Bill for AI agent work by choosing a pricing model (per-task, cost-plus, time-based, or outcome-based), tracking agent activity at the task level, and presenting clients with itemised breakdowns of what the agent did and what it cost.
AI agents now draft contracts, review code, generate reports, and run data analysis for paying clients. The work is real. The value is measurable. But the billing? Most firms are still guessing. A 2025 Deloitte study found that 25% of enterprises expected AI agents to perform autonomous work by year-end — yet no industry-standard billing model exists for this work. The firms that figure out billable hours for AI agent work first will set pricing expectations for everyone else.
Why Does AI Agent Work Need Its Own Billing Model?
Traditional billable hours assume one person, one task, one clock. AI agents break all three assumptions.
An agent can run five research tasks in parallel. It can finish a code review in 90 seconds that would take a developer 45 minutes. It might cost £0.03 in compute or £3.00, depending on the complexity. Billing by the hour — the default for human work — creates a mismatch. Charge at human rates and you overcharge for trivial tasks. Charge at compute cost and you undervalue the expertise baked into your agent’s configuration.
The market has no standard yet. According to industry surveys, fewer than 15% of firms using AI agents for client work have a formal pricing model for that work. Most either absorb the cost or bundle it into existing human rates without disclosure. Neither approach is sustainable. Clients will eventually ask what they are paying for — and firms without a clear answer will lose trust.
This is different from the technical question of how to track time for AI agents. Tracking captures the data. Billing turns that data into revenue.
What Pricing Models Work for AI Agent Billing?
Four models have emerged. Each fits different types of work.
| Model | How It Works | Best For | Risk |
|---|---|---|---|
| Time-based | Log agent compute time, bill per minute/hour | Long-running workflows | Undervalues fast agents |
| Per-task | Fixed fee per completed unit | Repeatable tasks (reports, reviews) | Scope creep on complex tasks |
| Cost-plus | Pass through API/compute costs + markup | Transparent client relationships | Requires detailed cost tracking |
| Outcome-based | Charge by results delivered | High-value deliverables | Hard to standardise |
Time-Based Billing
Log how long the agent actively processes a task. Bill by the minute or hour, similar to human billing. This is the most familiar model for clients accustomed to hourly rates.
The problem: a well-tuned agent completes work faster over time. If your agent finishes a task in 3 minutes instead of 30, your revenue drops 90% for the same output. Time-based billing penalises efficiency.
Per-Task Billing
Charge a fixed fee per completed unit. A contract review: £15. A data analysis report: £50. A code review with annotations: £25. The client knows the cost upfront. You keep the margin whether the agent takes 2 minutes or 20.
This is the clearest value exchange. Industry practitioners working with project-based billing tools recommend this model for repeatable, well-defined tasks where scope is predictable.
Cost-Plus Billing
Pass through the actual API costs, compute charges, and token usage — then add a markup (typically 30-100%). The client sees exactly what the agent consumed. You earn a margin on infrastructure.
This works when trust is high and tasks are variable. It requires granular cost tracking — you need to know that a specific research task consumed 12,000 tokens at £0.002 per 1,000 tokens, plus two web search API calls at £0.01 each. Without proper AI agent cost tracking, this model falls apart.
Outcome-Based Billing
Charge based on the result, not the effort. An agent that generates 50 qualified leads: bill per lead. An agent that resolves 200 support tickets: bill per resolution. This creates the strongest alignment between your costs and client value.
The difficulty is standardisation. Not every agent task has a clean, measurable outcome. And clients may dispute whether an outcome meets the agreed quality bar.
How Do You Track AI Agent Costs for Billing?
Accurate billing requires accurate tracking. Three layers of data matter.
Task-level logging. Every agent action needs a start timestamp, an end timestamp, and a task identifier linked to a client or project. Traditional timers do not work here — agents do not “start” and “stop” like humans pressing a button. You need event-based logging built into your agent framework.
Token and API cost capture. Record input tokens, output tokens, and the model used for every API call. Different models have different costs. A task using a smaller model at £0.001 per 1,000 tokens costs a fraction of the same task on a larger model at £0.06 per 1,000 tokens. Without this granularity, cost-plus billing is impossible.
