Vendor diligence

ChatGPT wrapper vs custom AI: how to tell

A vendor selling a "proprietary AI" at six-figure budget might just be calling the OpenAI API. Here is how to tell what you are paying for.

Score a vendor in 5 min Read the methodology

The wrapper pattern

Proprietary model claim + fixed-price retainer + budget over 50K + no model disclosure clause = wrapper pattern. The product is a thin layer over a frontier API; the value is the prompt engineering, not the model.

Wrappers can be valuable

Some wrappers are excellent businesses. The problem is opaque pricing. If you are paying frontier-model resale + 10x markup with no transparency, the value capture is wrong.

How to verify

Ask for the model disclosure clause. Ask for the data flow diagram. Ask for the switching policy. Ask for the per-token cost pass-through.

When wrappers are fine

When the wrapper does meaningful prompt engineering, retrieval, evaluation, and orchestration. When the pricing reflects the engineering work, not just the model resale.

Try it

Run the 12-question wizard

Five minutes. Free preview is anonymous; the full scorecard + PDF is gated by a work email.

Vendor diligence

Question 1 of 12 · 8%

Auto-saved locally. Refresh-safe.

What kind of vendor are they?

Different archetypes carry different risk profiles. Boutique consultancies, agencies, SaaS vendors, freelancers, and staff-aug shops all fail in distinct ways - we adapt the rest of the questions to the archetype you pick.

Want a second opinion?

Run the proposal past buzzi - free 30 min.

No pitch. Senior delivery consultants will sanity-check the contract and flag the three things to negotiate hardest.

Book the call