An AI agent rate card lists published prices per agent-hour or per task, tiered by complexity, that professional services firms present to clients alongside human rates — turning opaque compute costs into clear, predictable pricing.
A client asks for your firm’s rate card. You hand them the standard sheet — senior consultant at £250/hour, junior at £120/hour. Then they ask: what about your AI agents? If the answer is silence, you are behind. Professional services firms deploying AI agents need published rate cards that give clients clear pricing, protect firm margins, and position agent work as a defined service tier — not an afterthought billed on an ad-hoc basis.
Key Takeaway: Calculate agent cost per hour (typically £3-8), apply a 3-5x margin multiplier, and tier rates by task complexity. Review quarterly.
Why Do AI Agents Need Rate Cards?
Rate cards bring structure to agent pricing. Without them, every engagement becomes a custom negotiation that wastes time and signals immaturity.
Clients expect published rates. Professional services clients are accustomed to rate cards. They budget based on published rates, compare firms based on rates, and build project estimates from rates. If your AI agents do not have published rates, clients cannot factor them into their planning — and your firm looks unprepared.
Rate cards protect margin. A well-calculated rate card includes cost, overhead, and margin by design. Every bill generated from the rate card carries the intended margin. Without a rate card, pricing is improvised — and improvised pricing tends to be either too high (losing competitive deals) or too low (eroding margin).
They simplify internal processes. Estimating project costs, building client quotes, generating invoices, and analysing profitability all become faster when they reference a single set of published rates. Rate cards eliminate the “what should we charge?” conversation on every project.
Mixed billing becomes coherent. Most projects involve both human and AI agent work. A single rate card covering both — human rates in one section, agent rates in another — gives clients a unified view and makes blended billing straightforward. For firms billing clients for AI agent work, the rate card is the foundation.
How Do You Calculate Agent Cost Per Hour?
The rate card starts with an accurate cost per hour of active agent work. Four cost components contribute.
Step 1: Model Inference Costs
This is the largest direct cost. Calculate the agent’s average token consumption per hour of active work, then multiply by the model provider’s per-token pricing.
A coding agent might consume 600,000 tokens per hour of active work (input and output combined). At £0.003 per 1,000 input tokens and £0.015 per 1,000 output tokens, with a 3:1 input-to-output ratio, the hourly inference cost is approximately £2.40.
A research agent consuming fewer tokens on a mid-tier model might cost £1.20/hour. A complex reasoning agent on a premium model might cost £6.00/hour.
Step 2: Tool Costs
Agents call external tools — APIs, databases, file systems, web services. Each call has a cost.
A research agent making 15 API calls per hour at £0.05 each costs £0.75/hour in tools. A coding agent running 8 build and test cycles per hour at £0.10 each costs £0.80/hour.
Tool costs vary significantly by workflow. An agent with no external tool calls has zero tool cost. An agent making hundreds of API calls per hour could have tool costs exceeding inference costs. Track actuals for accuracy.
Step 3: Infrastructure Overhead
Shared infrastructure costs amortised across agent-hours: hosting, orchestration platform fees, monitoring and logging, data storage, and network transfer.
For a typical deployment, infrastructure overhead runs £0.50-1.50 per agent-hour. A small deployment (2-3 agents) has higher per-agent overhead because fixed costs are spread across fewer hours. A larger deployment (10+ agents) achieves lower per-agent overhead through scale.
Step 4: Operational Costs
The ongoing human effort to maintain the agent: prompt engineering, testing, updates, monitoring, incident response. Amortised across the agent’s total productive hours.
If an agent requires 10 hours of maintenance per month from a £60/hour engineer (£600/month), and the agent runs 200 productive hours per month, the operational cost is £3.00 per productive hour. As agent usage increases, this cost per hour decreases.
Worked Example
| Cost Component | Research Agent | Coding Agent | Summarisation Agent |
|---|---|---|---|
| Model inference | £1.20/hour | £2.40/hour | £0.30/hour |
| Tool costs | £0.75/hour | £0.80/hour | £0.10/hour |
| Infrastructure | £0.80/hour | £0.80/hour | £0.80/hour |
| Operational | £1.50/hour | £2.00/hour | £0.80/hour |
| Total cost/hour | £4.25/hour | £6.00/hour | £2.00/hour |
These costs change as model pricing shifts and agent usage scales. Track actual costs monthly and update calculations quarterly.
