annually on its pricing page.

Those prices explain why clients challenge agency setup fees. If the tool costs less than lunch for the team, why is the agency asking for £5,000 before anything goes live? The answer should be concrete. The agency should be able to show the hours, deliverables, risks and outcomes behind the setup fee. If it cannot, the client should assume the fee is partly margin protection.

A fair setup fee is normally attached to work such as process mapping, data source review, prompt and workflow design, permissions, integration with tools like HubSpot, Xero, Microsoft 365, Slack or a custom CRM, testing with real examples, staff onboarding, documentation and a handover plan. A weak setup fee is attached to vague language like AI transformation, onboarding, optimisation or proprietary configuration without showing what is actually being done.

What should a fair AI setup fee include?

For a UK SME, a setup fee should be broken down into visible pieces. If you cannot see the pieces, you cannot judge whether the price is fair. A simple implementation might need two to five days of specialist work. A more serious deployment can need several weeks across discovery, build, testing and training.

These ranges are not magic numbers. They are a sanity check. A £1,500 setup fee for a narrow sales follow-up automation may be reasonable. A £20,000 setup fee for turning on a standard chatbot widget with five FAQs is difficult to defend. A £35,000 setup fee for a secure internal assistant connected to SharePoint, CRM records and finance documents may be entirely reasonable if the deliverables are clear and the testing is serious.

The key is that the fee should map to work that survives scrutiny. Ask the agency to show which staff are involved, how many days are allocated, what will be produced, what evidence you will receive, and what happens if the tool does not perform well enough to use.

Why UK businesses should not treat free AI tools as free implementation

There is a real adoption gap in the UK market. DSIT reported in its AI Adoption Research that around 1 in 6 UK businesses, 16%, were currently using at least one AI technology, with a further 5% planning to adopt. The same research found that natural language processing and text generation were used by 85% of AI adopters, while agentic AI was only used by 7%. That matters because most businesses are experimenting with accessible text tools, not deploying controlled AI systems across operations.

The government research also found that limited AI skills and expertise are among the common barriers to adoption, and that ethical concerns, high costs and unclear regulation matter to businesses. In plain English, the hard part is rarely opening the account. The hard part is deciding what staff are allowed to put into the tool, what outputs must be checked, who owns errors, and whether the workflow produces value after the novelty fades.

Cyber security is another reason setup work matters. The UK government Cyber Security Breaches Survey 2025 to 2026 says 21% of businesses reported adopting some AI tools, but among organisations using, adopting or considering AI, only 24% of businesses had security practices or processes in place to manage AI risks. That figure comes from the GOV.UK survey. This is exactly where cheap tools become expensive. A staff member can paste customer data into the wrong service for free. Fixing the policy, permissions, client communication and incident response later is not free.

The ICO is clear that organisations need to apply UK GDPR principles when personal data is used in AI systems. Its AI and data protection guidance includes resources for assessing risks to individual rights and freedoms. A responsible agency setup should therefore include at least a basic data protection review. If personal data, employee data, customer records, special category data or automated decision support is involved, setup is not admin. It is risk management.

What are agencies actually charging for behind the scenes?

A good agency setup fee usually covers work you do not see in the demo. Demos are clean. Real businesses are not. Data is duplicated, CRM fields are inconsistent, staff use different names for the same process, permissions are messy, and edge cases appear immediately once the system touches real customers.

Here is a practical example. A free or low-cost tool can summarise sales calls. Turning that into a working sales process may require call transcript access, CRM field mapping, consent wording, prompt design, lead scoring rules, account owner routing, exception handling, testing against old calls, sales team training, manager dashboards and a way to correct bad outputs. If the agency charges £6,000 to do that properly, the fee may be justified. If it charges £6,000 simply to connect Zoom, OpenAI and HubSpot without testing or handover, it is probably inflated.

Another example is a customer support assistant. The cheap version is a chatbot trained on website pages. The proper version needs a knowledge base review, banned answer list, escalation logic, tone rules, complaint handling, hallucination testing, GDPR review, logging, ownership and a weekly improvement process. The price difference between those two versions is not about the model. It is about operational reliability.

Day rates also explain part of the cost. Experienced UK automation, AI, data, privacy and software specialists are not priced like SaaS subscriptions. A modest specialist day rate might be £500 to £900. Senior implementation, data protection, security or solution architecture time can be £900 to £1,500+ per day. A four-week project with two specialists involved part time can become a five-figure setup fee quickly. That does not automatically make it fair, but it explains why serious implementation cannot be compared with a £20 per user licence.

When is an AI setup fee a red flag?

A setup fee becomes a red flag when it is disconnected from deliverables. The most obvious warning sign is a flat onboarding fee with no scope. If the proposal says £7,500 setup but cannot explain what gets configured, what gets tested, who attends workshops, what documents are delivered, or what happens if the system fails acceptance testing, you are not being shown enough.

Be careful with agencies that resell cheap tools behind a confusing wrapper. There is nothing wrong with building on OpenAI, Microsoft, Google, Anthropic, Zapier, Make, n8n, Airtable, HubSpot or other mainstream platforms. In many cases, that is the sensible route. The issue is when the agency presents a basic configuration as proprietary AI infrastructure and uses that language to justify a high setup fee.

