The cost of three dashboards

Three dashboards do not produce three views of one account. They produce three accounts, each with its own vocabulary, and every question that spans them requires a translation that nobody wrote down.

A conversion in one is not a conversion in another: each counts what it can claim, over its own window, with its own definition of the event. Reach is de-duplicated within a platform and cannot be added across them. Money is in whatever currency each account bills in, so the totals are not addable even when the definitions agree. A day starts at different moments depending on each account's time zone.

So the sum at the bottom of the spreadsheet is not a total. It is a number that would only be a total if four things were true, and the person who built the sheet knows which four and the person reading it does not.

In somebody's head, and that is the part that costs.

Every organisation running several platforms has one person who knows that the numbers from one source run higher than the other, that a particular campaign's conversions are double-counted, that the figures before a certain date use a different attribution window, and that a particular week is unreliable because a sync failed. None of it is written anywhere. All of it is load-bearing.

The bill arrives in three ways. New people make decisions on numbers they have read correctly and understood wrongly. The person who holds the translation cannot take a holiday during a reporting period without the reporting quietly degrading. And when that person leaves, the organisation does not know what it no longer knows — the numbers keep being produced, they keep looking the same, and the corrections stop being applied.

Putting three charts on one page does not fix any of this. It makes the incomparable numbers adjacent, which is worse, because adjacency implies comparability.

What helps is normalising the grain before anything is aggregated: one row per entity per day, holding counts and money only, with the account's own currency stamped on the row and the platform recorded. No rates stored, because a stored rate can drift from its own inputs. Nothing converted on the way in.

Then the rules about combination have to be explicit rather than assumed. A scope covering one currency is exact and nothing is converted. A scope covering several is converted when read, labelled an estimate, and stamped with the date of the oldest rate involved. Where a rate is missing nothing is invented and the total is refused. A list that cannot be ranked comparably comes back in raw order with that fact attached.

None of that is glamorous, and all of it is the difference between a consolidated view and three views in a trench coat.

It should not decide whose conversion count is right.

That distinction matters because it is where reporting tools usually overreach. Normalising the shape of the data — the grain, the units, the currency, the clock — is a mechanical problem with a correct answer. Deciding which platform deserves credit for a sale that three of them touched is a judgement about your business, your margins and your tolerance for double-counting, and it has no correct answer that a tool can compute.

A system that quietly picks one has replaced a problem you knew you had with one you do not. The honest arrangement presents the parts, presents the blend with its caveats attached, and leaves the attribution judgement where it belongs.

Reading is only half of the scatter. The other half is that every platform has its own way of making a change, its own idea of what a budget attaches to, its own answer to whether a deletion can be undone, and its own opinion about whether a change can be tested before it is made.

Consolidating that means normalising the entities every platform has in common and being explicit about what each platform can actually express, rather than pretending they are the same and discovering the difference at the worst moment. Each connection carries its own statement of what it supports, which is the honest alternative to a lowest-common-denominator interface that fails silently.

That one shape holds on the writing side too, whichever platform is the destination. Reading is what a connection does. A change is written down before it happens, with the value it would replace and the checks it cleared, and it waits for an authority — you, or a rule you wrote. Beyond that, no platform write reaches a customer account until that platform has granted write access; the applying step is in private beta.

One record of what was decided beats three histories in three interfaces, for the same reason one normalised table beats three exports: not because it is faster, but because it is the only version anybody can check.

Not hours. Hours are what people report, because hours are easy to notice.

The saving is that the translation stops living in a head. When the grain is normalised and the combination rules are stated, the reconciliation is a property of the system rather than an act of memory, and the number a new person reads is the number the experienced person would have read.

That is the whole argument, and it is an argument about institutional knowledge rather than about convenience.

The numbers your dashboard rounds off is what each of those three dashboards is doing on its own, and Why platform numbers disagree is how to sort the differences that matter from the ones that do not.