Why review metrics weekly?

A week smooths out daily noise and still catches problems before they cost much. Daily numbers move with weekends, time zones and single creator videos, and a monthly review is too slow to catch a bad build. This split is our guidance: alerts for urgent problems, the weekly review for decisions.

Steam data also arrives late. Wishlist data is updated daily for the day before. UTM conversions are final 4 days after the visit. Players can request a refund within 14 days of purchase. Game data settles too: Unity notes that the data of the current day is still in flux, and that a D30 chart needs 30 days to be complete for an install date.

Sources: Steamworks Wishlists and UTM Analytics documentation; Steam Refunds policy; Unity Analytics dashboards documentation.

What is the agenda of the weekly review?

Seven items, in a fixed order, from stability to demand. The order matters: a crash on the latest build explains a drop in retention, and retention explains part of the sales. Each row has a question, the place to read the number and a threshold for action.

The thresholds are our suggestions for a small team, not industry rules. Two of them use published figures. The D1 threshold is the bottom quarter of PC games in the GameAnalytics Benchmarks Hub for October 2026, which is 4.23%. The refund rate is the Steamworks definition: units returned in a period divided by units purchased in that period. Replace our thresholds with your own once you have eight weeks of history.

The weekly review agenda. The action thresholds are our suggestions.
MetricQuestionWhere to read itAction threshold (our suggestion)
1. Crash-free sessions and errors per session, by buildIs the current build stable?Game health boardCrash-free sessions on the newest build are 1 point or more below the build before it
2. New players and the first session funnelDo new players get through the start?Game KPIs board and a Funnel chartOne funnel step converts 5 points or more below last week
3. D1 for last week, D7 for the week beforeDo they come back?Game KPIs board and the Install cohorts chartD1 is below 4.23%, or D1 falls two weeks in a row
4. DAU, returning and resurrected playersIs the player base growing?Game KPIs board and the Users pageDAU falls two weeks in a row with no beat that explains it
5. Units sold, units returned and net salesWhat did last week earn?Steam view, from the daily Steam sync (Early access)The refund rate is above your average of the last four weeks, outside a sale
6. Wishlist adds, deletes and purchasesIs future demand growing?Steam view, from the daily Steam sync (Early access)Deletes are higher than adds two weeks in a row
7. New reviews and their themesWhat do players say?Steam store page (reviews are not in Manalens)The same problem appears in three or more new reviews

Sources: GameAnalytics Benchmarks Hub; Steamworks Developer Refund Reporting and User Reviews documentation; Manalens dashboards, retention, users and Steam documentation.

How do I run the 30 minutes?

Open the same dashboard every week and follow the same steps. The 30 minutes and the steps are our suggestion for a solo developer or a small team. Valve suggests that you set a limit on the time you spend reading reviews, so cap that part.

  1. Set the range to the last 7 full days, and compare it with the 7 days before.
  2. Check stability. If crash-free sessions fell on the latest build, that is the first task of the week.
  3. Check new players. Note the new players and the step with the largest drop in the first session funnel.
  4. Check retention. Read D1 for last week's cohorts and D7 for the cohort before. Do not read D7 for a cohort less than 7 days old.
  5. Check Steam. Read units, returns and net sales, then wishlist adds, deletes and purchases.
  6. Read the week's reviews for 10 minutes and note the themes that repeat.
  7. Write three lines: what changed, why you think it changed, and one action for this week.

Sources: Manalens product documentation; Steamworks User Reviews documentation.

What does one sample week look like?

Here is one review written out. Every number is invented for this example. It is not data from a real game and it is not a benchmark. The game is a small premium PC game, six weeks after launch, and build 1.0.4 shipped on Tuesday of the week under review.

Read the table from the top. Row 1 passes the threshold: crash-free sessions on the new build are 1.9 points below the build before it. That one fact explains most of the rows below it. The tutorial step converts 5 points lower, D1 falls from 8.9% to 7.1%, and the refund rate goes from 6.5% to 9.2%. D7 does not move, because the D7 cohort started before the new build. Three reviews name the same crash.

Note the sample size before you trust each row. With 310 new players, one D1 point is about 3 players, so a fall of 1.8 points is 5 or 6 players. Alone, that is weak evidence. Together with the crash data and the reviews, it is enough to act.

The three lines for this week are these. What changed: crash-free sessions fell to 97.2% on build 1.0.4, and D1, the tutorial step and refunds moved with it. Why: a new crash on load in build 1.0.4, which is the top error in the Game health board. Action: ship a fix as build 1.0.5 by Wednesday, and read crash-free sessions for 1.0.5 and the D1 of the next cohort in the next review.

