Lead Source Performance Analysis in 14 Days, Not 14 Weeks

Cardinal Direct, an illustrative Nashville lead buyer running about 9,000 auto leads a month, killed a vendor in June that had been unprofitable since March. Three months of overspend, and the reason was mundane: their lead source performance analysis ran on calendar months, and bind data for a month does not close until the following month, so every review was reading data that was six to ten weeks stale. Nobody was incompetent. The reporting cadence was just built wrong.

The calendar-month trap

Monthly reporting mixes cohorts. The leads you bought on the 2nd have had 29 days to bind. The leads you bought on the 30th have had one. Averaging them produces a number that means nothing, and it means nothing in a specific, dangerous direction: it always makes recent performance look worse and last month look better than it was.

Then someone tries to fix it by waiting longer. Now your lead source performance analysis is accurate and eight weeks late. Accuracy you cannot act on is a hobby.

Cohorting leads properly

A cohort is every lead from one source, one sub-ID, received in one week. You then measure that group at fixed ages (day 2, day 7, day 14, day 30, day 60) and compare it to the same source's earlier cohorts at the identical age.

The discipline is that you never compare a day-7 read to a day-30 read. Ever. Half the bad decisions in lead source performance analysis come from that one apples-to-oranges slip.

Early signals that predict bind

Cardinal Direct backtested twelve months of history and found their day-2 contact rate correlated tightly enough with eventual bind rate that a cohort landing 8 percentage points below its source's trailing average almost never recovered. That is an illustrative finding from one buyer's book, not an industry constant, but the method of deriving your own version of it takes an afternoon.

The 14-day loop

Here is what Cardinal Direct actually runs now. The whole lead source performance loop fits inside one recurring meeting plus two automated reports, which is the only reason it survived contact with a busy quarter.

Four possible outcomes and no fifth. Scale means raise the cap 25%. Hold means no change, review next cohort. Throttle means cut volume in half while you talk to the vendor. Stop means stop. Having exactly four options removes the endless middle ground where a bad source lives for another six weeks.

Throttle before you stop

Throttling is underrated. A vendor with one poisoned sub-ID is not a bad vendor, and offboarding them costs you volume you will scramble to replace. Cut to 50%, send them the sub-ID breakdown, and give them one cohort to fix it. Most will.

Writing kill criteria you will actually honor

Write them before the test. Put them in the vendor onboarding doc so both sides have seen them. Vagueness here is how a source survives on relationship rather than results, and it is the single most common reason lead source performance reviews produce discussion instead of decisions.

• Minimum cohort size before any decision: below this, you observe and say nothing
• Day-2 contact rate floor, expressed as a gap from the source's own trailing average
• Day-14 application-rate floor, absolute
• Maximum consecutive failing cohorts before automatic throttle (two is reasonable)
• A named owner who executes the stop without needing a meeting
• A defined re-test path, so a fixed vendor can earn volume back

Confounders that will fool you

Cohort math is only as good as what you control for. Four things wreck it regularly.

The honest trade-off: rigorous lead source performance analysis will occasionally kill a good vendor during a bad fortnight. That is the price of moving fast. Mitigate it with a documented re-test path rather than by slowing everything down.

Two objections you will hear

Both come up on the vendor call. Both deserve a straight answer rather than a defensive one.

"Fourteen days isn't enough time to judge us"

Correct, if you mean bind rate. Wrong, if you mean contact rate. Take a real Cardinal Direct cohort: 180 leads received in week 31, of which 54 were reached inside 48 hours. That is 30% against a trailing source average of 41%. No amount of extra waiting turns unreachable people into reachable ones, and the gap is 11 points on a base big enough to mean something. The day-14 read is not a verdict on bind. It is a decision about how much more money to expose to a batch whose early signals already sit outside the source's own historical band. Framed that way it is a volume decision, not a character judgment, and most vendors stop arguing.

"Your close rate is the problem, not our leads"

Sometimes fair. Which is exactly why lead source performance analysis is comparative and never absolute. If every source's day-2 contact rate fell in the same week, the problem is yours. If one source dropped 11 points and the other six held flat, it is not. Run that cross-source check before the call. You arrive with an answer instead of an argument, and the conversation gets shorter.

Measuring whether your lead source performance work paid off

A reporting change can fail quietly. You keep producing the cohorts, nobody acts on them, and a year later you are back to reading month-end summaries. So measure the process itself, once a quarter, on three numbers.

The third row is the one that pays for the whole exercise. Cardinal Direct's numbers, illustrative but the shape holds: recovered vendors are cheaper than replacement vendors, and the only reason they recover is that somebody handed them a sub-ID breakdown early enough to fix it. If your lead source performance analysis only ever produces stops and never produces recoveries, your kill criteria are too blunt.

Making it a habit, not a project

The framework is not hard. Sustaining it is. Cardinal Direct's version survives because it takes twenty minutes on a Tuesday, the alerts come to a shared channel, and one person is named on every source. No quarterly business review, no forty-slide deck.

Start with your three largest sources this week. Cohort them by receipt week, read day 2 and day 7, and write down what you will do at day 14 before you get there. Everything else, the backtest, the tripwires, the automated alerts, can come later. Good lead source performance analysis begins as a habit and only afterwards becomes a system.