An e-commerce company’s monthly review shows average order value (AOV) up from ₹800 to ₹846. Good news, surely. On the next slide, revenue is down 17.5%. Both numbers are correct. The puzzle dissolves once you stop looking at metrics as standalone facts and start treating them as products and sums of smaller pieces. That habit, metric decomposition, is the single most useful technique for answering “why did this number move?”

Step one: write the metric as an equation

Revenue is orders multiplied by average order value. Orders, in turn, are visitors multiplied by conversion rate. So revenue = visitors × conversion rate × AOV. Each factor has an owner and a different set of possible causes: traffic belongs to marketing, conversion to product and pricing, AOV to merchandising and promotions.

With the equation in hand, a revenue change can be split into contributions. Suppose orders fell from 50,000 to 40,000 while AOV rose from ₹800 to ₹900, taking revenue from ₹4 crore to ₹3.6 crore. One standard way to attribute the ₹40 lakh drop:

  • Volume effect: the change in orders valued at the old AOV, (40,000 − 50,000) × ₹800 = −₹80 lakh.
  • Rate effect: the change in AOV applied to the new order count, (₹900 − ₹800) × 40,000 = +₹40 lakh.

The two add to −₹40 lakh exactly. The split is not unique: if you value the volume change at the new AOV and the rate change at the old order count, you get slightly different pieces with the same total. That ambiguity is inherent whenever two factors change together. State which convention you used, or use the midpoint method, which averages the two.

Step two: look for mix effects

Now return to the opening example, where AOV rose without anyone changing a single price. The business sells two kinds of product with very different basket sizes:

SegmentLast month ordersThis month ordersAOV (both months)
Grocery40,00030,000₹500
Electronics10,0009,000₹2,000
Total50,00039,000₹800 → ₹846

Within each segment, AOV did not move at all. Grocery orders fell by 25% and electronics by 10%. Because the low-value segment shrank faster, the overall mix tilted towards expensive orders, and the blended AOV rose. Revenue went from ₹4 crore to ₹3.3 crore, a 17.5% fall. The rising AOV is not a sign of success; it is a side effect of losing grocery customers.

Any average across segments of different sizes can move purely because the segment sizes moved. Average salary rises when you lay off junior staff. Average loan size rises when small-ticket lending is paused. Average delivery time improves when you stop serving outlying pin codes. Whenever an average changes, ask whether the underlying segments changed or only their weights.

The extreme case: Simpson’s paradox

Mix effects can be strong enough to reverse a conclusion. Take a site’s conversion rate across two months:

  • Month 1: 80,000 mobile visits converting at 2.0% and 20,000 desktop visits at 5.0%, giving 2,600 orders from 1,00,000 visits, or 2.6% overall.
  • Month 2: 50,000 mobile visits at 1.9% and 50,000 desktop visits at 4.8%, giving 3,350 orders, or 3.35% overall.

Overall conversion rose sharply, yet conversion fell on both mobile and desktop. The overall gain came entirely from traffic shifting towards desktop, which converts better. This reversal is known as Simpson’s paradox. If a product team claimed credit for the overall improvement, they would be claiming credit for something that actually got slightly worse on every device.

Which view is “right” depends on the question. If you want to know whether the checkout experience improved, the segment-level view answers it. If you want to know how many orders the business got, the total is what matters. The mistake is using the total to answer a question about the segments.

A practical order of investigation

When a top-line number moves unexpectedly, a disciplined sequence saves time:

  1. Confirm the data. Check for pipeline delays, duplicated loads, timezone shifts (UTC versus IST can move a day’s orders across a month boundary) and definition changes.
  2. Decompose by formula. Is the change in traffic, conversion, or order value?
  3. Decompose by segment. Region, channel, category, platform, new versus returning. Find where the change is concentrated.
  4. Separate rate from mix. Within the segments, did rates change, or only the weights?
  5. Look for external causes. Festivals, competitor sales, stock-outs, price changes, payment outages.

Each step narrows the search. A 15% sales drop that turns out to be entirely one category in two cities, driven by a stock-out of three products, is a very different conversation from a vague “sales are down, demand must be weak”.

Decomposition does have limits. Factors in an equation are not always independent: raising prices lifts AOV but may cut conversion, so attributing the change to separate factors can suggest separate causes when there is really one decision behind both. Use the decomposition to locate the movement, then reason about causes with that interaction in mind.

Written by the RP Strativa Team. Figures in the examples are illustrative.

Leave a Reply

Trending