A campaign launches. The purchases start rolling in. Everyone in the room relaxes — the number went up, the ad worked, ship the next one. That instinct is exactly the trap Omar Eddaoudi warns against: if you only ever look at purchases, you're managing a business by looking at the one metric that lags every other signal by weeks, and you'll find out something went wrong only after it's already gone wrong.1 Purchases are the scoreboard at the end of the game. They tell you almost nothing about how the game is actually being played while it's still underway.
Before the purchase, there's a whole chain of smaller signals that predict it: hook rate (did the ad stop the scroll at all), video hold rate (did people keep watching once stopped), click-through rate (did the interest convert to intent).2 Each of these measures a different failure point. A campaign can have a strong hook rate and a weak hold rate — meaning the opening promise didn't match what came after, and the audience is dropping off mid-story. A campaign can have strong hold and weak click-through — meaning people were engaged but never given a reason to act. Purchases alone collapse all of that diagnostic detail into a single pass/fail number, which means when purchases dip, you have no idea which upstream link actually broke.
Eddaoudi pushes the frame further than most performance-marketing advice does: he wants cohort analysis — are people who bought once actually coming back — because repeat behavior is the strongest available signal that the brand identity, not just the individual ad, is landing.3 This matters especially in fashion and lifestyle categories, where a single purchase proves the ad worked but says nothing about whether the brand worked. A customer can be talked into buying once by a clever hook and never think about you again. A customer who comes back on their own is telling you the storytelling and identity actually stuck.
The sharpest version of the argument: "how is your brand actually scaling, your brand equity? Are you converting new people into the ecosystem?"4 This reframes growth itself. Revenue can grow while brand equity quietly erodes — you can buy short-term purchases with discounting and aggressive targeting while the underlying asset (how much people trust and identify with the brand) is flat or declining. Revenue is the lagging output. Brand equity is closer to the actual machine generating it, and it's the machine that a metrics dashboard built only around purchases will never show you breaking down.
You're reviewing last month's campaign results with a client. Purchases are up 15% and the room wants to call it a win and move on. You pull the funnel apart instead: hook rate is flat, hold rate dropped 6%, click-through is up. What actually happened is the ads got worse at holding attention but the offer got more aggressive, and the aggressive offer is masking the attention problem in the topline number. If you'd only looked at purchases, you'd have congratulated the team on exactly the wrong thing and let the real problem — content that's losing people mid-watch — keep degrading, invisibly, until the discounting stops working too.
You check one more number nobody asked for: are April's buyers still buying in June, or was April a one-time discount-driven spike? The cohort curve answers a question purchases alone never could — whether you built a customer or just closed a sale. From now on, before any campaign gets called a success, you walk the whole chain: hook, hold, click, purchase, return. A win at every earlier stage with a purchase dip somewhere upstream tells you exactly where to fix it. A purchase win sitting on top of a collapsing hook rate tells you the win is borrowed against a future you haven't paid for yet.
Every metric in a marketing funnel sits at a different distance from the decision that produced it. Hook rate is nearly instantaneous — it reflects whether the first three seconds of an ad worked, today. Hold rate reflects the next ten. Click-through reflects the offer. Purchase reflects all of the above, plus price sensitivity, plus competitive context, plus whatever mood the buyer happened to be in — compressed into one binary outcome that arrives days or weeks after the creative decisions that actually shaped it.2 The lag is the problem: by the time purchases move, the specific upstream failure that caused the move has already been made, shipped, and is sitting three or four production cycles in the past, which means a team managing only by purchases is always diagnosing symptoms that appeared weeks after their actual cause.
A purchase number can be flat, rising, or falling for reasons that have nothing to do with brand health, which is exactly what makes it a dangerous single source of truth. Watch for purchases holding steady while discounting quietly increases — that's not stability, that's the brand paying more to hold the same ground. Watch for purchases rising while hold rate falls — that's often aggressive offers compensating for weakening creative, a trade you're making without deciding to make it. And watch, especially, for a team that can tell you last month's purchase number instantly but can't tell you last month's hook rate without a special request — that asymmetry in what gets checked reflexively versus what gets looked up on demand tells you which number the organization actually manages by, regardless of what the dashboard nominally includes.
Cohort return is the sharpest of Eddaoudi's metrics precisely because it's the hardest one to fake with spend.3 You can buy a purchase with a big enough discount or a hard enough retargeting push regardless of whether the brand identity landed. You cannot buy a second, unprompted purchase the same way — a customer who comes back without being re-targeted, re-discounted, or re-convinced is reporting something no amount of ad spend can manufacture: that the first experience was good enough, and the identity compelling enough, to be worth repeating on its own. This is why the page treats cohort return as closer to the "actual machine" than revenue — revenue can be produced by force; voluntary return can only be earned.
None of this is free to implement, and the honest caveat belongs here rather than only in the Evidence section: full-funnel-plus-cohort tracking requires enough transaction volume and enough time elapsed to generate a meaningful cohort curve, plus the tooling and discipline to actually look at hook and hold rate as routinely as purchases. A brand-new operator with three months of sales data and a handful of ad variants doesn't yet have a cohort to analyze — the framework describes a discipline that becomes available, and valuable, at a specific stage of maturity, not a starting requirement for anyone running their first campaign.
Eddaoudi cites his own agency's track record (Slate Swim taken from zero to seven figures) as the evidence base — this is [POPULAR SOURCE] practitioner testimony, not independently audited data, and should be weighted as one operator's reported experience rather than an industry-wide finding. The genuine tension: a small or resource-constrained operator may not have the volume to run meaningful cohort analysis at all, making this advice more actionable for an established brand with real customer history than for someone just starting. The open question the source doesn't answer: at what stage of a business's life does the shift from purchase-only tracking to full-funnel-plus-cohort tracking actually pay for the added complexity?
This directly complements Misaligned Optimization Metrics (same batch) as the deliberate antidote to the trap that page documents. Where the Eurostar case shows an organization accidentally narrowing its attention onto one legible metric and losing the customer experience it doesn't measure, this page documents the opposite discipline: consciously building a wider metric set — hook, hold, click, purchase, retention — specifically so no single number can quietly become the whole picture. Read together they form a diagnosis-and-treatment pair.
Business — On-Brand Latitude Testing Matrix (same batch). That page documents how much creative variance a brand can test before it drifts off-identity; this page documents how you'd actually know if that drift happened — a falling hook or hold rate on new creative variants is the early-warning signal that a test has wandered outside the brand's latitude before purchases even move. The insight neither page reaches alone: creative testing and metric depth are two halves of the same discipline — testing without funnel-depth metrics is flying blind about whether variance stayed on-brand; funnel metrics without deliberate testing latitude have nothing systematic to diagnose.
Behavioral-mechanics — Goal Dilution Effect. Goal dilution shows how adding weak proxy goals reduces motivation toward a real target; this page is the inverse discipline — deliberately adding more measurement points along the real causal chain (not weak proxies, but genuine upstream signals of the same outcome) to prevent any single distorted metric from dominating attention. The distinction the pairing sharpens: not all additional metrics dilute focus — metrics that sit on the actual causal path to the outcome (hook→hold→click→purchase→retention) concentrate understanding, while metrics unrelated to that path (vanity metrics, off-target proxies) are what actually dilutes.
Sharpest implication: purchases are the metric furthest downstream from the truth, which makes them the most reassuring number to look at and the least useful one to manage by — by the time it moves, everything that actually caused the move already happened, unmeasured, weeks earlier.
Generative questions: