Your AI agent finished a competitor analysis in 90 seconds. A junior analyst would have taken four hours. The deliverable is identical. Now you need to write an invoice. What do you charge?
This is the billing question that every professional services firm deploying AI agents now faces. The average firm spends 5–15% of project costs on AI tools, yet fewer than 15% of firms using AI agents for client work have a formal pricing model. The gap between adoption and billing maturity is widening fast. Firms that establish clear, defensible billing practices now will win client trust. Firms that avoid the conversation will face disputes later.
Why Is Billing for AI Work So Difficult?
The hourly billing model breaks down the moment an AI agent enters the picture.
Hourly billing assumes a rough correlation between time spent and value delivered. An experienced consultant bills more per hour because their expertise produces better work. But AI agents shatter that link. An agent can complete research tasks 10–40x faster than a human equivalent. Billing at human hourly rates for 90 seconds of agent compute feels dishonest. Billing at raw compute cost — a few pence — feels like giving away expertise for free.
The tension sits between speed and value. Your firm invested months building, configuring, and testing that agent. The agent’s output reflects your team’s domain knowledge, prompt engineering, and quality assurance processes. The cost of running it is almost irrelevant to its worth.
Clients make the problem harder. Some expect a discount because “the machine did it.” Others don’t care who or what did the work — they care about the result. Without an industry standard, every client conversation starts from scratch.
There is also the risk of getting it wrong in both directions. Overcharge and clients leave. Undercharge and you cannibalise your own margins while training clients to expect AI-era prices for everything, including the human work that still takes real time.
Which Billing Model Works for AI Agent Services?
Four models have emerged for billable hours for AI agent work: time-based, per-task, cost-plus, and outcome-based. The sibling guide covers each model in detail. Here, the question is: which model fits which situation?
| Situation | Recommended Model | Why |
|---|---|---|
| Repeatable tasks with clear scope | Per-task | Clients get price certainty; you keep the margin |
| Variable, exploratory work | Cost-plus | Transparency builds trust when scope is unclear |
| High-value deliverables with measurable results | Outcome-based | Aligns your fee with client value |
| Clients who insist on time-based contracts | Time-based (blended rate) | Familiar format, but use a blended human-agent rate |
| Mixed human + agent projects | Hybrid | Separate line items for human hours and agent tasks |
Decision framework. Ask three questions before choosing a model for each engagement:
- Is the scope fixed or variable? Fixed scope suits per-task pricing. Variable scope needs cost-plus or time-based.
- Can you measure the outcome? If yes, outcome-based pricing captures the most value. If no, fall back to per-task or cost-plus.
- What does the client expect? Some industries default to hourly billing. Meet clients where they are, then educate them on alternatives.
Most firms will end up using a hybrid approach. Human work billed by the hour. Agent work billed per task or at cost-plus. The key is making the distinction visible — not hiding agent contributions inside human line items.
How Should You Communicate AI Usage to Clients?
Transparency about AI usage increases client retention by up to 20% compared to firms that hide it. That statistic alone should settle the debate: tell your clients.
Frame AI as a quality and speed advantage, not a cost-cutting measure. Clients respond well to “we used an AI research agent to analyse 200 sources in 10 minutes, giving you broader coverage than a manual review” and badly to “we saved money by using AI instead of a person.” The framing matters more than the fact.
Show value on invoices, not just cost. Compare these two invoice line items:
| Weak Line Item | Strong Line Item |
|---|---|
| AI agent — 2 min — £0.30 | Competitor analysis (12 competitors, 48 data points, synthesised report) — £45 |
| Automated review — 90 sec | Contract clause review (flagged 3 risk areas, 2 missing provisions) — £25 |
The second column communicates what the client received. The first communicates what the machine consumed. Clients pay for outcomes, not compute time.
Build trust through reporting. Send clients a monthly or quarterly summary showing how AI agents contributed to their projects. Include metrics: tasks completed, time saved versus manual processing, and quality indicators. This positions your firm as a partner investing in better tools — not a firm cutting corners.
Set expectations before you start. Add an AI work clause to your scope of work. Specify which tasks may be handled by agents, which pricing model applies, and how agent work will appear on invoices. Clients who agree upfront rarely dispute at billing time.
Watch for emerging guidelines. Professional bodies in law, accounting, and consulting are beginning to issue guidance on AI disclosure. Stay ahead of these requirements rather than reacting to them.
How Are Different Industries Approaching AI Billing?
No single industry has solved this yet. But each is learning something useful.
