Product strategy from user data · August 2026
What 74,613 accounts reveal about reachability, friction, and the next AstroMatrix product.
Synastry is 23% of chart views and relationship material ~17% of all interpretation reading. 22.9% of accounts find the screen — a Home button and the Matrix switch. But it has zero contextual links: nothing in the app ever offers it at a moment of intent, which is why 63% open it once or twice and stop.
Accounts holding 8+ charts read 0.1–0.5 interpretations per chart. They are not content customers who lapsed — 97.9% were seen inside three months. They are running a workspace the product does not provide.
In their first three visits, 37.8% of a newcomer's activity is the home screen and 2.7% is an actual interpretation. 29.2% of those who stall never see a second screen.
AstroMatrix's biggest opportunities are not primarily missing features. They are places where demonstrated demand is obscured by navigation, operational friction, or surfaces that fail to deliver the value already inside the product.
The strongest examples are relationship astrology, chart management, first-session personalisation, and several finished features users can barely reach.
Usage measures our design, not their appetite. A low number tells us a feature went unused; it cannot tell us whether it was unwanted or simply unreachable — and in this product it has repeatedly turned out to be the second.
Progressed charts sit under 0.4% of reading with no route of their own and ninth position in a ten-item selector. Relocation is behind a disabled flag. The weekly timeline and the synastry wheel have no entry point at all. None of these has ever had a fair test. Treat every low number as a hypothesis about the interface until it has been given one.
Four patterns in the data indicate confusion rather than disinterest. The first is the largest single problem found in this review; the last corrects an assumption made earlier in it.
Some of what looks like disengagement is repeated task failure.
5,090 accounts opened six or more screens and never read a single interpretation, averaging fifteen visits. A further 11,112 opened six or more and read three or fewer — 16,202 in total who explored broadly and consumed almost nothing.
People who return fifteen times and open six screens are not uninterested. They are trying to accomplish something and failing to reach it. Read alongside where support contact concentrates, that reframes a share of our engagement problem as an operational one: usage being consumed by tasks that do not complete.
Opening six screens across fifteen visits scores as high engagement on any conventional measure. In this cohort it means the opposite: repeated unsuccessful search. Any metric counting screens or sessions without checking whether the primary value was reached will read this population as thriving.
It is the behavioural counterpart of the governing principle. Usage measures our design; movement measures our navigation. Neither measures satisfaction.
Support contact is caused by chart management, not by content. Comparing accounts that reached the support screen against everyone else, the lift is entirely on the people-and-data screens:
| Screen | Support users | Everyone else | Lift |
|---|---|---|---|
| edituser | 19.7% | 4.6% | 4.33 |
| login | 5.9% | 1.9% | 3.17 |
| chartpickuser | 82.6% | 38.7% | 2.13 |
| profiles (add person) | 22.6% | 11.2% | 2.01 |
| component | 72.2% | 61.0% | 1.18 |
| matrix | 43.8% | 37.9% | 1.16 |
And they are not struggling beginners. Support users average 60 visits, 9.3 screens and 7.7 charts against 27, 5.6 and 5.1 for everyone else. The people writing in are the engaged ones with collections to manage — which is the operator workflow again, arriving through the support inbox.
The same pattern shows in birth-data churn: 355 accounts have opened the edit screen three or more times, they hold 7.3 charts against a fleet average of 5.1, and 12.1% of them contacted support — 5.8× the base rate. Repeatedly correcting birth data is a reliable precursor to writing in.
A metric worth naming: the gateway ratio. Opens of an index divided by opens of the thing behind it. Above 1 the screen is a corridor; below 1 it is handing off. It is comparable across screens and before and after a change, which makes it the natural measure for any of the navigation work here.
| Gateway → destination | Gateway ratio | Accounts above 2× | Reads as |
|---|---|---|---|
| tarot → reading | 2.67 | 55.2% | Corridor |
| journal → report | 2.00 | 44.8% | Corridor |
| home → component | 1.00 | 15.6% | Even |
| chartpickuser → chartview | 0.67 | 4.5% | Converts |
| facets → component | 0.50 | 5.5% | Converts |
| matrix → component | 0.44 | 1.6% | Converts |
| Surface | Diagnosis |
|---|---|
| Matrix | Fundamentally working — leave the hierarchy alone |
| Relationship | Under-surfaced |
| Chart management | High-friction |
| Tarot / Journal | Unnecessary hallway before value |
| Home | Weak first-value delivery, not bad navigation |
That distinction is worth holding onto: without it, months could go into simplifying a hierarchy that is already doing its job.
