Track AI Costs Per Client: Usage Attribution for Agencies and Consultancies

The agency question is always the same: whose usage is this? When one team runs dozens of clients through shared models, the bill is a single number — and every client line item has to be reconstructed from memory. That breaks both billing and trust.

Per-client tracking has three layers: per-client keys, tagged usage, and a review cadence. This guide covers all three. LayerFlow's client tags implement it; the docs show the setup.

Layer 1: Per-client keys

Give each client its own API key, named after the client, with its own cap. Attribution becomes structural: the provider dashboard groups by key, and the bill maps to clients without any tagging discipline. Per-client keys are also the compliance answer — client data runs under a key that is only ever used for that client.

Layer 2: Tagged usage

• Tag prompts by client, project, and task type at creation time.
• Store tags on prompts, versions, and runs — not just the final outputs.
• Use tags for reporting AND routing: cost attribution and budget enforcement share the tag.
• Keep a fixed tag taxonomy; free-text tags decay into unusable data.

Keys give the ceiling; tags give the breakdown. Without tags, a client key tells you the total, but not which project or task inside the client burned it.

Layer 3: The review cadence

1. Monthly: reconcile provider spend against client budgets.
2. Quarterly: re-estimate client AI budgets from actuals, not guesses.
3. Incident: alert at the cap threshold, before the client invoice, not after.
4. Every review ends with the question: does the client's budget match the usage pattern?

Mistakes that ruin attribution

• One shared key across all clients — attribution is impossible, not difficult.
• Tags added after the fact from memory.
• No caps per client, so one client's workload bleeds into another's budget.
• Provider dashboards as the only report — nobody reads them monthly.
• Treating AI costs as overhead instead of pass-through.

Internal next steps

Foundations first: BYOK for Beginners and Team API Keys. Then Hard Budgets for AI Teams and AI Spend Analytics Dashboards.

Attribution from day one: sign in to LayerFlow and set up per-client keys and tags, or check pricing.

FAQ