The cost question nobody answers clearly

Ask an AI consultancy how much an agent costs and you will get "it depends." That is technically true but completely unhelpful. Here is an actual breakdown based on what we charge and what the underlying infrastructure costs. No vague ranges. No "contact us for pricing" evasions. Real numbers.

The four cost components

Every AI agent has four cost layers: development, infrastructure, API usage, and maintenance. Most quotes you see online only cover development. The other three are where surprises live.

Build costs: the one-off investment

Building a custom AI agent typically involves four phases. Here is what each costs:

1. Scoping and architecture (1-2 days)

Defining what the agent does, what it connects to, and where the boundaries are. This includes prompt engineering, tool selection, and integration mapping. We also identify which AI model fits the task - using the most powerful model for everything is a common and expensive mistake.

Cost: £500-£1,000

2. Core development (1-3 weeks)

Building the agent logic, integrating with your systems, and implementing the conversation flows or automation sequences. This is the bulk of the work and where the cost range widens most.

Cost: £2,000-£8,000

The range depends on complexity. A customer support agent that answers FAQs from a knowledge base is on the lower end. A multi-step operations agent that processes invoices, updates your CRM, and generates reports is on the higher end.

3. Testing and hardening (3-5 days)

Testing across edge cases, adversarial inputs, and failure scenarios. Building error handling, fallback logic, and monitoring. This phase catches the problems that would cost far more to fix in production.

Cost: £500-£1,500

4. Deployment and monitoring (1-2 days)

Deploying to production, setting up logging and alerting, configuring the monitoring dashboard, and documenting the system for your team.

Cost: £500-£1,000

Total build cost: £3,500-£11,500

Most agents we build fall in the £3,500-£7,000 range. The expensive builds involve multiple system integrations, complex decision logic, or regulatory compliance requirements.

Running costs: the monthly expense

Once your agent is live, there are ongoing costs. This is where many businesses get caught off guard, so let us be specific.

AI model API costs

This is the biggest variable. It depends on the model you use, how much text the agent processes per interaction, and your total volume.

Cost per 1,000 interactions by model:

Claude Sonnet (most common for business agents): £1-£3
GPT-4o: £0.50-£2
Claude Haiku or GPT-4o-mini (for simpler classification tasks): £0.05-£0.20

Monthly cost by agent type:

Support agent handling 500 tickets/month: £5-£50
Content agent generating 100 articles/month: £20-£100
Research agent running daily monitoring across 50 sources: £30-£150
Lead qualification agent processing 200 leads/month: £10-£40
Email triage agent handling 3,000 emails/month: £15-£60

The key insight here is that API costs are almost always trivially low compared to the human labour they replace. A support agent costing £50/month in API fees replaces work that would cost £2,000+ in staff time.

Infrastructure costs

These are the servers, databases, and tools that keep the agent running.

Hosting (Vercel, AWS, or similar): £0-£20/month for most agents
Database (Supabase, PostgreSQL): £0-£25/month
Monitoring and logging tools: £0-£20/month
Vector database for knowledge retrieval (if needed): £0-£30/month

Total infrastructure: £0-£95/month, with most agents under £50.

Maintenance costs

This is the cost most quotes omit. Agents are not set-and-forget systems. They need:

Prompt tuning as edge cases emerge (2-4 hours/quarter): £100-£200
Model updates when providers release new versions (1-2 hours/quarter): £50-£100
Integration maintenance when connected systems change their APIs (varies): £0-£200/quarter
Performance reviews to ensure accuracy has not drifted (1-2 hours/month): £50-£100

Total maintenance: £50-£200/month on average.

Total monthly running cost

For a typical business agent: £70-£400/month. Most fall between £100-£250.

Pricing models: how agencies charge

When you hire a team to build your agent, the pricing structure matters as much as the total number. Here are the three models you will encounter:

Fixed price per project

You agree on a scope and price upfront. The agency delivers the agent for that price regardless of how long it takes them. This is how we work at Bloodstone - it protects you from scope creep and gives you certainty.

Best for: Most businesses. You know what you are paying before work starts.

Time and materials (hourly/daily rate)

You pay for the hours spent. Rates for AI development in the UK typically range from £80-£200/hour depending on the agency.

Best for: Exploratory projects where the scope is genuinely unclear. But be careful - without a cap, costs can escalate quickly.

Retainer with usage-based pricing

A monthly fee for ongoing development and maintenance, plus variable costs based on agent usage (API calls, interactions processed).