The Matrix hierarchy works, and an earlier draft of this review implied otherwise. People open the Matrix and its facets roughly half as often as they open the interpretation behind them — the index is doing its job and handing off. The same is true of the chart picker. Whatever is wrong with the newcomer experience, it is not that the Matrix fails to convert.
Tarot and journal are the genuine corridors: 2.67 and 2.00 openings per thing reached, with more than half and nearly half of accounts respectively above a 2× ratio.
Every astrology experience needs to answer three questions, and the product currently answers one and a half of them.
| Saturn conjunct Moon | What the product should say |
|---|---|
| Orientation | Saturn meeting your Moon can intensify emotional responsibility. |
| Navigation | This is one of your strongest transits this month. |
| Integration | Reflect on where you are carrying emotional responsibility that no longer belongs to you. |
The integration layer is already half-built.
Every interpretation carries three fields — concept, opportunity and pattern — and the component screen already renders all three. “Opportunity” is the application field. What this document has been calling a missing layer largely exists in the content model; it is presented as a third block of prose beside the other two, so it reads as filler rather than as the thing to do about this.
That makes integration far cheaper than it first appears — a presentation change, not a writing project — and it resolves into a UI rule:
The three needs map onto who has them. Orientation serves the 40.8% who visit less than monthly and the newcomers. Navigation serves the 46.8% in the middle. Integration serves the 12.4% who come weekly or more — which is why interest in practice tracks frequency (9.0% and 7.1% among frequent users against 2.9% and 2.7% among rare) rather than any particular mode.
The quickest way to locate a gap: ask what someone of that type would naturally go looking for, then set it against what they actually do. The difference tells you which of the three is missing.
| Mode | Would normally look up | What the numbers show | Gap |
|---|---|---|---|
| Newcomer | “What does my chart say about me?” | 37.8% of first-visit activity on home, 2.7% on an interpretation | Orientation |
| Self-explorer | Placements and aspects — then what's coming | Interpretations 88%, but progressed under 0.4% | Navigation |
| Chart operator | A client's chart, their transits, what was said last time | Chart views 100%, 0.1–0.5 readings per chart, no notes exist | Integration |
| Relationship | The two charts overlaid, the composite, the friction | Synastry 23% of chart views — wheel has no route | Navigation |
| Timing | What's active, is the moon void, which hour suits this | Daily 95%, but moon and calendars bounce inside ten seconds half the time | Orientation Integration |
| Tarot | A draw, and somewhere to put what it stirred | The reading holds 71 seconds; the entry loses 43% inside ten | Navigation |
There is no row where the thing does not exist. The failures cluster by mode rather than by feature — which is exactly why sectionalising is the right instrument and a bigger menu is not.
Relationship is found, used, and never offered.
Synastry is 23% of chart views and relationship material ~17% of interpretation reading, with Synastry Aspects the fourth-largest content category. 22.9% of accounts have opened the relationship screen — those figures agree, and an earlier draft here wrongly set the 23% against a 3% “dominant mode” figure to claim a six-fold gap. The real finding is narrower and more actionable: the screen has two links in and neither carries any state. Nothing that happens in the app produces the thought look at this relationship now.
The first experiments are cheap, because adding a person is the strongest predictor in the dataset: accounts that reach the add-person screen go on to do a relationship reading at 59.7% against 18.1%, and reach peaks at exactly two charts.
Partner added → offer the comparison immediately → persistent “You + Sarah” context → synastry, composite, and current transits to the relationship. That is a redesign of the second-person flow rather than a new feature, and 1,846 two-chart accounts have added someone and never read anything.
7,041 accounts hold eight or more charts and have never once compared any of them — and comparison gets rarer as collections grow, from 77% never at four charts to 86% never above fifty.
