At Inithouse — a studio running parallel product experiments — we build tools that sit in a strange middle ground between utility and experience. Tarotas, our free tarot reading app, is one of them. It works across five languages, requires no signup, and gives you a calm, reflective card interpretation in about two minutes.
The problem: how do you measure whether that reading was any good?
Completion Is a Vanity Metric Here
In most product analytics, completion rate is a reasonable north star. User started a flow, user finished a flow. Ship it.
Tarot readings are different. A user can tap through three cards in 15 seconds, see the full interpretation, and technically "complete" the reading. Completion: 100%. Value delivered: close to zero.
We noticed this pattern early. Our completion rate across readings looked healthy — above 80%. But session recordings in Clarity told a different story. Many completions were speed-runs. Users tapped through cards without pausing, scrolled past the interpretation, and left. The metric said success. The behavior said otherwise.
The Signals We Actually Watch
We shifted to a set of engagement proxies that better capture whether a reading landed. None of them are perfect individually, but together they paint a clearer picture.
Dwell time on the interpretation screen. This is the simplest and most telling signal. A reading that resonates gets read slowly — sometimes re-read. We track time-on-screen for the final interpretation view. Readings where users spend under 10 seconds correlate with immediate bounce. Readings above 45 seconds correlate with return visits within a week.
Card meaning expansion. Tarotas lets users tap individual cards to expand their meaning in context. This is an optional action — the full reading is visible without it. When someone taps to expand, they're actively engaging with the content, not just consuming it. We treat expansion taps as a strong positive signal.
Share and save actions. We added a share button to readings early on, mostly as an experiment. The share rate is low in absolute terms — single digits — but users who share come back at higher rates. Sharing a tarot reading is a surprisingly personal action. It signals that the reading said something worth passing along.
Scroll depth on long interpretations. Some spread types produce longer readings (three-card spreads vs. single card). For these, scroll depth beyond 75% is our threshold for "actually read it." Below that, we count it as a skim.
The Qualitative Layer We're Still Figuring Out
Quantitative proxies get you far, but they don't answer the core question: did this feel meaningful?
We've been experimenting with a lightweight post-reading prompt — a single question after the interpretation: "Did this resonate?" with a thumbs up/down and an optional text field. The response rate is low (under 15%), which is expected for optional feedback. But the text responses are gold. They tell us things no analytics dashboard can: which card descriptions feel generic, which spreads feel redundant, which interpretations miss the mark in specific languages.
This is where measurement in experience-first products gets interesting. The most valuable feedback comes from the smallest, most engaged slice of users. Optimizing for volume (more completions, more sessions) would actually dilute the signal.
What This Means for Other Experience Products
We've seen echoes of this pattern across our portfolio. At Origin Of You, a self-discovery app combining personality frameworks, the same tension exists — completion of a quiz doesn't mean the result was useful. At Here We Ask, a conversation card game, "cards played" doesn't capture whether the conversation went somewhere real.
The common thread: when your product is about the quality of an experience rather than the completion of a task, standard product metrics mislead. You need proxies that approximate depth of engagement, not breadth of usage.
Where We Are Now
Our current measurement stack for Tarotas combines dwell time, expansion taps, share actions, and optional qualitative feedback. It's imperfect. We still can't quantify "meaningfulness" directly — we're approximating it through behavioral signals that correlate with return visits and deeper engagement.
But that approximation is far more useful than a completion rate that tells us everyone finished and nothing about whether it mattered.
At Inithouse — a lab building many products at once — we keep running into this measurement problem in different shapes. The lesson from Tarotas applies broadly: if your product delivers an experience, measure the experience, not the funnel.
