You're the head PM of Facebook Groups product. MAP is steady, but DAP has declined by 20%. This trend has gradually occurred over the last 3 months. What do you want to do next?

Reviewed against current first-party Meta and Facebook documentation on August 2, 2026. The 20% decline in this prompt is a hypothetical interview scenario, not a reported Meta result.

Short answer

I would not start by changing the Feed algorithm, sending more notifications, or assuming a competitor caused the decline. I would first confirm what “Facebook Groups DAP” means and whether the decline is real. Then I would identify which segments and which step of the Groups experience account for the lost daily activity. Only after one hypothesis explains both the size and the three-month timing would I choose a fix and test it.

Steady MAP with DAP down 20% tells us that a smaller share of the monthly-active population is active on a typical day. It does not prove that every user visits less often, that members no longer value Groups, or that discovery is the cause. A change in cohort mix could produce the same aggregate pattern.

Clarify the metric and scope

I would begin with a precise internal metric definition:

For this case, I will treat Groups DAP as the number of deduplicated, logged-in Facebook users who complete at least one qualifying Groups action during a calendar day. MAP is the monthly equivalent over a rolling window.

I would ask whether a passive group-post view qualifies, or whether the user must open a group, react, comment, post, or complete another action. I would also confirm which surfaces count: the Groups tab, Feed posts from groups, recommendations, notifications, search, direct links, and any separate community products.

This matters because Meta's public DAP is a different company-level metric. Meta's current SEC filing defines Family DAP across Facebook, Instagram, Messenger, and WhatsApp, using models to deduplicate people. It does not publish a Facebook Groups DAP definition. The interview metric therefore needs an internal definition before it can be diagnosed.

My remaining clarifying questions would be:

• Is the 20% change relative or an absolute percentage-point change?
• Is the decline visible in raw daily counts, the DAP-to-MAP ratio, or both?
• What comparison period and time zone are used? Does the result survive weekday and seasonal adjustment?
• Did MAP remain steady in every major market and platform, or only in aggregate?
• Was the decline gradual within cohorts, or caused by new monthly users who rarely returned?
• Were there instrumentation, identity, enforcement, privacy, ranking, navigation, or release changes near the start of the trend?
• Did any quality or safety metric change at the same time?

Read the DAP-to-MAP signal correctly

If MAP is unchanged and DAP falls by 20%, then DAP divided by MAP also falls by 20%. For example, a ratio of 0.25 would become 0.20. That means the product is generating fewer daily active people for the same size monthly-active population.

The ratio alone does not reveal why. At least four stories fit it:

1. Existing members return on fewer days.
2. A growing share of MAP consists of low-frequency or newly acquired users.
3. People still visit monthly but encounter less relevant content on most days.
4. Logging or deduplication undercounts daily activity while monthly identity resolution remains stable.

I would separate those stories with cohort and event data before discussing solutions.

Validate that the decline is real

First I would audit the metric pipeline from event to dashboard:

• Compare the event definition and identity rules before and after the trend began.
• Check event volume by app version, operating system, region, surface, and server versus client logging.
• Look for delayed ingestion, backfills, time-zone changes, bot removal, account enforcement, consent changes, and deduplication-model changes.
• Sample raw event records and reconcile unique users with an independent query.
• Compare Groups DAP with nearby proxies such as group opens, post views, comments, reactions, posts, and active members.

Facebook's Group Insights documentation lists active members, posts, comments, reactions, popular times, and top posts among the signals available to group admins. Its Post Insights documentation separately defines reach and engagement. These can serve as directional cross-checks, but I would not substitute them for the internal DAP definition.

A measurement problem becomes more likely if the decline begins sharply on one logging version while independent actions remain steady. It becomes less likely if multiple independent server-side and client-side signals fall together.

Localize the lost activity by contribution

Once the metric is trusted, I would calculate each segment's absolute contribution to the decline:

baseline segment DAP minus current segment DAP

I would rank segments by lost people, not just percentage change. A tiny cohort can have a dramatic percentage decline without explaining the product-level result.

I would cut the data by:

• country and region
• iOS, Android, mobile web, desktop, and app version
• public versus private groups
• small, medium, and large groups
• new members, established members, and reactivated members
• passive readers, contributors, and admins
• new versus established groups
• entry surface: Feed, Groups tab, search, recommendation, notification, and direct link
• action: view, open, react, comment, post, moderation, and return
• content and safety cohorts, including recommendation eligibility

Public and private groups have different visibility and participation mechanics, as Facebook's group privacy documentation explains. I would therefore avoid combining them until their trends are understood.

I would also build weekly acquisition cohorts and compare their daily return curves. If established cohorts are stable but recent cohorts return less often, the aggregate decline is a mix or onboarding problem. If most cohorts bend downward at the same calendar date, a shared product, ecosystem, or measurement change is more plausible.

