What good looks like: how to set digital marketing KPIs that actually measure success
"How do we measure success?" is a question I'm often asked. The expected answer is a single number: a target ROI, a magic ROAS, the one KPI that tells you if it's working. The real answer is that there is no single metric, and chasing one leads good programs to make bad decisions.
Success looks different at every funnel stage because each has a different job. Judging them all by the same bottom-line metric will quietly defund the activity that feeds your conversions. Below is the framework I use to set KPIs across a full-funnel programme and how I read the numbers afterwards, since no KPI tells the whole truth alone.
Start with the job, not the metric
The most common measurement mistake I see is holding upper-funnel activity to lower-funnel standards—expecting an awareness campaign to post a direct return and cutting it when it doesn't. It's like weighing yourself to see if you're getting taller.
Each funnel stage exists to do a specific job, and the right KPI is the best measure of that job done well. Awareness exists to be seen by the right people. Conversion exists to drive income. Asking the same question of both makes no sense. Before setting a target, get clear on what each activity layer is for and then measure accordingly. That logic carries through the framework.
Here's how that maps out in practice.
Awareness Reach the right cold audiences
Engagement Earn attention, pull people in
Consideration Build intent and frequency
Lead generation Capture data and permission
Conversion Drive income
Allow me to walk through the logic because the nuance is where the value lies. I'll move stage by stage to keep transitions clear.
Awareness is about buying attention efficiently from cold, well-chosen audiences. I lead on CPM, with reach and impressions supporting. Channels like connected TV, YouTube, display, and Amazon do the heavy lifting. Don't ask this layer for income. Ask if you reached the right people at a sensible cost.
Engagement is where you show if anyone actually cares. I watch video completions and site visits as the primary signals—did people choose to lean in?—with wider engagement metrics in the background. A three-second view is a trace of interest; a video completion is a minor act of consent worth retargeting. This messy, important middle turns mild interest into authentic intent and builds the frequency that gets you remembered. This is where retargeting earns its keep: impression retargeting to lift frequency, video-completion retargeting to re-engage the warm. It's also where non-brand search starts catching people as they begin to look and, increasingly, where newer answer-engine placements live. The right measures here aren't financial. They are quality-of-engagement signals: engaged session rate and the rate at which people take non-financial micro-actions that show they're moving closer. From here, the next step is to generate leads.
Lead generation is about obtaining data and permission to nurture people over time. CPL is the headline but on its own it's a trap—the most useful thing I can tell you about this stage. A low CPL is worthless if leads don't convert or if you can't scale it. So I always read CPL alongside response rate (to qualify lead quality) and volume (to check scalability). A quirk: on Meta lead-gen forms, broad, almost "open" targeting often outperforms tightly defined audiences on cost, though quality isn't always as high. That's why you can't judge lead gen on CPL alone.
Conversion is the only stage genuinely all about money. ROI (or ROAS) is the primary measure, supported by conversion rate and average gift. This is the home of brand search, donate-form abandonment retargeting, and low-funnel social—the warmest audiences captured at the moment of decision. It's also the stage last-click attribution tends to over-reward. This brings me to the part most KPI conversations skip.
No KPI is exhaustive — and your attribution model is quietly choosing your winners.
Here's the uncomfortable reality under every dashboard: the numbers depend entirely on the attribution model that generated them. Change the model and you change which channels look like heroes and which look like dead weight without moving a single pound of spend.
Last-click attribution gives all credit to the final touch before conversion. It's clean, simple and badly biased toward the bottom of the funnel. It lavishes credit on brand search and retargeting while making awareness and engagement work look worthless. Useful, but only half the story.
Data-driven attribution spreads credit across touchpoints based on observed contribution and improves on last-click attribution. But it's an algorithmic black box and still relies on user-level tracking that privacy changes and signal loss steadily erode.
Then there's the wider family of models — first-click (over-credits the top), linear (splits evenly), time-decay (more credit the closer to conversion), position-based or U-shaped (rewards the first and last touch). Each is an alternative lens, and each tells a slightly different story about the same campaign. Beyond click-based models entirely sit marketing mix modelling (a top-down, statistical read of how spend drives outcomes that doesn't depend on tracking individuals — far more resilient in a privacy-first world) and incrementality testing (geo-holdouts and the like, which get closest to the only question that really matters: did this spend cause income that wouldn't have happened anyway?).
My practice is to never trust one model. I look at three lenses in parallel: usually data-driven, last-click, and a holistic model I build by hand. I triangulate between them. Last-click shows the floor; the holistic view shows the fuller picture; data-driven sits in between. The discipline isn't picking the "right" model but refusing to be governed by any single one.
Why this matters more as we move to a PMax-style world
All this is becoming more important given where platforms are headed. Performance Max, Demand Gen, and the new wave of AI-search ad products (OpenAI opened ChatGPT advertising to US businesses in 2026, with other markets following) share a direction: the machine takes over more targeting, spans more of the funnel, and gives you fewer granular levers. You buy an outcome and let the algorithm decide where, when, and to whom your money is spent.
That has real consequences for measurement. When a single PMax campaign operates across awareness, consideration, and conversion at once, neat channel-by-channel last-click reporting stops describing reality. The more buying consolidates into AI-run black boxes, the more you need two things: stage-appropriate KPIs that respect each layer's purpose and a holistic, incrementality-minded read of whether the whole system grows your income. I wrote more about this shift and what AI search means for purpose-driven organisations on our SEO, AI Search & Content Strategy page. It's the same story: less manual control, more need for genuine measurement maturity.
So, what does "good" actually look like?
Not a number. A framework.
Good looks like a measurement approach in which every activity stage is judged by its job—CPM at the top, engagement signals in the middle, CPL where you capture data, ROI where you drive income. It means reading those numbers through multiple attribution lenses to triangulate the truth rather than being governed by your platform's default model. Increasingly, it means orienting the whole toward incremental growth—proving your spend causes income, not just sits in front of conversions that would have happened anyway. That is how the stages fit together.
The organisations that win at this aren't the ones with the most impressive single ROAS figure to wave at the board. They're the ones who know what good looks like at every step and can tell the difference between activity that's genuinely working and activity that merely looks busy.
If you'd like help building a measurement framework or KPI scorecard that does this properly — one your team and your leadership can actually align around — drop us a line. We’ll help you make it work in practice.