One sample week. All numbers are invented for this example.
Metric (example numbers)Last weekWeek beforeReading
Crash-free sessions, newest build97.2% (build 1.0.4)99.1% (build 1.0.3)Down 1.9 points: act
New players310292Stable
Funnel step: tutorial_complete58%63%Down 5 points: act
D1 of the new players7.1% (22 of 310)8.9% (26 of 292)Down 1.8 points, above 4.23%
D7 of the cohort one week older2.1% (6 of 292)2.0% (6 of 300)No change
DAU140150Down once: note it
Units sold and returned240 sold, 22 returned (9.2%)230 sold, 15 returned (6.5%)Refund rate up, no sale: act
Wishlist adds, deletes, purchases420, 130, 95400, 120, 90Adds are higher than deletes
New reviews12, with 3 about a crash on load9, with no repeated problemOne problem in 3 reviews: act

Sources: our own example; Manalens dashboards and error tracking documentation.

What counts as a real change?

A change that holds for two weeks, or one that you can tie to an event. Write down every beat next to its date: builds, sales, fests, announcements and creator videos. Most big moves in one week come from one of them.

Small samples mislead. With a few dozen new players a week, D7 can swing by many points by chance. Pool two or more weeks before you call a change, or test the gap with a significance calculator.

Valve also notes that most games' review scores slowly trend down over time, because later buyers are less sure that they will like the game. A slow slide in the recent score is normal. A sudden drop needs a reason. Refunds behave the same way: Steamworks says the weekly refund rate can be volatile after a launch and after a discount.

Sources: Steam blog, User Reviews (Valve, 19 September 2017); Steamworks Developer Refund Reporting; Manalens A/B test significance calculator.

Which alerts should run between reviews?

Alerts catch what cannot wait a week. In Manalens, an alert counts one named event in each five minute window, and it can filter by app fields such as the app version or build. Alerts go to Slack or to a signed HTTPS webhook. Two alerts cover most needs.

  • session_start with "fewer than" a count: the game stopped sending events, for example after a bad build.
  • app_unclean_exit with "at least" a count: a crash spike on the new build.

Sources: Manalens alerts documentation.

How is the weekly review different from launch week?

Launch week is daily and the weekly review is not. In the first days after release, a bad build or a creator video changes the numbers within hours, so you read one dashboard every day. The launch week dashboard template lists the charts for that week and how to build each one.

Change to the weekly review when the daily numbers stop changing your decisions. For most small games that is our suggestion of about two weeks after launch. Keep the launch dashboard: its player, health and Steam charts are the same charts that the weekly agenda reads, with a range of 7 days.

Sources: Manalens dashboards documentation.

What should a monthly review add?

The trends a week hides. Once a month, read D30 by cohort, net revenue per month, the wishlist balance, and the result of each beat of the month. Then plan the next month: one Steam beat, one update and one thing to fix.

Sources: Manalens retention and Steam documentation; Steamworks Wishlists documentation.

Can I see the whole review on one board?

Manalens is the analytics and growth engine built for games. Create the Game KPIs and Game health boards from templates, connect a Steamworks Financial API key (the automatic Steam sync is in early access) for daily wishlists and sales, and add your own charts to the same dashboard.

Sources: Manalens product documentation.

Frequently asked questions

Which day should I hold my weekly review?

Pick one day and keep it. Monday works well, because the review can cover the last full week, and Manalens weekly cohorts start on Monday in UTC. Steam wishlist days are in GMT and sales days in Pacific Time, so check that the last day has arrived before you read it.

How long should a weekly metrics review take?

About 30 minutes, which is our suggestion for a solo developer or a small team. Read the same charts in the same order every week: stability, new players, retention, Steam sales and wishlists, then reviews. At the end, write three lines with one action for the week.

Should I check my numbers every day after launch?

In the first two weeks, yes, for crashes and reviews. After that, daily numbers move with weekends and single videos, and they lead to rushed changes. Let alerts watch for crashes and missing events, and make your decisions in the weekly review instead.

What threshold should make me act?

Use your own history first. Our starting points are these: crash-free sessions 1 point below the build before, a funnel step 5 points below last week, a D1 below 4.23% (the bottom quarter of PC games in the GameAnalytics Benchmarks Hub for October 2026), or the same problem in three new reviews.

Why do my Steam numbers change after the week ends?

Because Steam data arrives late and gets corrected. Wishlist data is updated daily for the day before, UTM conversions are final 4 days after the visit, and players can request a refund within 14 days of purchase. Read last week, and check the week before again.

Sources and implementation