What Data to Track After Launching an Interactive Film Game: Choice Rates, Completion Rates, Replay Rates, and Ending Distribution

After launch, the priority is not to build a wall of metrics, but to distinguish four questions: whether players understand choices, whether they complete the experience smoothly, whether they want to explore, and at which nodes content or technical issues block them. Choice rates, completion rates, replay rates, and ending distribution must use actual exposure and opportunities to play as their denominators, segmented by version, platform, and first playthrough versus replay.

An option’s choice rate = the number of players who selected it ÷ the number who saw it and were eligible to select it. Do not divide by all players, because conditions, language, resources, and preceding routes change exposure. For timed choices, also track timeouts, no input, and interface focus loss separately.

Also examine the distribution of decision times. If almost everyone instantly selects the same option, the answer may be too obvious, or the wording’s position may make it the default. Long hesitation may indicate a successful conflict of values, or options that are hard to understand. Interpret this alongside test recordings and feedback.

Build a funnel from launch through prologue completion, reaching each chapter, the first ending, and normal completion. At each step, also examine media errors, crashes, loading times, and voluntary exits. Do not conflate content-related attrition with technical attrition.

Examine completion rates by cohort: first play date, platform, version, language, and device tier. An overall increase after a hotfix may simply reflect a different mix of new players; compare similar cohorts to assess changes.

Replay can mean clicking a new route after an ending, returning to an early chapter, using the skip-read-content feature, or starting a second session within a certain number of days. First choose a definition that matches the product’s goals, such as “starting a new timeline within seven days of the first ending.” Exclude restarts after crashes and loading saves to correct mistakes.

Then examine replay depth: did players merely open it, or reach a new node or ending? Many starts with little progress may indicate too much repeated content, a skip feature that is difficult to use, or prompts that offer no direction.

Use players who have completed at least one playthrough as the denominator for each ending’s share, and distinguish first-playthrough results from cumulative results. A rarely reached ending may be intended as a hidden reward. If almost nobody reaches a major ending, first verify that its conditions, saves, and paths work properly before discussing player preferences.

A small number of endings does not mean the experience lacks variety. Display major endings and epilogue variants in separate layers, and examine the key states required to reach them. Do not force changes to the story just to make a bar chart look even.

For each key node, display arrivals, option exposure, submissions, media preparation failures, exits, average wait time, and arrivals at the next node side by side. This can reveal problems where button clicks work normally but large numbers of players disappear after a video transition. Looking only at choice rates will miss these breakpoints.

Also create health metrics for save restoration, skipping read content, and accessibility settings: restoration success rate, the rate of reaching new content after skipping, and completion rate after using timer assistance. The aim is to confirm that features deliver on their promises, rather than evaluate players.

Regularly categorize customer support messages, reviews, community discussions, and interviews, preserving their meaning rather than merely counting positive and negative sentiment. Link recurring themes to nodes and versions, such as “misleading options,” “subtitles too fast,” and “unclear character motivations.” Data shows where something is unusual; players’ words help explain why.

Do not treat a small sample of comments as representative of everyone, or use overall averages to dismiss specific obstacles. Cross-check the two types of evidence.

Before changing option wording, write down your expectations: fewer misunderstandings and a higher submission rate, while satisfaction with relationship branches must not decline. After a hotfix, monitor both target and guardrail metrics to avoid increasing completion rates at the expense of meaningful hesitation.

Not every distribution needs optimization. A work may intentionally make a choice painful or an ending rare. If players understand it, the technology is reliable, and it serves the creative goals, there is no need to pursue an even distribution.

Collect only necessary anonymous events, provide appropriate notice and controls, and restrict access and retention according to policy. Filter out internal tests, bots, duplicate reports, and anomalous clocks. Display event versions, sample sizes, and data delays on the dashboard. When the data pipeline is unreliable, pause major content decisions.

On day one, focus on blockers involving launch, crashes, downloads, and saves. In the first week, examine chapter funnels and recurring feedback. Analyze choices and endings once the sample is stable. Changing the story too early based on a few dozen core fans can amplify noise. Record observations, interpretations, evidence strength, decisions, and follow-up dates at every review.

Separate content experiments from emergency fixes. Fixes for erroneous paths can ship quickly; changes to option meanings or ending conditions require version labels, guardrail metrics, and player communication. Otherwise, earlier and later cohorts face different works, and their data cannot be compared directly.

Next step: write a precise definition, denominator, segmentation dimensions, and actionable triggers for each of choice rates, chapter completion, replay, and ending distribution. Manually recalculate them using ten test sessions to confirm that the dashboard matches the raw events.