Provider Lock-In Risk Assessment (Onboarding)

Provider Lock-In Risk Assessment (Onboarding)

1. 01

   Inventory every AI model and API dependency across active client projects

   List each provider, the specific model or endpoint, monthly token volume, and the business function it powers, from content generation to coding assistance.

2. 02

   Flag any workflow where a single provider handles more than 60% of total AI calls

   Concentration at this level means a pricing change or outage at that vendor directly threatens delivery timelines and margins on multiple retainers.

3. 03

   Review each provider's pricing model and recent change history for volatility signals

   Compare per-token rates and any announced adjustments across Anthropic, OpenAI, and Google Vertex AI to spot which vendors have shifted costs most in the past year.

4. 04

   Test whether your current codebase can route requests through an abstraction layer without refactoring

   Gateways like Helicone or OpenRouter sit between your app and the model, so confirm your API calls already use a compatible SDK or can be pointed at a proxy endpoint.

5. 05

   Document the data retention and privacy terms for each provider handling client data

   Some clients in regulated industries require that prompts and outputs stay within specific regions or are not used for training, which may rule out certain vendors entirely.

6. 06

   Estimate the migration cost to a second provider for each flagged dependency

   Factor in prompt rewriting, output format differences, and regression testing time, then rank which workflows are cheapest to make portable first.

7. 07

   Define a decision rule for when to trigger a switch, such as a 20% price increase or two consecutive availability incidents

   A pre-agreed threshold removes emotion from the call and lets you act before a cost spike erodes the project margin.