How to check where the figures came from

An answer you cannot check is an opinion. Every analyst answer carries its own evidence, and this page is how to read it. See What is the analyst? for the analyst itself.

What are the chips under an answer?

A list of what it actually looked at: "Queried Shopify sales", "Checked inventory", "Read Klaviyo segments" and so on.

Read them first, because they tell you whether the answer is built on the thing you meant. An answer about stock that never checked inventory is answering a different question.

A chip with a warning tint is a read that failed. The answer was written without it, and that is exactly when to look at the caveats.

What is in "What I checked"?

Expand it and you get three things.

• Confidence: how sure the analyst is about what it has told you.
• Caveats: what it could not see. This is where a disconnected integration, a failed read or a capped result gets declared.
• The evidence: the specific figures the answer was built from.

The evidence is the part to use. When somebody asks where a number came from in a meeting, it is on screen rather than being something you have to go and reconstruct.

How should you read confidence and caveats?

As the useful half of the answer.

A high-confidence answer with no caveats can be acted on. A confident-sounding answer with a caveat saying the last 30 days of Klaviyo data were unavailable is a different thing, and the prose alone will not always make that obvious.

Check the caveats before you quote a figure to anybody. It takes five seconds and is the difference between the analyst being trustworthy and merely fluent.

What does "not recorded" mean?

That the figure was never written down, as against being zero.

The analyst is careful about this. A delivery whose landed cost nobody entered is reported as not recorded rather than as no cost, and a supplier with no scored deliveries reports no data rather than 0% on time.

That rule runs right through Tideswell: a blank means "could not read" and a zero means zero. See How to read your store dashboard.

What does a capped result look like?

The analyst says so. It reads a bounded number of suppliers, purchase orders, lines and decisions in one answer, and it flags when a result hit that edge rather than presenting a partial list as complete.

Narrow the question when you see it. "Late purchase orders for Portside Mills" returns everything where "all purchase orders" may not.

The same honesty appears on the store dashboard, where a card tells you when it read only recent orders.

Why are money figures missing from my answer?

Because you do not hold the permission that covers them, so they are left out entirely rather than shown as zero.

That is a deliberate rule: a zero would imply a cost was nothing. The analyst would rather say less than mislead you. See How to work out who can do what in your workspace.

What will it never claim?

That one thing caused another.

When it reports a decision beside a measured outcome, it puts them side by side and does not join them. "We ordered early and sell-through was 18% up" is two facts, and only a person who knows the business should decide whether one explains the other.

That restraint is what makes the decision log worth reading, per What are decisions and track records?.

Can you get the underlying screen instead?

Usually, and it is worth doing when a number is going to drive a purchase. Product chips in an answer are live links to that product's record, and the same figures have full screens at What is the store dashboard? and What is the inventory dashboard?.

Use the analyst to find what to look at, then look at it.