Sixteen charts is a hundred and twenty possible pairings. Asking someone to pick one is not a navigation problem, it is a discovery problem, and the answer is for the product to do the choosing:
“Three relationships stand out in your charts.”
For a first version, four rankings the engine can defend: strongest overall synastry · most supportive · most challenging · most currently activated. Richer framings — karmic, attraction, communication — imply scoring methodology we have not built; add them once people use the first four.
The chart-count paradox resolves once you stop assuming everyone opened the app for the same reason. These are jobs, not identities — a practitioner reads deeply sometimes, a self-explorer accumulates ten family charts. The same account moves between them, which is why the product should adapt to the job in front of it rather than classify the person permanently.
| The meaning job | The management job |
|---|---|
| “Help me understand this chart.” | “Help me manage these people, and tell me when something matters.” |
| 2–3 charts, 5.61 interpretations per chart, returns weekly or daily | 8.4 charts, 0.1–0.5 per chart, returns about 0.6 times a month |
| Matrix, facets, journal, reports | Searchable people list, pinning, last viewed, notes, session history, transit alerts, upcoming dates, export, sharing |
Fifteen thousand accounts are doing the management job with only the meaning job's tools. The content gate makes the mismatch concrete: five interpretations per session is one client reading.
The move that changes the economics is transit alerts across saved people. Today forty stored charts are forty items buried in a picker. With alerts they could become forty recurring reasons to open the app — “Mars squares Emma's natal Moon tomorrow.”
Two explanations are ruled out. Not the paywall — the median account views four interpretations in its life and the gate fires at five in a session. Not setup burden — accounts that stalled hold 5.2 charts against 6.3 for engaged ones. They entered birth data, added people, and did not come back.
And not an empty screen. Home renders eight panels of real astrological content, each tapping through to an interpretation. Three of its four groupings are time-based — what is moving today, the current journey, important events. But the fourth, Core Themes, is natal: the birth chart's most significant placements, drawn from a different filter, cached separately, and ordered last.
So the screen already contains something that answers “what does my chart say about me?” — it is simply placed fourth, behind three panels about a sky the newcomer has no reason to care about yet.
Lead with identity for the first three sessions. Lead with the sky after that.
A newcomer sees Core Themes first: what is constant in your chart. From the fourth session on, the current map returns to the top and identity drops back down. Same screen, same panels, same content pipeline — a reorder, not a redesign.
The session counter already exists and is already banded at 1 / 4 / 10 by the content gate, so there is no new state to introduce. And because the natal panel is cached independently of the daily forecast, it can paint before the current map resolves — making a newcomer's first screen faster as well as more relevant.
This also settles where a long-form personal synthesis belongs. Not here: Home's median dwell is 20 seconds, and a 200-word essay is the same failure as a wall of category labels, just better written. It belongs as the interpretation behind the top Core Theme, where dwell is 44 seconds and the surface supports a real read.
Two supporting fixes: the onboarding wizard exists and reaches 1.4% of the people who stall — a defect to investigate before it is a feature to design. And nothing should promote tarot, reports, calendars or moon at this stage; appetite for all four sits at 0.10–0.33 of baseline.
Still unanswered: we cannot say why they leave. The analytics hold counts and averages with no ordering or exit events, so “bounced in ten seconds” and “read for four minutes and never returned” are indistinguishable. An ordered first-session trail, or asking twenty of them, settles it cheaply. We have never asked.
Nothing here proposes rebuilding. Reading the current screens against the findings, most of what is needed is presentation of things that already exist. Graded by size of change across eight screens:
| Change | Count | Examples |
|---|---|---|
| Copy only | 5 | Panel titles that say why rather than what; chart-type labels; naming the picker's mode |
| New content in an existing slot | 5 | A lead statement above Home's panels; a generate action in the Reports tab; tarot → chart link |
| Reorder or change a default | 8 | Facet sort by significance; picker recency; the chart-type selector; component bottom bar |
| Contextual hint | 1 | One line explaining what a life area covers |
| Genuinely new screen | 1 | The per-person workspace for operators — notes, last seen, session history |
The integration layer is already half-built.
Every interpretation in the product carries three fields: concept, opportunity and pattern — and the component screen already renders all three. “Opportunity” is the application field. The layer we have described as missing largely exists in the content model; it is presented as a third block of prose beside the other two, so it reads as filler rather than as the thing to do about this.
