What Are You Actually Paying Your Supplier For?

A supplier does not need to invent every component to create value. Ordinary software combines frameworks, databases and cloud services. A machine learning product may use established techniques such as gradient boosting, clustering or embeddings.

The commercial value can sit in choosing the right approach, preparing the data, integrating it with a customer’s operations and keeping it reliable. A common algorithm can support a valuable service. A proprietary label can conceal a weak one.

Generative AI makes this distinction more pressing because more of the visible output can be produced using widely available tools.

The AI Value Chain Test

The AI Value Chain Test is the practical assessment proposed in this publication. Start with six questions:

1. What technology or capability is actually proprietary?
2. Whose data makes the result useful?
3. Where does specialist expertise sit?
4. Who validates that the output is correct?
5. Who understands failures and unusual cases?
6. Who remains accountable for the delivered service?

Each question should lead to evidence. Evaluation results, architecture decisions, incident records and demonstrations of difficult cases reveal more than a claim to have a sophisticated AI platform.

If we removed the supplier’s logo, what part of the service could only they provide?

This question is intended to expose the contribution, not to assume that a product assembled from common components lacks value. Dependable operation, accumulated knowledge and effective support can be the hardest parts to reproduce.

Look beyond the demonstration

A convincing demonstration shows what happens on a selected task. Procurement also needs to understand performance across the customer’s actual workload, including exceptions, changing data and mistakes.

For a forecasting service, ask what happens when historical patterns stop holding. For a recommender, ask how quality is measured beyond short term engagement. For an AI support tool, ask when a human takes over and whether that person understands the system’s history.

These are questions about the supplier’s continuing capability. They remain relevant if the underlying model improves or becomes cheaper.

Compare complete costs

Comparing token prices with salaries omits much of the operating model. AI delivery also involves integration, evaluation, oversight, security, failure handling and migration. Human delivery involves recruitment, training, management and tools as well as pay.

The useful unit is reliable, accountable output at an acceptable level of risk. A supplier might automate routine work while improving that outcome. It might also increase output while quietly removing the expertise needed to recognise a failure.

Customers should therefore connect price and staffing discussions to evidence of service quality. Ask what has changed, which capabilities remain and how the supplier demonstrates the promised outcome.

A stronger value proposition can survive this scrutiny. It should make the reasons for the commercial relationship clearer to both parties.

Source note

The AI Value Chain Test is an editorial framework reconstructed from the existing Third-Party Shadow AI master article. It is a set of assessment questions, not an externally validated scoring model or a claim that component reuse diminishes supplier value.