Best for: Businesses that need continuous development and iteration on their agents, not just a one-off build.

Real cost examples

Let us look at three actual agent types and their full cost of ownership over 12 months.

Example 1: Customer support agent

What it does: Answers common questions from a knowledge base, collects information for complex issues, routes to the right team member, follows up on unresolved tickets.

Build cost: £5,000
Monthly API costs: £30-£80
Monthly infrastructure: £25
Monthly maintenance: £75
Year 1 total: £6,560-£7,160

Alternative: Junior support hire at £26,000/year plus £3,000 recruitment costs. The agent handles the same volume at roughly a quarter of the cost, works 24/7, and never calls in sick.

Example 2: Content generation agent

What it does: Takes a brief and generates long-form articles, social posts, and email copy. Pulls in research from specified sources. Outputs in your brand voice.

Build cost: £4,000
Monthly API costs: £50-£100
Monthly infrastructure: £15
Monthly maintenance: £100
Year 1 total: £5,980-£6,580

Alternative: Freelance content writer at £1,500-£3,000/month producing 10-20 pieces. The agent produces 50-100 pieces for a fraction of the cost, though human editing is still needed.

Example 3: Operations agent

What it does: Processes invoices from email, extracts line items, matches against purchase orders, flags discrepancies, updates the accounting system, and notifies the relevant team member.

Build cost: £8,000
Monthly API costs: £40-£60
Monthly infrastructure: £30
Monthly maintenance: £150
Year 1 total: £10,640-£10,880

Alternative: Part-time bookkeeper at £800-£1,200/month. The agent is slightly cheaper in year 1 and significantly cheaper from year 2 onwards, with the added benefit of processing invoices in minutes rather than days.

Hidden costs to watch for

These are the expenses that do not appear in any quote but will hit your budget:

Data cleanup. If your systems are messy - duplicate records, inconsistent formats, missing fields - you will need to clean them before an agent can work reliably. Budget £500-£2,000 for this.

Team training. Someone needs to understand how the agent works, how to review its outputs, and when to escalate. This is usually a few hours but it is real time.

Iteration cycles. The first version of an agent is rarely perfect. Budget for 2-3 rounds of refinement in the first month as you discover edge cases. A good agency includes this in the build cost, but ask explicitly.

Scaling costs. API costs are linear with usage. If your agent processes 500 interactions this month and 5,000 next month, your API bill goes up 10x. This is still usually cheaper than hiring, but it should not be a surprise.

How to reduce costs without sacrificing quality

Match the model to the task. Use cheaper, faster models (like Claude Haiku) for simple classification and routing. Reserve powerful models (like Claude Sonnet) for tasks that require reasoning and nuance. A well-architected agent uses different models for different steps.

Cache common responses. If your support agent answers the same 20 questions repeatedly, cache those answers instead of calling the API every time. This can cut API costs by 40-60%.

Batch processing. Instead of running the agent on every event in real time, batch similar tasks and process them together. This reduces API overhead and infrastructure costs.

Start narrow and expand. Build the simplest version that delivers value first. Add complexity only when the data shows it is needed. Many businesses over-engineer their first agent and spend twice what was necessary.

The ROI calculation framework

Here is a straightforward framework for calculating whether an agent is worth the investment:

Step 1 - Calculate the current cost. How many hours per week does the target task consume? Multiply by the hourly rate of the people doing it. Include overhead - management time, error correction, training new staff.

Step 2 - Estimate the agent cost. Use the examples above as a guide. Add 20% for contingency.

Step 3 - Calculate the payback period. Divide the build cost by the monthly savings. Most agents pay for themselves in 1-4 months.

Step 4 - Factor in the intangibles. Speed improvements, consistency, 24/7 availability, and reduced error rates all have value that is hard to quantify but real. A lead response that happens in 30 seconds instead of 3 hours has a measurable impact on conversion rates.

If the payback period is under 6 months and the monthly savings exceed the running costs by at least 3x, it is almost always worth building. If the numbers are tighter than that, an AI strategy session can help you identify whether there is a better starting point.

How to get started

The best way to understand what your specific agent will cost is to talk to us. We will scope it in a free initial call and give you a fixed-price quote before any work begins. No surprises, no scope creep.

Our agent development service covers the full lifecycle - from scoping through deployment and ongoing maintenance. And if the numbers do not make sense for your business, we will tell you that too. We would rather point you to an off-the-shelf tool than build something that does not deliver ROI.

What Does an AI Agent Actually Cost to Build and Run?