Per-Client AI Billing for Agencies: Complete Guide

Without per-client tracking, you can't answer the most important question: Which clients are profitable?

Many agencies discover too late that 30-40% of their clients are unprofitable—consuming more in AI costs than they're paying in retainers. Per-client billing visibility solves this problem by showing exactly where your money goes.

The Problem: Invisible AI Costs

Scenario: You're an agency serving 8 clients. Monthly AI bill from OpenAI: $847. Question: How much of that $847 is Client A vs Client B?

Without per-client tracking, you have no idea. You might be:

• Losing money on 3 clients (undercharging for their usage)
• Overcharging 2 clients (they're subsidizing the unprofitable ones)
• Wasting 2-3 hours per month manually tracking costs in spreadsheets

Two Pricing Models for Agencies

Model 1: Retainer with Included Tokens

Charge a flat monthly retainer that includes X tokens. If client exceeds, they pay overages or upgrade.

Pros of Retainer Model

• Predictable revenue: You know exactly what you'll earn each month
• Easy to explain: Clients understand flat-rate pricing
• Encourages upgrades: Clients naturally move up tiers as usage grows

Cons of Retainer Model

• "Wasteful" usage: Clients may overuse if tokens aren't scarce
• Complex tier management: Need to monitor who's approaching limits

Model 2: Cost-Plus Markup

Charge actual cost + markup percentage. Client pays exactly what they use.

Pros of Cost-Plus Model

• Fair pricing: Clients pay for what they use, nothing more
• No tier management: Usage scales automatically
• Higher margins on heavy users: 30% of $500 > 30% of $100

Cons of Cost-Plus Model

• Variable revenue: Harder to forecast monthly income
• Client education: Must explain why bill varies month-to-month

How to Implement Per-Client Tracking

Step 1: Tag Every Request

Add a client identifier header to all LLM API requests:

Step 2: Export Monthly Usage

At month-end, export usage by client from your LLM gateway:

Step 3: Apply Markup and Invoice

For cost-plus model, apply your markup percentage:

Total time: 30 seconds per client instead of 2-3 hours of manual tracking.

What Markup Percentage Should You Charge?

Industry standard markup ranges from 20-50% depending on value-add:

Handling Unprofitable Clients

What do you do when per-client tracking reveals a client is costing more than they're paying?

Option 1: Optimize Their Usage

Before raising prices, try optimization:

• Are they using GPT-4o when GPT-4o-mini would work? (16x cost reduction)
• Are prompts verbose? Trim unnecessary context (can reduce tokens 30-50%)
• Can you cache repeated requests? (reduces duplicate API calls)

Real example: Client was using 2,400-token prompts that included entire conversation history. Optimized to only include last 3 exchanges. Result: 67% token reduction, client became profitable without price increase.

Option 2: Raise Retainer

If optimization isn't enough, raise the retainer:

Option 3: Switch to Cost-Plus

Move unprofitable retainer clients to cost-plus pricing:

Advanced: Multi-Project Tracking

Some clients have multiple projects. Track costs per project with nested tags:

Then invoice with project-level breakdown:

Automating Client Billing

For agencies with 10+ clients, manual invoicing becomes tedious. Automate it:

Automation Flow

1. On 1st of month, export previous month's usage via API
2. Calculate markup for each client
3. Generate invoice in QuickBooks/Xero/Stripe
4. Send invoice automatically

Time saved: 2-3 hours per month for a 10-client agency.

Related: Complete Guide to LLM Cost Optimization