When the Ad Platform Grades Its Own Homework, Budgets Bleed

The platform reports a strong month. Sales has no record of those customers.

Everyone in the room has a number and no two of them reconcile, so the meeting settles into the familiar shape: marketing defends the platform, sales defends the pipeline, and the discrepancy survives to appear again next month.

The platform may be telling the truth

The instinct is to suspect the platform of marking its own work. That instinct has a real basis — an advertising system does define, count and report the outcome it is being judged on, which is a structure worth being alert to.

It is also the wrong first explanation, and reaching for it ends the investigation early. When three systems disagree, it is entirely possible that all three are counting correctly. They may be counting different events, across different windows, under different definitions of one customer.

Once that is understood, the disagreement stops being an accusation and becomes a definition problem, which is the kind that can actually be worked on.

Three counts of the same month

What follows is a clearly labelled hypothetical, written for this article. No figure in it is observed, and any number appearing here should be read as illustration, not as evidence.

An air-conditioning service company runs paid ads through a summer season. Three parts of the business report on the same period, and each produces a different figure.

The platform counts conversions. Its rule is that someone saw or clicked an ad, and then completed the form on the website within the attribution window the platform applies by default. That window may be days long, and it may include views without a click. Under those rules the count is correct.

The CRM counts enquiries. Its rule is that a contact record was created. It merges duplicates by phone number, so a customer who submits the form twice appears once, and a household where two people enquire may appear once. It also holds enquiries that arrived by phone, by referral and through the map listing, with no channel field filled in for most of them. Under those rules the count is also correct.

Finance counts invoiced jobs. Its rule is that work was completed and billed in the period. That excludes every enquiry still in the pipeline, includes jobs booked two months earlier from any source, and quietly drops the enquiries that were quoted and lost. Correct again, under its own rules.

Three departments, three honest counts, three genuinely different questions being answered. The gaps between them are the arithmetic of those definitions, and they will reappear every month until someone maps them.

The Three Counts

The practice is to put the three figures beside each other for a fixed period and, for each gap, name the rule that produces it.

Between the platform and the CRM, common causes include the attribution window, view-through counting and deduplication. Someone should be able to say which of those explains most of the gap, and roughly how much of it remains unexplained after the known rules are accounted for.

Between the CRM and finance, common causes include lag, lost opportunities and work arriving through other channels. Enquiries that have yet to convert, enquiries that were quoted and refused, and invoiced work that reached the company by another route entirely.

That is The Three Counts, and the output is a short list of named causes with a residual at the end. The residual is the useful part. A gap you can explain is a definitional artefact and needs no action beyond understanding. Writing it down once converts it from a monthly puzzle into a known property of the reporting. A gap that remains after the known rules are accounted for is the part worth investigating for missing data, changed configuration or broken measurement.

Which gaps are worth acting on

Many gaps will prove structural rather than erroneous. The correct response is to document the rule and stop reopening the same disagreement each month.

A few are worth work. When the residual is large relative to the counts around it, when it moves sharply from one period to the next, or when it points in a direction the known rules would push the other way, something in the measurement has changed and finding out what is a real task.

The distinction matters because the alternative is a monthly argument in which each party defends its own figure, which produces heat and leaves the definitions exactly where they were. Reconciling is a different activity from disputing, and it has an output.

Why this is a standing job

The temptation after a first reconciliation is to treat the map as finished. It rarely stays finished, for reasons that have little to do with anyone’s diligence.

Attribution defaults change on the platform side, sometimes announced and sometimes noticed later. Deduplication rules change when a CRM is configured, or when someone tidies the contact database. Channels are added. A form gets a new field. Each of those quietly moves one of the three counts relative to the others, and the map made in March describes a system that no longer exists by September.

Handled as a recurring review, this takes a short meeting and produces a running record of which rules explain which gaps. Handled as a one-off, the map decays silently and the monthly argument returns with no one quite sure why.

An agenda for the review

Set a recurring meeting with the same four items each time, and keep the previous month’s answers visible so changes are obvious.

First, the three figures for the period, from their own systems, with no adjustment. Second, the named causes for each gap, carried forward from last time and amended where a rule has changed. Third, the residual after those causes, and whether it moved. Fourth, one item: anything either system changed since the last review — a setting, a field, a merge rule, a new channel.

The fourth item is the one that earns the meeting. Most of the surprises in this work arrive as a configuration change no one mentioned, and a standing question about it converts a mystery in three months into a note today.

Run it once with whatever three figures you already have. The first pass will be untidy, and several gaps may resist explanation. That is useful information. Run the review before the next budget discussion, so the decision begins with definitions everyone can see rather than a figure only one department recognises.

Before you bolt on another tool, it is worth knowing whether your business runs on systems or on you. I put together a free 2-minute assessment that gives you a straight read on exactly that, and the first thing to fix. Take the free assessment.