Multi-Provider Cost and Lock-In Review (Retention)

Multi-Provider Cost and Lock-In Review (Retention)

1. 01

   Pull 90 days of per-model spend and token volume from the gateway or provider billing export

   Segment by client account and by model family so a single retainer's exposure is visible rather than blended into agency-wide totals. Tools such as Helicone and Portkey log request-level cost and latency, which makes this export a query rather than a spreadsheet reconstruction.

2. 02

   Flag any account where one provider exceeds 70 percent of that account's inference spend

   Concentration above that line means a single pricing change or deprecation notice can move the account from profitable to underwater inside one billing cycle.

3. 03

   Re-price the top three workflows against at least two alternative endpoints

   Run the same prompts through a second frontier model and through an open-weight deployment to get a real cost-per-completed-task number. OpenRouter and similar routing layers let you compare without rewriting application code.

4. 04

   Check whether client data enters a shared training pool under current terms

   Forrester's September 2026 position is that private deployments outperform public ones for B2B marketing precisely because shared model access erases differentiation. Document the answer per client, not per tool.

5. 05

   Confirm the fallback path works before you need it

   Trigger a deliberate failure against the primary provider in a staging environment and time how long the secondary takes to absorb traffic. A documented fallback that has never been exercised is a plan, not a capability.

6. 06

   Update the retainer's AI cost line and margin assumption in writing

   Send the client a one-page summary showing current spend, the concentration risk, and the cost of the recommended routing change. Cost visibility tools like ClaudeBill exist because per-session spend is otherwise invisible until the invoice lands.

7. 07

   Set the next review date and the trigger that would force an earlier one

   A 90-day cadence is standard; a provider announcing a price change, a model retirement, or a new safety restriction should pull the review forward immediately.