How Do You Set Tiered Rates and Margins?
Cost per hour gives you the floor. Margin and tiering give you the rate card.
Applying Margin
Professional services firms typically bill human time at 3-5x the fully loaded employee cost. The same principle applies to AI agents. A 4x multiplier on a £5.00/hour agent cost gives a £20.00/hour client rate.
Why 3-5x rather than 2x or 10x? Below 3x, the margin does not cover business overhead, sales, and profit. Above 5x, the rate starts to lose its cost advantage over human rates — the whole point of agent pricing is that it is materially cheaper than human alternatives while remaining profitable for the firm.
Tiered Rates by Complexity
Not all agent tasks are equal. A data formatting task is simpler, faster, and cheaper than a multi-step research analysis. Tiered rates reflect this difference.
| Tier | Task Types | Typical Cost/Hour | Multiplier | Client Rate/Hour |
|---|---|---|---|---|
| Tier 1 — Simple | Data extraction, formatting, summarisation, classification | £2-4 | 4x | £8-16 |
| Tier 2 — Standard | Research, analysis, drafting, comparison, review | £4-8 | 4x | £16-32 |
| Tier 3 — Complex | Multi-step reasoning, creative output, cross-referencing, synthesis | £8-15 | 4x | £32-60 |
Some firms add premium rates for: urgent processing (1.5x standard rate), after-hours execution (1.25x), high-security environments requiring isolated infrastructure (2x), and custom agent configurations for specific client workflows (negotiated).
Example Rate Card
A mid-sized consulting firm’s AI agent rate card might look like this:
| Resource | Hourly Rate |
|---|---|
| Human — Partner | £450 |
| Human — Senior Consultant | £250 |
| Human — Consultant | £150 |
| Human — Analyst | £100 |
| AI Agent — Complex Tasks | £50 |
| AI Agent — Standard Tasks | £25 |
| AI Agent — Simple Tasks | £12 |
The visual hierarchy communicates value. Human expertise commands the highest rates. AI agents handle volume work at a fraction of the cost. Clients see the full resource spectrum and can make informed choices about how their project is staffed.
How Do Agent Rates Compare to Human Rates?
Agent rates should sit at 10-40% of the equivalent human hourly rate. This range reflects three realities.
Lower cost base. An AI agent costs £3-8/hour to run. A human consultant costs £40-150/hour in fully loaded compensation (salary, benefits, office, training, management). The cost differential is 10-40x.
Different accountability model. A human professional is personally accountable for their output. They bring judgement, client relationship management, and professional liability. An AI agent does not. The lower rate reflects the lower accountability and the need for human oversight.
Value positioning. Agent rates must be low enough to be obviously cheaper than human alternatives. If agent rates creep above 40% of human rates, the cost saving no longer justifies the perceived risk of AI-generated output. The sweet spot is dramatic enough to make the choice easy for clients.
Blended Rate Calculations
Most projects involve both human and agent work. A blended rate simplifies quoting and budgeting.
Project example: A due diligence project requires 10 hours of AI agent research (at £25/hour) and 5 hours of senior consultant analysis (at £250/hour).
- Agent cost: 10 x £25 = £250
- Human cost: 5 x £250 = £1,250
- Total: £1,500 over 15 hours
- Blended rate: £100/hour
Compare this to the all-human alternative: 15 hours of senior consultant time at £250/hour = £3,750. The blended approach saves the client 60%.
Blended rates make it easy for clients to compare AI-augmented delivery against traditional human-only delivery. The saving is the selling point. For firms considering different billing models, blended rates work within time-based, hybrid, and outcome-based frameworks.
How Do You Present Rate Cards to Clients?
The rate card presentation determines whether clients see agent pricing as a cost reduction or a compromise.
Keep It Simple
A one-page rate card with human and agent rates side by side. No jargon. No footnotes longer than the card. If clients need a paragraph to understand the pricing, it is too complicated.
Explain What the Rate Covers
Clients will ask: “I’m paying £25/hour for a computer?” Frame the rate card in terms of what the client receives:
“The agent rate covers: AI model processing, quality assurance checks, output monitoring, technical support, and access to our proprietary agent configurations. All agent output is reviewed by a qualified professional before delivery.”