That makes integration substantially cheaper than this document has been implying. For interpretation content it is a presentation change, not a writing project.
Two corrections that came from reading the screens rather than the data:
Home is not a menu, and it is not purely a current map. It renders eight panels of real content, and its fourth grouping — Core Themes — is drawn from natal rather than transit data. An earlier draft here recommended “put content on Home,” written without opening the file. The content is there, including the identity content; it is ordered last. That is a reorder, not a redesign.
The astrology-linked tarot is not a feature to build — it is a feature to surface. Component-to-spread matching is implemented, the reading opens pre-loaded with the placement as its question, and the content library holds 130 spreads, all active, around 117 of them named for specific transits. Yet of 20,000 sampled readings, only 1.9% ask an astrological question at all. The most common questions in the product are “How does my lover feel today?” and “What does my lover think about me?” — generic fortune-telling that any tarot app can answer. The one genuinely defensible thing here is built, populated, and almost nobody arrives through it. The link is one of seven equal actions on a bottom bar, and the tarot page opens with a free-text box that invites exactly the generic question people type.
Three of Home's six buttons pass context; three pass nothing. Birth Report, Weekly and Monthly arrive at the Matrix already filtered by category. Tarot Reading, Love Match and Reflections pass no parameters at all — and those are precisely the three features with usage problems. Love Match arrives at a relationship screen with no pair chosen; Reflections at a meditation configuration screen; Tarot at an empty question box. This is the smallest, most repeatable change in the review.
docs/product/screen-interventions.md in the server repository.
The roadmap above ordered by value across the product. This is the same work arranged by who it serves, with the friction findings folded in. Ship is backed by evidence in hand. Learn is where a decision is blocked on a measurement. Avoid is where the data argues against an intuitive move.
The widest gap between demand and design in the product.
Reaching the add-person screen predicts a relationship reading at 59.7% against 18.1% — and there is no link there. profiles has four contextual links pointing in and none pointing out. Highest demand-to-effort ratio in the review.
Reaching the add-person screen predicts a reading at 59.7% against 18.1%. 1,846 two-chart accounts added someone and read nothing.
~7% of interpretation reading and absent from the old data entirely. It currently rides behind synastry with no entry of its own.
7,041 accounts hold 8+ charts and have never compared any. Prototype “three relationships stand out” before committing to the full explorer — and cap collection size, since ranking is O(n²).
Reach peaks at exactly two and falls away above eleven. That shape is the feature's limit, not the audience's.
Not lapsed — 97.9% seen within three months. And now identified as the source of most support contact.
The only re-entry reason that scales with collection size. Forecast data runs to end of 2026; delivery already works.
Support contact lifts 4.33× on the edit screen, 3.17× on login and 2.13× on the picker — while content screens sit at 1.16–1.18. The friction generating support tickets is entirely here. Recency, pinning, search, and a birth-data entry flow that does not need three attempts.
0.10 readings per chart above fifty charts. They may want calculation and storage, in which case the content model and its per-session gate are the wrong product for them. Ask twenty before building either way.
All three below their own baseline. And five interpretations per session is one client reading.
Ruled out: the paywall (median four interpretations a lifetime, gate at five per session) and setup burden (stalled accounts hold 5.2 charts against 6.3 for engaged ones).
Not a link — the interpretation itself. 37.8% of first-visit activity is home; 2.7% is content. This replaces a menu with what the menu points at.
Five thousand accounts opened six or more screens and read nothing, at fifteen visits each. They are searching. Any screen they land on repeatedly should offer content, not another set of choices.
Stats only write once a profile resolves, so login, register and password-reset are invisible here — the 1,447 login accounts are survivors. Grafana counters fire regardless: compare usage_login_total_count against what 1,447 accounts should produce. A loop would show there and nowhere else.
“Bounced in ten seconds” and “read for four minutes and never returned” are opposite problems and currently indistinguishable. We have never asked.
They already hold 5.2 charts. Appetite for tarot, reports, calendars and moon sits at 0.10–0.33 of baseline. Breadth is what maturity looks like, not what causes it.
