Separate an AI sales recommendation from a forecast you can defend

An AI tool can suggest a next action without proving what the action will earn. Keep observations, assumptions and forecasts separate before changing prices or spending.

Ask what the recommendation is based on

Require the relevant time period, product group and calculation. A statement that one item is growing may reflect a small base, a promotion or a stock change.

Do not accept a confident conclusion if the tool cannot identify the data behind it.

Distinguish revenue from contribution

More sales are not automatically more useful if the offer leaves little after product, packing, payment and delivery costs. Define the cost boundary before asking for an expansion plan.

Use the margin calculator with your actual inputs to make the arithmetic visible. It does not predict demand.

Turn the suggestion into a scenario

Write the proposed action and the assumptions required for it to work. For example, an advertising increase may assume enough stock, a stable conversion rate and an acceptable acquisition cost.

Run a conservative case rather than relying on a single expected outcome. If the recommendation only works under optimistic assumptions, treat that as a decision constraint.

Check what the model cannot observe

Customer intent, future demand, supplier reliability and competitor changes may not be in the dataset. The tool should not convert missing information into certainty.

Likewise, a correlation between a campaign and orders does not by itself establish causation. Ask what else changed during the period.

Choose a bounded test

Define the spending or operational limit, the measurement and the stopping condition. Keep a record of the result so the next recommendation can be compared with reality.

This is a decision framework, not personalised financial advice or a promise about an AI product. Use pricing decision guides to examine the trade-off before letting a generated forecast become a live business commitment.

Illustrative case to check

A tool predicts that doubling advertising spend doubles orders. Ask which evidence supports a constant acquisition cost and whether stock and fulfilment capacity can support the result. Turn it into an explicit scenario, not a forecast presented as fact. Set a limited test and preserve the assumptions so the outcome can be interpreted later.