Does engagement stay a KPI?

The question

If the system works well the interaction gets shorter. Do we still want to measure time spent, sessions, pages?

Why it is a decision, and why it is the most uncomfortable one

Because engagement is not a neutral metric: it is a metric that selects product behaviours. The corpus shows this from three different angles that all point the same way.

• “Engagement metrics reward systems that create attachment” — and the taxonomy of compulsive use names the design mechanisms that feed it: systematic agreeableness, persona personalisation, immediate gratification.
• Product metrics “reward silence”, that is, they reward not disclosing.
• Onboarding, tours and wizards “should be considered a cost”: time spent understanding the product is not value.

Put together: if engagement stays the KPI, the product will optimise towards attachment, reticence and dwell time. Three things the rest of this site treats as damage.

The cost of either branch

Abandoning it means being left without a daily, legible, shared indicator, and replacing it with slower ones — first-attempt success, conversion, qualification, return.

Keeping it means that the decisions on trust calibration, on the delegation scale and on empathetic register will be lost one at a time, each against a weekly number.

Not deciding means engagement stays the KPI by inertia, and keeps selecting product behaviours without the organisation ever declaring which outcomes it actually wants to measure.

Criterion for deciding

Replace engagement only if the organisation accepts slower, less legible indicators. Use first-attempt success and the four behavioural trust indicators to judge whether the product helps the user — not just how long it keeps them.

What the literature says

The corpus offers two explicit candidates: first-attempt success, which the phenomenon on the learning curve proposes in place of learnability; and the four behavioural trust indicators — cancellation after an automatic action, external verification of the output, drop-off at confirmation points, switching automation off after a trial.

Both candidates find a precise home in Google’s HEART framework (Kerry Rodden, Hilary Hutchinson and Xin Fu, ACM CHI 2010): five categories of UX metrics — Happiness, Engagement, Adoption, Retention, Task Success — built to replace small-scale qualitative observation with quantitative measures usable on mass-market products. First-attempt success, in this scheme, is a case of Task Success; the four trust indicators sit closer to Happiness (actual, not declared, satisfaction) and to Retention read in reverse, where switching off automation signals lost trust rather than lost interest. HEART does not resolve this card’s decision by itself — its own authors insist the five categories must be chosen by product purpose, not applied all at once by default — but the process that comes with it, Goals-Signals-Metrics, is directly reusable here: for each relevant category, write the business goal in one sentence, then the observable behavioural signals that indicate progress towards it, and only then the precise metric to instrument. Applied to this decision, the process forces an explicit answer to the question this card asks: is the objective retention (Engagement) or the useful outcome (Task Success)? The two categories exist to be chosen separately, not summed into a single number.

Neither candidate is a full substitute for engagement, and this has to be said: they are slower, less legible and they do not answer the question “how did this week go?”. Whoever decides to abandon engagement accepts no longer having that answer.

Who decides

This is not a product decision: it is a decision about how the company judges the product. It goes where objectives are approved. If it stays with product, product will defend every individual choice against its own KPI, and lose.

Where it falls on roles

Engagement is not simply a number someone looks at: it is the number against which every other choice is defended. Keeping or replacing it changes the work of those who produce it, are measured by it and approve it.

• Leadership & Strategy — Approves the objectives and must fund work assessed over quarters rather than weeks.
• Operations & Data — Defines, collects and explains the replacement measures.
• Marketing & Growth — Open rate and time on site lose reliability while AI reads, summarises and filters messages before the user. Keeping engagement leaves reporting legible but increasingly disconnected from outcomes; replacing it requires renegotiating what the channel promises to budget owners.
• Product & Design — Pays the cost of a decision made elsewhere: confirmation points, uncertainty signals and shorter paths must be defended against a weekly number that penalises them. The same mechanism makes the delegation scale implicit.
• Research & Information Architecture — Supplies research measures rather than analytics: first-attempt success, time to first useful result, and trust observed in behaviour rather than declared. If engagement remains, this work has no commissioner.

Related phenomena

• Empathy and social presence — Shows what engagement rewards most directly.
• Trust, AI literacy and calibration — Provides concrete replacement indicators.
• Zero learning curve — Supplies first-attempt success as an alternative.
• The customer journey shortens — Explains why sessions can fall while the system works.
• Cognitive load and wellbeing — Poses the same question from the user’s side: if engagement measures dwell time, the product gets pushed towards response speed rather than the understanding a user retains.
• Memory and context adaptation — A system that remembers shortens interaction by definition.