Legal: the 6-minute increment problem. Law firms bill in 6-minute increments. An AI research agent that reviews case law in 45 seconds doesn’t fit neatly into that model. Some firms now bill AI legal research at a flat rate per query, treating it like a disbursement (similar to database access fees). Others apply a reduced hourly rate — typically 30–50% of an associate’s rate — to agent work. The firms seeing the least client pushback are those that itemise agent work separately and explain the value of billable hours the agent delivers.
Consulting: AI-driven analysis. Strategy firms use AI agents to process market data, run scenario models, and draft preliminary findings. Most are bundling agent costs into project fees rather than billing them separately. This works when the project price is value-based. It becomes problematic on time-and-materials contracts where clients expect a breakdown.
Development agencies: AI coding agents. Software teams using AI coding agents face a unique challenge. The agent might generate code in seconds, but a human developer still reviews, tests, and integrates it. The billing split is still taking shape: some agencies bill the human review time at full rate and treat agent-generated code as a tool cost. Others charge per feature delivered, regardless of whether a human or agent wrote the initial code.
Marketing: AI content generation. Content agencies use AI agents for first drafts, SEO analysis, and content briefs. The most common approach is per-deliverable pricing — a blog post costs £X whether an agent drafted it or a human did. The margin difference is the agency’s reward for investing in AI tooling.
The common lesson across industries: clients care about what they get, not how it was made. Pricing models that focus on deliverables rather than labour inputs face the least resistance.
What Do You Need to Track for Accurate AI Billing?
You cannot bill what you cannot measure. Accurate AI billing requires a system of record for agent activity.
Five data points to capture for every agent task:
| Data Point | Why It Matters for Billing |
|---|---|
| Task duration (wall-clock time) | Required for time-based billing models |
| Token consumption (input + output) | Drives compute cost calculations |
| API and tool costs | Needed for cost-plus billing accuracy |
| Task output description | Supports value-based invoice line items |
| Client/project attribution | Ensures costs land on the right invoice |
Time tracking is the billing foundation. Understanding what AI agent time tracking involves is the first step. Every agent action needs a timestamp, a task identifier, and a link to a client or project. Without this, you’re estimating — and estimates lead to disputes.
Integrate agent data with your invoicing workflow. Agent tracking data should flow into the same system your team uses for human time tracking. A single platform that shows both human hours and agent tasks per project gives you one source of truth for billing. It also makes reporting to clients straightforward — one view, two types of work, clear attribution.
Audit your agent costs regularly. AI model pricing changes frequently. Token costs drop. New models arrive with different pricing tiers. Review your cost-plus markups quarterly to make sure your margins hold. Track your actual AI agent costs against what you’re billing to catch discrepancies early.
Automate where possible. Manual logging of agent work defeats the purpose. Use event-based logging built into your agent framework. Every task start, tool call, and task completion should generate a record automatically. The goal is zero manual effort to capture agent billing data.
Key Takeaway
Bill AI agent work based on the value it delivers, not the seconds it runs. Track every agent action, communicate transparently with clients, and build billing terms into your contracts before the first invoice goes out.
Start Billing Accurately for AI Agent Work
Keito tracks both human hours and AI agent activity in one platform — giving you the billing data you need for every pricing model.
Frequently Asked Questions
How do you bill clients for AI agent work?
Choose a billing model that fits the task. Per-task pricing works for repeatable work with clear deliverables. Cost-plus works when tasks vary and clients want cost transparency. Outcome-based pricing aligns your fee with results. Whichever model you use, itemise AI work separately on invoices and show what the agent produced, not just how long it ran.
Should you tell clients you used AI?
Yes. Firms that disclose AI use report up to 20% higher client retention than those that don’t. Frame AI as a quality and speed advantage. Include an AI work clause in your scope of work and present agent contributions clearly on invoices. Transparency builds trust; hiding AI involvement creates risk.
What is the best pricing model for AI agent billing?
Per-task billing is the most practical starting point. It gives clients price certainty, protects your margins regardless of how fast the agent works, and avoids the awkward mismatch of charging hourly rates for work that takes seconds. For more detail on each model, see the guide to billable hours for AI agent work.
How do you track AI agent work for invoicing?
Use event-based logging built into your agent framework. Capture task duration, token usage, API costs, output descriptions, and client attribution for every task. Feed this data into your time tracking platform alongside human hours. This gives you one system of record for all billable work — human and AI.
Do different industries bill differently for AI work?
Yes. Law firms are adapting 6-minute increment billing or using flat-rate disbursements for AI research. Consulting firms tend to bundle agent costs into project fees. Development agencies split billing between human review time and agent-generated code. Marketing agencies use per-deliverable pricing. The common thread is a shift towards output-based billing rather than time-based billing.