Guide

Five marketing numbers that are probably lying to you.

Most marketing metrics are reported in the way that flatters them. None of these five is wrong to track, each is just easy to read in a way that leads to confident, expensive mistakes. Here is how each one misleads, and how to read it honestly.

Last-click attribution

Why it misleads

It hands 100% of the credit to the final touch before the sale, usually a branded search or a retargeting ad, and zero to everything that created the demand in the first place. The channels that introduced you look worthless; the channels that closed an already-decided buyer look like heroes.

How to read it honestly

Look at first-click and linear models alongside last-click, and pay attention to the spread between them. A channel that ranks high on last-click but low on first-click is harvesting demand others created, not generating it. For the channels you spend real money on, the only way to know their true contribution is an incrementality test, a geo or spend holdout that measures what happens when you turn them down.

Revenue-based LTV

Why it misleads

Lifetime value quoted on revenue counts money you partly gave back and never actually kept. It ignores refunds, returns and the cost of goods, so it overstates what a customer is worth, often by a wide margin in categories with high returns or thin margins.

How to read it honestly

Compute LTV on contribution margin, net of your real refund rate. State the assumptions on the figure, the margin, the refund rate, the time horizon, so anyone can challenge them. An LTV you can't inspect is a number you can't trust, and an LTV:CAC ratio built on inflated LTV will green-light spend that loses money.

Blended CAC

Why it misleads

A single, averaged cost-to-acquire hides the truth that one channel is usually subsidising another. A cheap channel and an expensive one blend into a number that looks acceptable, while you keep funding the expensive one because the average never flags it.

How to read it honestly

Break customer acquisition cost down by channel and by segment, and judge each against your target, not against the blend. The goal isn't a tidy headline number; it's knowing which specific channels clear the bar and which are quietly dragging the average up.

Platform-reported ROAS

Why it misleads

Every ad platform is graded by its own homework. Each claims the conversions it can plausibly touch, so when you run several at once their reported results, added together, exceed the revenue you actually booked. Trust them at face value and budget drifts to wherever the reporting is most generous.

How to read it honestly

Treat platform ROAS as a starting point, not proof. Reconcile reported conversions against your real revenue and assume overlap. For the bets large enough to matter, run a holdout, the only honest measure of what a channel adds is what you lose when it's switched off.

Social impressions and likes

Why it misleads

Totals, total reach, total impressions, total engagement, climb whenever you post more, whether or not any single post is landing. They feel like progress and are the easiest metric to grow while per-post performance quietly falls.

How to read it honestly

Measure per post, normalised to your own baseline: engagement rate and reach relative to your norm, not raw counts. Track whether reach is genuinely growing or just your posting volume. And be honest that organic social rarely attributes cleanly to revenue, measure what's measurable and don't dress up the rest.

The pattern under all five

None of these is fixed by a prettier dashboard. The fix is to compute the honest version of the number, net of refunds, split by channel, normalised to baseline, and to be candid about what a number can and can’t prove. That is the discipline Decifer is built around: every figure computed from your data and traced to its source, and honest about its limits rather than flattering.

See it applied in the Analytics, Paid and Discoverability modules.

Read your numbers the honest way.

Connect your sources and Decifer computes the margin-true, channel-split versions, and is candid about what the data can and can't settle.

Request access How the engine works