The answer is never “you’re paying for compute.” The answer is “you’re paying for a result delivered faster and cheaper than the human alternative, with quality guarantees.”
Show Comparative Examples
Include 2-3 worked examples on the rate card or in an accompanying sheet:
Example 1: “Market research report — Agent: 2 hours at £25/hour = £50. Human equivalent: 8 hours at £250/hour = £2,000. Saving: 97.5%.”
Example 2: “Document review (100 documents) — Agent: 4 hours at £12/hour = £48. Human equivalent: 20 hours at £150/hour = £3,000. Saving: 98.4%.”
These examples make the value proposition concrete rather than abstract.
Include Quality Guarantees
Pair the rate card with SLA commitments: turnaround time guarantees, accuracy targets, and revision policies. This addresses the unspoken client concern — “cheaper means worse” — with measurable commitments. Firms that track agent SLAs can back these guarantees with data.
How Do You Keep Rate Cards Current?
Model costs are falling. Major providers have reduced pricing by 50-80% over the past 18 months (2024-2025). Agent costs per hour are trending downward. Rate cards need to keep pace.
Quarterly Review Cycle
Every quarter, recalculate agent cost per hour using actual spend data from the past 90 days. Compare the new cost against the published rate. If costs have dropped significantly (more than 20%), decide: pass savings to clients or maintain rates.
Pass savings to clients — reduces the rate card, increases competitive advantage, strengthens client relationships. Works when the market is competitive and price is a differentiator.
Maintain rates — holds the current rate card, expands margin. Works when the firm has strong client relationships, differentiated services, or limited competition.
Most firms do a blend: pass partial savings to clients (reducing rates by half the cost reduction) while retaining partial margin expansion. A £5.00/hour cost drop might translate to a £2.50/hour rate reduction and a £2.50/hour margin increase.
Version and Communicate
Date every rate card version. Communicate changes to clients with 30-60 days’ notice. Honour existing engagement rates for their contracted period. This professionalism builds trust and avoids surprises.
Track Trends for Forecasting
Model pricing trends help forecast future rate card changes. If costs have dropped 50% in 18 months, plan for further reductions. Build this into engagement pricing for multi-year contracts. A rate that is profitable at today’s cost is even more profitable at next year’s lower cost — as long as the rate card is not renegotiated down at the same pace.
Keito calculates your AI agent cost per hour automatically — tracking model inference, tool costs, infrastructure, and operational overhead in real time. Build and maintain accurate rate cards backed by actual cost data.
Frequently Asked Questions
What is an AI agent rate card?
An AI agent rate card is a published pricing sheet listing the hourly or per-task rates for AI agent work, tiered by task complexity. It sits alongside human professional rates and gives clients clear, predictable pricing for agent-delivered services. Rate cards protect firm margins and simplify estimating, quoting, and invoicing.
How do you calculate AI agent cost per hour?
Add four components: model inference costs (token consumption x per-token price), tool costs (API calls and external services), infrastructure overhead (hosting, orchestration, monitoring — amortised per hour), and operational costs (maintenance and engineering — amortised per hour). A typical mid-complexity agent costs £3-8 per hour.
What margin should you add to AI agent rates?
Professional services firms typically apply a 3-5x multiplier on agent cost per hour, matching the margin structure used for human billing. A 4x multiplier on a £5/hour agent cost gives a £20/hour client rate. Below 3x, margins are too thin to cover business overhead. Above 5x, rates lose their cost advantage over human alternatives.
How do AI agent rates compare to human rates?
AI agent rates typically sit at 10-40% of equivalent human hourly rates. A task billed at £250/hour for a senior consultant might be billed at £25-50/hour for an AI agent. The differential reflects the lower cost base, different accountability model, and the need for the pricing to clearly communicate cost savings to clients.
How often should you update AI agent rate cards?
Review rate cards quarterly using actual cost data from the past 90 days. Model provider pricing has dropped 50-80% over recent 18-month periods, which affects agent cost per hour. Decide each quarter whether to pass savings to clients (competitive positioning) or maintain rates (margin expansion). Version and date every rate card, and give clients 30-60 days’ notice of changes.