Answer
How do you tell if an AI vendor is selling a wrapper?
Being a wrapper is not the problem. Being a wrapper with no data, no integrations and no fallback is.
Ask what stays if the underlying model changes. A thin product has no proprietary data, no real integrations, no fallback when the model is wrong, and no way to export your history. Building on someone else's model is normal.
The word wrapper is used as an accusation and it is not a useful one, because nearly every business-facing AI product is built on a model somebody else trained, including the good ones. That is a sensible engineering decision and it says nothing about whether the product is worth buying. What distinguishes a durable supplier from a thin one is what exists around the model, and it can be established with a handful of questions that vendors answer very differently.
The first is data. Does the product accumulate anything that makes it better for you specifically — your past conversations, your service catalogue, your pricing rules, your history of what customers actually ask? A supplier with a real answer will describe how that store is built and what happens to it. One without will talk about the quality of its prompts, which is a description of something that can be reproduced by anyone in an afternoon.
The second is integration, and it is where thin products are most exposed. Genuinely connecting to the systems a business runs on is unglamorous, slow work, and it is exactly what a product assembled quickly will not have done. Ask which systems it connects to, in which versions, whether the connection is bidirectional, and what happens when the other system changes. A supplier who describes integrations as coming soon is describing the largest part of the work.
The third is fallback, which matters more than accuracy and is asked about less. What happens when the model is wrong or unavailable? Does the system detect that it is uncertain and escalate, or does it produce a confident answer regardless? Is there a path that still works if the underlying provider has an outage? A product with no answer here is passing the entire risk of the model's behaviour through to the buyer while charging for having managed it.
The fourth is portability, and it is the one that decides how much leverage you retain. Can you export your conversation history, your configuration and your accumulated data in a usable form? A supplier who cannot say yes clearly is selling something you can only leave by starting again, which is a commercial position rather than a technical limitation, and it should be priced accordingly.
The fifth is more diagnostic than any feature question: ask what the product will not do. A supplier who names limits — the cases it escalates, the questions it declines, the situations where a person is required — has thought about failure, which is the thing that will determine your experience of it. A supplier for whom everything is possible has either not deployed the product much or is not describing it accurately, and both are the same problem for you.
Every serious question about an AI supplier reduces to the same one: what would still be here if the model underneath it were swapped tomorrow?
Answer Production Engine, Context Theory
Related questions
Does it matter which underlying model a vendor uses?
Less than how they handle it changing. Models are updated, deprecated and repriced on the provider's schedule, not the vendor's, so the meaningful question is whether the vendor has abstracted that away and tests behaviour when it moves. A vendor whose product is tied to one specific model version, with no evaluation process for changes, will pass every provider change through to you as a surprise.
Is a cheaper thin product ever the right choice?
For a narrow, low-stakes task with a short commitment, yes, and it is often the sensible way to establish whether the workflow helps at all before committing to anything larger. What makes it a bad choice is scope: the moment it holds data you would miss, sits between you and customers, or becomes the only record of something, the portability and fallback questions stop being theoretical.
METHOD
Every figure below carries its source and the date it was verified. Nothing on this page is asserted.
The numbers on this page.
| What | Value | Specific to |
|---|---|---|
| Buyers preferring a rep-free purchase path | two-thirds | Category-wide |
| Buyers who eliminate vendors publishing no pricing, before contact | 60% | Category-wide |
| Share of the buying journey completed before contacting a vendor | 60% | Category-wide |
Gartner · March 2026 · verified
2026 B2B buyer surveys · verified
What is specific to this page.
| Kind | Claim | Check it against |
|---|---|---|
| Procurement | Building a business product on a model trained by someone else is standard engineering practice and does not distinguish suppliers, so the durable differences lie in accumulated data, integrations, fallback behaviour and portability. | The supplier's own technical documentation, checked for what it describes beyond prompt design. |
| Software | Connecting to the systems a business runs on requires version-specific, bidirectional work that a quickly assembled product will not have done, so a supplier describing integrations as forthcoming is describing the largest remaining part of the build. | The supplier's published integration list with named systems and versions, and its stated behaviour when a connected system changes. |
| Software | A product with no defined behaviour for model error or provider outage passes the entire risk of the model's behaviour to the buyer while charging for having managed it, which is testable before purchase. | The supplier's documented escalation and degradation behaviour, plus its status history for the underlying provider's outages. |
| Procurement | Inability to export conversation history, configuration and accumulated data in a usable form is a commercial position rather than a technical limitation, and it determines the buyer's leverage at every subsequent renewal. | The supplier's contractual data export provisions and the formats it commits to on termination. |
Each row would be wrong on another industry's page. Where a sourced figure exists it is in the table above instead; these are the constraints that shape the work and do not happen to be numbers.
Start with the measurement.
Reading about a benchmark is not the same as knowing your own number. The audit produces yours, measured rather than estimated.
$497 · delivered in 5 business days · credited against month one