Also watch for setup fees that hide ongoing costs. AI systems often create monthly charges beyond the agency retainer: seats, API usage, vector storage, automation tasks, extra Microsoft licences, monitoring, backup, hosting, support and improvement time. If the setup proposal does not show expected running costs in pounds, ask for them before signing. A £4,000 setup that leads to £800 per month in unplanned subscriptions may be worse than a £7,500 setup with controlled running costs and clear ownership.

The clean test is this: would the agency be comfortable if you showed its setup breakdown to another competent AI implementation specialist? If the answer is yes, the fee is probably defensible. If the answer is no, or if the agency relies on mystery, urgency and jargon, slow down.

How should you challenge the fee without being unfair?

Do not challenge the fee by saying the AI tool is free. That is too blunt and often misses the point. Challenge the fee by asking what work the fee buys. A good agency will welcome that conversation because it separates serious implementation from tool reselling.

Ask five direct questions. First, what exact deliverables are included in setup? Second, what assumptions are you making about our data, systems and staff availability? Third, which parts of the setup are one-off and which will need ongoing support? Fourth, what would be cheaper if we did it ourselves? Fifth, what would you remove if we had to reduce the setup budget by 30%?

The last question is especially useful. If the agency can reduce scope intelligently, it understands the work. For example, it might suggest starting with one department, using manual file upload before a full CRM integration, or delaying a custom dashboard until usage is proven. If it simply says the price is fixed because that is our package, you may be buying a productised service rather than a tailored implementation. That is not automatically bad, but it should be priced and described honestly.

A fair agency should also tell you when not to pay them. If your need is basic staff access to ChatGPT Business, Copilot, Gemini or Claude, you may only need internal guidance and a short training session. If you need a single newsletter drafting workflow, you may be able to build it internally. The setup fee becomes justified when the work affects important business processes, sensitive data, customer experience, regulated obligations, or measurable operational cost.

What is the practical pricing benchmark?

For most UK SMEs, the practical benchmark is this: under £1,000 should buy light configuration or training, not a serious implementation. £2,000 to £5,000 should buy a narrow setup with clear deliverables. £5,000 to £15,000 should buy a proper departmental workflow, including discovery, build, testing and handover. £15,000 to £60,000+ should only appear when there are complex integrations, multiple teams, sensitive data, custom software, governance work or a route to measurable commercial return.

There are exceptions. A specialist agency with a proven reusable product may charge less because the setup is repeatable. A complex regulated use case may cost more because privacy, security, testing and audit work are heavier. But if the fee sits outside those bands, the agency should explain why in plain English.

The best setup fees feel boring when you inspect them. They are not justified by hype. They are justified by a scope table, delivery timeline, acceptance criteria, risk register, training plan, support terms and expected running cost. That is what you are paying for. Not AI magic. Not secret prompts. Not access to a model you could buy yourself.

If you want a simple rule, use this one: the more the AI touches live data, customers, money, staff decisions or regulated processes, the more setup work is justified. The more it looks like a generic content helper or personal productivity tool, the harder a large setup fee is to defend.

Is This Right For You?

This applies if you are buying an AI implementation, automation build, chatbot, internal assistant, reporting workflow, document processing system, or agent that will touch real business data or staff workflows. It is especially relevant if the agency is asking for a setup fee before a monthly retainer or software subscription.

It does not apply in the same way if you only need a simple ChatGPT Business workspace, a Microsoft 365 Copilot licence, or a single two-step Zapier automation with no sensitive data and no integration risk. In those cases, a large setup fee should be questioned hard.

Frequently Asked Questions

Is a £5,000 AI setup fee reasonable for a small UK business?

It can be reasonable if it includes discovery, configuration, testing, data protection checks, staff training and documentation for a real workflow. It is high if the agency is only setting up accounts, connecting one simple automation or providing generic prompt templates.

Should an agency show me the tool costs separately from its setup fee?

Yes. You should see software subscriptions, API usage, automation task costs, hosting, support and agency labour separately. Bundled pricing is not automatically wrong, but hidden tool costs make it impossible to compare options fairly.

Can I avoid setup fees by using ChatGPT, Copilot or Zapier myself?

Sometimes. If the use case is personal productivity, simple content drafting or a basic two-step automation, DIY may be sensible. If the workflow uses customer data, connects business systems, affects service quality or needs adoption across a team, you should budget for implementation work.

What deliverables should I expect from an AI setup project?

Expect a scoped workflow, configured tools, documented prompts or logic, data handling notes, user roles, test cases, acceptance criteria, training materials, known limitations and a handover plan. For higher-risk work, expect a DPIA input, security review and incident escalation route.

Is it fair for an agency to charge setup and then a monthly retainer?

Yes, if the two fees cover different work. Setup should cover initial discovery, build and launch. The retainer should cover monitoring, fixes, optimisation, reporting, support and improvements. If the retainer only keeps the system switched on, ask what platform cost is being marked up.

What is the biggest hidden cost in AI implementation?

The biggest hidden cost is usually staff time and process change. Workshops, testing, feedback, training, governance decisions and adoption all take time. Tool subscriptions are visible. Internal disruption is often under-budgeted.

When should I refuse to pay an AI agency setup fee?

Refuse or renegotiate when the agency cannot explain deliverables, will not separate tool costs from service costs, avoids data protection questions, gives no testing plan, or describes a standard third-party tool as proprietary without evidence.

How can I compare two agency setup fees fairly?

Compare scope, not just price. Look at the number of workflows, integrations, data sources, users, security requirements, testing depth, training, documentation and support. A cheaper setup can be more expensive if it leaves you with an unreliable system.