The deepest readers, and 95.9% of everyone who has viewed a report. The Matrix hierarchy already converts for them — index-to-content ratios of 0.44 and 0.50.
They are the report audience, and reports are generated in four screens but retrievable only through the journal — which is itself a corridor, opened twice for every report reached.
Journal separates this mode's thriving quarter from its struggling one, 39% against 71%. A prompt carrying the interpretation across, not a blank page.
Progressed is under 0.4% of reading with no route of its own and ninth place in a ten-item selector. Draconic outperforms it sixfold with no surfacing at all. Test with the audience that reads most deeply before concluding anything about demand.
We know relationship material is ~17% of reading; we cannot say who reads it. One label on component_category joins the content axis to the user axis.
People-management sits below fleet average for this mode and it reaches tarot least of any group. They hold 3.2 charts by choice.
Wants to know when. The answer is currently spread across screens they have to go and find.
Both already built inside Calendars; void-of-course already sits on the daily panel as a link out. 77.8% of the addressable audience is already on that screen, and it is where notifications land.
“Full moon tonight — three-minute reset.” Interest in practice tracks frequency (9.0% and 7.1% against 2.9% and 2.7%), and this is the most frequent mode.
Six tabs collapsed into one key until this month. Now instrumented; a fortnight settles it.
47% and 53% of visits end inside ten seconds. Reach doubled last quarter with no gain in depth. Move the content out; do not push more people in.
Owns 59% of the tarot audience and 72% of readings. The entry screen is the worst corridor in the product.
The tarot entry is opened 2.67 times for every reading reached — 55.2% of accounts are above a 2× ratio, the worst in the product. The reading itself holds people for 71 seconds. The room works; the hallway is too long.
“This card echoes the Uranus opposition currently active in your chart.” Their thriving quarter differs from their struggling one on facets (36% to 79%) and chart views (47% to 82%). The join is the thing a generic tarot app cannot copy.
Within engagement bands tarot and moon co-occur at 0.60 — below chance. Cards and calendars are different people.
Support contact lifts 4.33× on the edit screen, 3.17× on login and 2.13× on the picker, while content screens sit at 1.16–1.18. Before trying to move operator accounts from 0.6 visits a month to three, birth-data editing, login recovery, chart selection and adding people all have to reliably succeed.
Otherwise successful re-engagement simply exposes people to the same operational failures more often. The two navigation defects on that path are already written up as engineering issues.
Demonstrated demand or measurable failure. Mostly already built.
Clear audience, real build, little doubt that it is wanted.
Well-argued, mechanism unproven. Ship as experiments with a measure attached.
What I would not do yet is a large navigation redesign. The Matrix hierarchy converts at a gateway ratio of 0.44; once the product surfaces the right thing for the right job, much of the menu structure stops carrying weight. Fix the surfacing first, then re-measure what the menu is still doing.
Underneath all of it the strategic order is simply:
The finding that framed earlier drafts of this review, now supporting rather than leading: engagement runs backwards against investment. Chart count falls monotonically as return frequency rises.
They have not churned. Splitting the large-collection accounts by last-seen date, 97.9% opened the app within three months and only 87 accounts have been away longer than ninety days. This is a once-a-month rhythm, not an exodus — which is why the answer is a product change rather than a win-back campaign.
Chart views by type
Interpretations read, by family
| Link map — not visible vs not clicked | links | contextual | reach |
|---|---|---|---|
| chartpickuser | 11 | 5 | 39.6% |
| tarot | 7 | 2 | 15.7% |
| chartdaily | 6 | 0 | 53.3% |
| facets | 4 | 2 | 48.2% |
| profiles (add person) | 4 | 4 | 11.5% |
| moon | 3 | 0 | 15.2% |
| relationshipfacets | 2 | 0 | 22.9% |
| guide | 2 | 0 | 1.4% |
| harmonizer | 1 | 0 | 2.8% |
| messages | 0 | 0 | 0.0% |
Reach tracks contextual links, not total links. chartdaily has the most ways in of any screen and not one of them carries a reason — consistent with 40% never returning to it. profiles has four links, all contextual. relationshipfacets sits in the healthy band on reach with zero context.
Genuinely not visible: harmonizer (one link) and messages (none). Everything else is linked and under-contextualised.