Decide which change is being asked

“How much did it rise versus last week” can be a delta, a growth rate, or the difference of two rates. Sales from 1,000,000 to 1,100,000 rose by 100,000, which is 10 percent. A margin from 20 percent to 30 percent rose by 10 percentage points. As a rate of change, that same margin rose 50 percent. All three can be called “up” in speech. A finance review cannot mix them.

When the metric is an additive amount, a query can show both the delta and the rate, and name the base period. When the metric is already a rate, the default should be percentage points, with the growth rate stated separately. A bare 10 is not an answer. dbt’s metrics guidance says a ratio carries its numerator and denominator. The same rule applies here.

Do not average rates and then subtract

Store margin should sum margin and sales first, then divide. Averaging each store’s margin and subtracting last week’s average treats every store as the same size, and the result will not match the company profit statement. A star schema puts additive measures on the fact table and stores on a dimension so that recomputation happens at the right grain.

When the follow-up asks which store contributed most, say whether the contribution is a margin delta or a change in percentage points. A large store that gains 0.2 points can add more profit than a small store that gains 5 points. Keeping the previous sort key answers a different question. A new grain often needs a new comparison type.

A zero base, a changed definition, and an incomplete window

If base sales are zero, the growth rate is undefined. Return the delta and say the rate cannot be computed. Do not print infinity or divide by one. New stores, closed stores, and newly listed products land here. The comparable set must be chosen first: all stores, or only stores with sales in both periods. Changing the set changes both the delta and the rate.

Tax-inclusive versus tax-exclusive, or refunds in one period only, is not an operating change. An incomplete month-to-date compared with a full month last year should not produce an unqualified rate. Switching from a rate to a delta does not make an incomparable pair comparable.

Two default strategies

Path one answers a single type after clarification. It fits teams whose metric catalog already maps the word “increase” to either a rate or percentage points. Path two shows delta and rate together for additive metrics, and shows the point change plus the numerator and denominator for ratios. It fits reviews where the word is ambiguous. The first sentence may emphasize only the type the user named, and the table needs a unit column.

Do not let the model pick whichever sentence sounds smoother. If the user asked about sales and then asks about margin, the second turn should move to percentage points instead of inheriting “growth rate” as a slot. Inheriting a comparison type should be as visible and removable as inheriting a time window.

Counterexamples and acceptance

Counterexamples: a margin from 20 percent to 30 percent described as up 10 percent; a simple average of store growth rates presented as the company rate; a huge percent when the base is zero; a follow-up about margin that still ranks stores by the sales delta. Use a fixed fixture: amounts 100 to 110, rates 20 percent to 30 percent, one new store with a zero base, and one large store with a small rate gain. The output should carry the unit, the base, and the comparison type.

The explanation and the query plan must match. The prose cannot say percentage points while the SQL divides. AskTable.ai can be a candidate entry point. This article does not show that the current release already separates deltas, rates, and percentage points, or that it stops dividing when the base is zero. Replay the questions your own review meeting actually asks.

Public references

Ready to help your team start?

Talk through a real scenario and see how AskTable.ai can fit your business.

Book a demo