Read Your Conversion Rate by Traffic Source, Not in Total

Your total conversion rate is an average of audiences who arrived wanting different things. Email subscribers convert like returning friends; cold social visitors convert like window shoppers; brand searchers arrive nearly decided. When the total moves, the first question is never what changed on the site - it is which source's share of traffic changed, and whether any source's own rate moved. Read the rate by source, over the same period, before touching a page. Half of all apparent conversion problems are audience changes wearing a site problem's clothes.

As a working model: direct and email traffic usually carries the highest intent, then branded search, then non-branded search, then shopping ads, then organic social, then display and cold social at the bottom. Your ladder will differ in detail, but the spread will be wide - several-fold between top and bottom is normal. This is why a viral social post can double your traffic and halve your rate while making you more money: the new visitors were never going to convert like the old ones, and judging the month by the blended rate marks a good event as a failure.

Do the arithmetic once and it becomes reflex. Illustration: last month 2,000 email sessions at 4 percent (80 orders) and 2,000 social sessions at 0.5 percent (10 orders): total 2.25 percent. This month social doubles to 4,000 at the same 0.5 percent (20 orders) and email holds: total is now 100 orders from 6,000 sessions, 1.67 percent. The rate fell by a quarter; nothing got worse; revenue rose. Every CRO review that skips this check risks redesigning a healthy site to treat a healthy audience shift.

The split also directs the work. Low search conversion often means the landing pages do not answer the query that was typed - a message problem. Low email conversion with high clicks suggests the offer or the product pages disappoint - an expectation problem set by the email itself. Low paid conversion with decent organic conversion points at targeting or at the specific landing page the ads use. The same symptom, three different benches. Source-level rates turn one vague worry into a routed ticket.

Set a simple monthly table: for each main source, sessions, orders, rate. Read the pairs together. Rising volume with a gently falling rate in one source usually means broadening reach - fine. Falling volume with a rising rate can mean a source is being starved to its keenest remnant - worth a look. A rate collapse inside one source with stable volume is the genuine alarm: something in that path - the ad, the link, the landing page, the tracking - broke. That is the pattern that justifies dropping everything.

Source data is only as good as your tagging. Use consistent campaign tags on email and social links, or those visits dissolve into direct and referral buckets and the table lies politely. Where the platform and analytics disagree on source, pick one as the record and note the choice in the log. Perfect attribution is not available at this scale; consistent attribution is, and consistency is what comparisons need.

Branded search - visitors who typed your name - deserves its own line in the table, separate from generic search. These visitors largely decided before they arrived; their high rate reflects your reputation, not your pages, and letting it blend into 'search' flatters the search figure and hides the generic-search problem underneath. It also misleads in reverse: a press mention or a marketplace promotion can flood branded search and lift the blended rate, tempting you to credit a page change that shipped the same week. Split branded from generic wherever your search reporting allows. The generic rate is the one your landing pages earn; the branded rate is the one your name earns, and they need different kinds of work.