How to check an AI company's projects and references
A portfolio is useful when the supplier’s contribution and the limits of its evidence are clear. Use this worksheet and reference conversation guide.
Syntalith
When checking an AI company's references, establish its own contribution, how the system is used and the source of any reported outcome. A client name and interface screenshot start the conversation. You still need to know whether the experience matches the work you intend to commission.
Syntalith provides AI implementation services. Apply these questions to our project descriptions as well.
Identify what the example covers
Ask whether the company built the system, one module, an integration or delivered training. Separate development from subsequent maintenance. Experience with one part of a larger platform can be relevant when the contribution is stated accurately.
A demonstration shows a mechanism using prepared data. A customer project may establish that particular work was delivered. A savings claim needs evidence of how the result was calculated. These provide different information for a buyer.
Project evidence worksheet
| Question | What to record |
|---|---|
| Who uses it? | User group and task, where disclosure is permitted |
| What did the supplier do? | Its scope and components delivered by others |
| Where does it run? | Demonstration, pilot or production, with the information date |
| What changed? | Result, source and measurement scope |
| How were errors assessed? | Corrections and cases requiring manual completion |
| Who operates it now? | Current maintenance responsibility |
| What remains unconfirmed? | Missing evidence and follow-up questions |
Do not demand confidential client data. A supplier can discuss the workflow, provide an anonymized example or arrange a reference call with consent. A client declining public disclosure does not establish poor delivery.
Reading a savings claim
For time savings, ask whether the measurement includes data preparation, review and corrections. Check the number and type of cases and the observation period. A result from an occasional task may not describe an entire department.
Generating an answer might be fast while checking its sources takes much longer. Time to an accepted result is the useful comparison. Plan your own check in a pilot.
Speaking with a user
If the client agrees to a conversation, ask how the task worked before implementation, what their team contributed and how problems were handled after launch. One useful question is what needed improvement before people used the system regularly. Establish whether the speaker is assessing development, maintenance or both.
Record the date and permitted use of the responses. Do not publish the conversation as a testimonial without agreement.
Turn evidence into your requirements
Choose a mechanism similar to your workflow and request a demonstration using an agreed scenario. The supplier selection guide covers wider purchasing criteria. For a Syntalith discussion, bring a specific task so we can relate available evidence to the proposed implementation scope.
Syntalith is an OpenAI Select Partner in the OpenAI Partner Network.
We help your team respond to customers faster and find information in company documents. We choose and set up the right AI tools, then teach your team how to use them.
How to introduce ChatGPT and OpenAI at workFind the right role for an agent in your process
Describe the work that currently needs repeated manual action. We will discuss the agent’s responsibilities, system connections and an initial delivery scope.
Explore AI agent development