Cost-per-action mapping. Beyond API costs, track tool usage: web searches, database queries, file operations, code execution. Each has a cost. Map every action to its compute cost so you can attribute total cost per task, per client, per project.
The infrastructure for this is straightforward: webhook listeners on your agent framework, a logging layer that writes to a time-tracking platform, and dashboards that aggregate by client. The technical setup is covered in detail in the guide on tracking time for AI agents.
How Should You Present AI Agent Work on Invoices?
Transparency is the difference between a trusted partner and a firm that hides its methods.
Itemise agent work separately. Do not bury AI agent contributions inside human line items. Create a distinct section on invoices: “AI-Assisted Work” or “Automated Analysis.” List the task, the duration or unit count, and the cost.
Show the value, not just the cost. A line item that says “AI research — 3 minutes — £0.45” looks trivial. A line item that says “Market competitor analysis (12 sources, 4,200 words synthesised) — £15” communicates value. Frame agent work by what it produced, not how long the machine ran.
Set expectations in the scope of work. Before the project starts, define which tasks will be handled by AI agents. Include this in your proposal or contract. Specify the pricing model. Clients who agree upfront rarely dispute at invoice time.
Address the ethics question directly. Should clients pay human rates for AI work? Most industry practitioners say no. Charging £250/hour for 3 minutes of agent compute destroys trust. But charging £0.45 for a task that replaced 2 hours of analyst work undervalues the output. The answer is somewhere in between — per-task or outcome-based pricing sidesteps this tension entirely.
According to early adopter surveys, firms that proactively disclose AI use in client work report higher client satisfaction than firms that do not. Transparency builds trust. Hiding AI involvement creates risk.
What Are the Practical Steps to Start Billing for AI Agent Work?
Five steps move you from ad-hoc to structured AI billing.
-
Audit your agent tasks. List every task your AI agents perform for clients. Categorise each as repeatable (fixed scope) or variable (scope changes per engagement).
-
Choose a pricing model per task type. Use per-task billing for repeatable work. Use cost-plus for variable work. Reserve outcome-based billing for high-value deliverables where you can measure results clearly.
-
Instrument your agents. Set up event-based logging that captures task duration, token usage, tool calls, and costs. Connect this to your time-tracking platform so agent work appears alongside human hours.
-
Build separated reporting. Create dashboards and invoice templates that show human contributions and AI contributions distinctly. Clients should see both in one view but understand which is which.
-
Update your contracts. Add AI work clauses to your scope of work agreements. Specify which tasks may be handled by AI, what pricing model applies, and how work will be reported. A single paragraph in your terms can prevent months of billing disputes.
Key Takeaway
Bill AI agent work by task or outcome, not by the hour. Track every action, be transparent with clients, and update contracts before the first invoice.
Track AI Agent Billable Hours Automatically
Keito logs every AI agent action — time, tokens, and cost — alongside your human team’s hours.
Frequently Asked Questions
How do you bill clients for AI agent work?
Choose a pricing model that fits the task type. Per-task billing works for repeatable work with clear scope. Cost-plus billing works when tasks vary and the client wants transparency into actual costs. Outcome-based billing aligns price with results. Whichever model you choose, itemise AI work separately on invoices.
What pricing model works best for AI agent billing?
Per-task billing is the most practical starting point. It gives clients price certainty, rewards you for building efficient agents, and avoids the awkward mismatch of charging hourly rates for work that takes seconds. Cost-plus is a good alternative when tasks are highly variable.
Should you charge human rates for AI-generated work?
No. Charging full human hourly rates for AI agent output erodes client trust when the work took seconds of compute time. Instead, price by the task or outcome. This captures the value of the work without the perception of overcharging for machine labour.
How do you track time for an AI agent?
Use event-based logging rather than manual timers. Instrument your agent framework to record start and end timestamps, token consumption, API calls, and tool usage for every task. Map each action to a client and project. See the full guide on tracking time for AI agents.
Do clients need to know when AI did the work?
Yes. Industry trends show that firms which proactively disclose AI use report higher client satisfaction. Include AI work clauses in your contracts, present agent contributions clearly on invoices, and frame AI assistance as a quality and speed advantage — not something to hide.