The Ask That Lived In The Copy

A Brief Diagnostic was booked through this website in early August. Google Ads and LinkedIn both reported zero conversions for the period.

Both are true. The event that fires when someone submits a booking was never marked as a key event, so nothing counted it.

Our homepage puts a standard to every brief that reaches us: agree the number before designing the room, one metric, in writing, before a single creative decision. We did not do it for our own marketing.

On 28 August we said we would publish what proper tracking showed, unflattering parts included. Before the fixed numbers arrive, here is what the broken ones showed. Everything below is Mochi's own first-party data, 1–10 August 2026. One caveat: 10 August was Singapore's observed National Day, which depresses that day's organic figures.

No instrument read wrong. Every tool did what it had been configured to do.

And the failures are not varied. They have one shape. What we wanted was written down clearly, in English, somewhere no system could act on it — a campaign name, a line of ad copy, a plan. Beside it sat a machine-readable field saying something else. The field won. It always does. Nothing ever contradicted us, because what we had written was not in a form anything could disagree with.

The campaign was named after what we wanted

Two LinkedIn campaigns ran from 3 August. One was named Follower Ads. Both were set to a brand awareness objective, bidding to maximise impressions, priced per thousand impressions.

The name is the ask. The objective is the instruction. Only one is executable.

That configuration buys impressions. It cannot buy follows — nothing in it treats a follow as the thing to get more of. The request existed in exactly two places: the campaign name and a line of ad copy. LinkedIn reads neither as an objective.

The brief that commissioned it did name a target, and named it properly: follows gained of at least forty, sixty or better. That target lived in the brief, and nowhere the platform could act on. Attributed follows: zero.

One caveat travels with that zero. LinkedIn credits a follow only when it happens inside the ad unit, so someone who clicks through and follows later is generally uncounted. Zero attributed follows is not proof of zero followers gained — only that the number the campaign was optimised against was zero, and that the target in the brief has no evidence behind it either way.

The destination field is the other machine-readable statement of purpose, and ours disagreed too. All three creatives point at our LinkedIn company page rather than the website — correct for a follower campaign, and also the reason paid LinkedIn cannot produce a site session, a blog read or a booking at all, while internally it was discussed as though it could. Nobody argued for that arrangement; the URL field settled it, unread.

Nor was anyone reading the account. One of the three creatives had been rejected by ad review and served nothing afterwards. No alert reached us, and a three-way creative test ran as a two-way test for its whole life.

The outcome nobody wrote into a field

The booking submission was not marked as a key event. Two of its neighbours were: the event that fires when someone completes the intake questions, and the one that fires when they click through to book. The account was not un-instrumented. It was instrumented around the outcome instead of on it, which is why the failure is easy to miss — either neighbour looks like proof the funnel is watched.

The consequence was total. GA4 counted no conversion, and Google Ads and LinkedIn had no key event to import. Every optimisation system in the stack was told nothing worked in a period when something did.

Everyone involved could have told you what success looked like. It was in the plan and in the copy. What did not exist was a checkbox.

That gap decides what any program can later prove. A brief is prose: read once, by people, before work starts. The things that report are fields — the run-of-show slot, the registration question, the CRM source field, the scan at the door. If the outcome has not reached one of those before the doors open, the report gets built from whatever the fields held. Which is attendance and satisfaction: the only things still collectable once it is too late to collect anything better.

When the machine does read your prose

There is one place in this stack where a machine will read your prose. It does not read it the way you meant it.

A Google Ads search campaign ran during those ten days on a keyword set that was 100% broad match — the setting that asks Google to interpret intent rather than match text. It is an instruction to fill in blanks, and it obliges.

The search terms report is the finding. The top-spending queries: "evening entertainment for corporate events", "corporate entertainment", "conference emcee" and "events services". Further down the same list: a design tool by name, a competitor by name, a query in Chinese, and a cluster of workout searches — zumba, barre, HIIT — which broad match appears to have read out of the phrase "wellness retreats".

Not one of them describes an organisation looking to commission measurable experience design.

That report is a continuous third-party description of what a machine believes Mochi sells. Nobody opened it.

The account said it a second way. The clicks arrived, and none of them converted. Most of the site's traffic that period landed on the homepage. Someone searching for a conference emcee reaches a page about experience design, does not find an emcee, and leaves. Average quality score across the set was 2.8, and landing page experience is one of its three components — a scored field reporting, continuously, that the query and the page did not match.

Leave an intention in prose and something downstream will specify it on its own reading, then bill you for the interpretation.

The apparatus that outran the instrument

A dashboard was built for this campaign with 24 columns. Exactly one could be populated from the data actually being collected.

A column header looks like a field and is not one. It is a question you intend to answer, and a reporting layout cannot create the instrument that answers it. Where nothing existed underneath, the column stayed empty and the intention stayed decorative.

The same absence shows across the stack. The content engine points at the blog, paid search at the homepage, paid LinkedIn at neither. Three engines, three destinations. That they were meant to be one funnel was written down — in a plan. No field had been told.

Configuration facts need no sample size

Our Impact Measurement page says that at forty guests there is no statistical significance, and anyone telling you otherwise is selling. That applies here, against us, and it disqualifies more of this period than it saves. Ten days is not a test. We cannot say Google Ads produced that booking, or that anything else did. Nor can the organic posts on our company page be ranked against one another — the differences between them are too small to separate a post that worked from a post that got lucky, and reading a preference into it would be the same error, one scale down. That is not a finding.

But none of the findings above are results. They are properties of the setup — true on day one, before any volume existed, readable by anyone who opened the account. A booking event that is not a key event will never be counted, at any volume. A campaign set to buy impressions will never optimise for follows, at any budget. Broad match will buy interpreted traffic at any scale. None of those statements needs a sample. Nobody opened the account.

Writing an outcome into a field is the same administrative act at every scale. Marking a key event takes about a minute; naming the CRM field that records source takes a line in a brief. Neither gets harder as the program gets bigger. What scales is the cost of finding out late, because that cost is the whole evidentiary record of the period, and it cannot be reconstructed once the room empties. Over one short campaign, the record lost is thin. On a year of programs defended in a budget review, the loss is the argument itself.

The test is short enough to run this afternoon, on anything already in your calendar. Point at the field where the outcome is written. Not the paragraph — the field.

Impact measurement is not a practice sold alongside the four. It is the operating system underneath all of them, which is why getting it wrong on our own marketing is the relevant test rather than an embarrassing aside.

The booking event is now a key event in GA4, created on 10 August. The rest of the remediation is written down and not yet applied — which is, precisely, the condition this essay is about. What we fix, we'll publish in October.

Point at the field

If something is already in your calendar and nobody has written what would count as it working into a field rather than a paragraph, take the Brief Diagnostic.

The Brief Diagnostic is a free 30-minute conversation. We will tell you what your setup can prove, and what it cannot.

No pitch. No deck. Just the questions we would ask anyway.