Know what changes
before you change models.

Yes. Use the free readiness guide to find existing logs, policies and outcome records, and note what is missing. It is available without an account. If you already have a suitable evaluation setup and can investigate the results, you may be able to carry out the comparison yourself. The assisted assessment is for teams that want Pisama to undertake that work with them.

The $500 introductory service fee covers the full assessment of one agreed workflow: evidence preparation, replay setup and incumbent capture, proposed criteria confirmed by your experts, comparison and investigation across up to three development configurations in total, one final assessment of a locked candidate, and an evidence-backed recommendation and handoff. Inference is separate. Payment is $250 (50%) at accepted scope and $250 (50%) at delivery of the agreed evidence package, whether the result is pass, fail or inconclusive.

You use your provider accounts and pay the providers directly for inference. Pisama charges the assessment fee without token markup. We agree on scope and estimate inference across the required runs before starting. BYOK is a billing choice; it does not mean execution is customer-hosted or that data stays inside your environment.

Yes, by agreement. The same assessment fee applies, plus an inference allowance agreed before work begins and billed transparently. We agree any additional allowance before further testing.

One text workflow with structured decisions, up to three model calls and one read-only tool. Fine-tuned incumbents can be assessed. Training delivery, write tools and autonomous agents are outside the initial scope.

We compare model responses to recorded requests, using the context and prior tool responses already in your logs. Live tool execution and end-to-end workflow behavior need separate validation.

No. Start with the logs, policies, tests and outcome records you already have. Pisama identifies usable evidence, prepares the replay and proposes criteria for your experts to confirm. We run the model comparisons and investigate the results. Replay feasibility and any missing inputs are checked before the engagement is accepted.

Your team supplies approved cases. We agree what data can be used and how it will be handled before testing. Customer-prepared anonymized or synthetic cases must preserve the context needed to judge outcomes, and any coverage limits are recorded. Do not include sensitive records in the initial enquiry.

No. A candidate may pass, fail or remain inconclusive within the tested scope. You are paying for the evidence needed to make the decision, including a clear finding that a configuration is unsuitable.

Yes. We start with the provider’s recommended replacement and the minimum integration changes it needs. Other models or serving routes are considered when the evidence gives a reason to explore them.

Share existing evidence you are permitted to use, identify your current configuration and name a domain contact. Tell us about your decision deadline. We agree provider access and data handling with you before testing.

Your experts confirm the proposed criteria, answer focused questions with the evidence alongside them and review an agreed audit sample. Pisama handles evidence preparation, model comparisons and investigation. Where existing evidence is insufficient, we identify the gaps and agree any additional expert review before proceeding.

Before work begins, we agree the review plan and estimate expert review and engineering effort separately. Your team also owns deployment and the rollout decision.

Yes. Approved rules and reviewed examples provide a starting point for comparable cases. We check that they still apply when your workflow or policies change, and bring exceptions back for review. Assessment criteria are agreed before model comparisons begin and stay fixed for final confirmation.