Find the broken step in the experience

I would use this diagnostic funnel:

eligible user → discovery or notification exposure → group or group-post open → relevant content available → qualifying Groups action → next-day return

This is an analytical model for the case, not an official Meta funnel. I would examine both demand and supply.

On the demand side, I would compare exposure, open rate, content depth, qualifying-action conversion, and return rate. On the supply side, I would compare active groups, active admins, posts per active group, approval latency, contributor retention, content freshness, and the share of content eligible for distribution.

Facebook says eligible Groups can appear through recommendations such as suggested Feed content and “Groups You Should Join.” Its recommendation guidance also explains that recommendation eligibility can be stricter than basic platform eligibility. A fall in recommendation exposure is therefore a valid hypothesis to test, not a cause to assume.

Likewise, Facebook's notification documentation says group notifications default to Highlights and can be changed or turned off. Notification delivery and opens belong in the investigation, but the existence of notifications does not prove that they drove this decline.

Test falsifiable hypotheses

The three-month slope is a clue, not a verdict. It could fit a staged rollout, a gradually changing cohort mix, weakening content supply, or a policy and eligibility change. It is less characteristic of a single global outage, but even that should be checked rather than dismissed.

I would place every relevant launch and configuration change on the same chart as the affected metric. For a case set in 2026, that change log could include the separate Forum Groups app that Meta said it began testing in May 2026. Meta describes it as synchronized with Facebook Groups in its July 2026 product update. I would investigate it only if the scenario's dates, markets, and cohorts overlap the test. Its existence is not evidence that it caused the decline.

Choose the next action from the evidence

I would select the hypothesis that explains the largest absolute share of lost DAP, matches the onset, and survives an explicit falsification check. The next action then depends on the result:

• If instrumentation is wrong, repair and backfill the metric before changing the product.
• If one release or navigation path is responsible, compare exposed users with an unaffected version or holdout, then roll back or patch the confirmed regression.
• If discovery exposure fell, inspect eligibility and ranking changes, restore only the affected path in a controlled experiment, and measure downstream quality as well as DAP.
• If notification delivery broke, fix delivery. I would not compensate by sending more notifications or overriding member preferences.
• If content supply weakened, test the specific contributor or admin friction identified in the funnel rather than launching a generic engagement campaign.
• If moderation or recommendation eligibility explains the loss, improve the relevant workflow or content quality without weakening safety standards.
• If newer cohorts are low-frequency, test more relevant group discovery and clearer first-session value for those cohorts while leaving healthy established cohorts alone.

Each experiment needs a control or credible comparison group, enough power to detect the predeclared meaningful effect, and a readout long enough to capture return behavior. I would not invent a universal target or deadline without baseline variance, sample size, and business constraints.

Metrics and guardrails

The primary recovery metric is validated Groups DAP, supported by DAP divided by MAP and cohort return frequency.

Leading indicators depend on the diagnosed step:

• discovery and notification exposure
• group and group-post opens
• qualifying-action conversion
• active contributors and active groups
• fresh content per active group
• next-day and weekly return
• admin approval and moderation flow health

Guardrails should include hides, reports, blocks, group leaves, unfollows, notification opt-outs, spam prevalence, recommendation quality, admin workload, crashes, latency, and integrity incidents. A change that raises low-quality opens while damaging trust is not a successful recovery.

A concise interview answer

“I would first define Groups DAP because Meta's public Family DAP is not a Groups metric. I would confirm which users, surfaces, and actions count, then validate the 20% decline against raw events and independent usage signals. With MAP flat, the DAP-to-MAP ratio is also down 20%, but that does not tell us whether existing users visit less often or whether MAP now contains more low-frequency users.

“Next I would rank segments by their absolute contribution to lost DAP across platform, version, region, public or private group, user tenure, role, entry surface, and action. I would map the loss onto the funnel from exposure to open, relevant content, qualifying action, and return, while checking content supply and moderation signals.

“I would align the dominant break with the product and measurement change log, then test a falsifiable hypothesis against an unaffected cohort. The fix would follow the diagnosis: repair instrumentation, patch a release regression, restore a confirmed discovery path, address specific supply friction, or improve onboarding for low-frequency cohorts. I would judge recovery on validated Groups DAP and return behavior, with safety, content quality, notification opt-outs, reliability, and admin workload as guardrails.”

Sources

• Meta Q1 2026 Form 10-Q: current public Family DAP definition and measurement limitations
• Facebook Help: view Group Insights
• Facebook Help: view insights about a group post
• Facebook Help: recommendations on Facebook
• Facebook Help: group notification settings
• Facebook Help: differences between public and private groups
• Meta Newsroom: 2026 Facebook and Groups product changes