Daily Active Users (DAU): Formula, 2026 Benchmarks, Infra

Updated October 2026: DAU formula and system-design math added, Q2 2026 DAU figures from Meta, Reddit and Duolingo filings, current Agora, R2, B2 and analytics pricing, Hetzner’s 2026 price rises.

Why Fora Soft wrote this playbook

Every Fora Soft scoping call starts with one question: how many people do you expect to use the product in month one, and how many concurrently? Founders sometimes hesitate. They should not. The answer determines server architecture, SFU choice, recording pipeline, CDN budget, load-test plan, and whether your app reaches the App Store on the timeline you wrote in the deck. Our product planning and analytics work turns that number into a forecast, an analytics setup and a sized roadmap.

This playbook is the conversation we have every week with founders, CTOs, and product leads. It explains what DAU is and why it matters more than any other metric, how to forecast it, what good DAU looks like by category, how DAU shapes infrastructure cost, and why “build for millions on day one” is the most expensive mistake you can make.

Fora Soft has shipped 250+ projects since 2005, from live commerce platforms with thousands of concurrent viewers to corporate telemedicine apps with closed user bases of dozens. We size every infrastructure decision against the DAU forecast.

Figure 1. An honest DAU forecast plus concurrent peak sizes every downstream infrastructure choice.

What are daily active users (DAU)? Definition and formula

Daily active users (DAU) is the number of unique users who complete at least one meaningful action in your product during one calendar day. The formula: DAU = the count of distinct user IDs with one or more qualifying events between 00:00 and 23:59 in a fixed time zone, usually UTC. In SQL that is COUNT(DISTINCT user_id) over the day’s qualifying events.

Worked example. On Monday, user A joins two calls, user B opens the app and sends one message, user C installs the app but never signs in. If “active” means “sent a message or joined a call”, Monday’s DAU is 2: A counts once despite two sessions, and C does not count. Count people, not devices: one user on a phone and a laptop is one DAU. Exclude bots, background syncs and silent push wake-ups.

Average DAU is the mean of daily DAU over a period: add up each day’s DAU and divide by the number of days. That is how public companies report it. Snap’s Q2 2026 10-Q defines quarterly average DAU as the sum of DAUs for each day divided by the number of days in the quarter. A 30-day average smooths out weekends and launch spikes, so use it for trend lines and keep daily values for capacity planning.

The definition of “meaningful” is still yours: opening the app, sending a message, joining a call, completing a workout, finishing a lesson. The discipline is to define it once, instrument it, and never quietly change it.

DAU is the single most predictive metric for product viability. Investors look at DAU growth before revenue. Engineering looks at DAU before scaling. Marketing looks at DAU per channel before doubling spend. A product without a defensible DAU number is a product running on hope.

Related metrics. WAU (weekly active users) and MAU (monthly active users) tell you the breadth of your audience. The DAU/MAU ratio (called stickiness) tells you how often the average user comes back.

The DAU/MAU stickiness ratio — the most overlooked product KPI

DAU/MAU expresses, on average, how many days per month a user shows up. A 20% ratio = 6 days / month. A 50% ratio = 15 days / month. The benchmark band tells you what kind of product you have.

These bands are the working ranges we use on client products, not a published standard, and most “industry benchmark” numbers online cannot be traced to data. The stickiness figures you can check come from company filings: Duolingo’s Q2 2026 shareholder letter reports 58.7M DAU and 140.6M MAU, which works out to about 42%. Measure against your own category and your own trend, not a number from a listicle.

There is no universal “good” DAU/MAU. There is the right band for your category. Setting the wrong target wastes engineering effort and confuses growth decisions.

DAU benchmarks in 2026: what public companies report

The largest audited DAU numbers in 2026 come from quarterly filings: Meta reported 3.60 billion daily active people in June 2026, Reddit 130.3 million daily active uniques in Q2 2026, and Duolingo 58.7 million DAU in Q2 2026. Each company counts “active” differently, which is exactly why you should fix your own definition before comparing.

Two lessons for a product team. First, the definition moves the number: Reddit’s DAUq includes logged-out readers, Meta’s DAP de-duplicates people across four apps. Second, even giants grow DAU slowly once they are big (Meta +3%), while focused habit products still grow fast (Duolingo +23%). Your own month-one number will be tiny next to these, and that is fine: what matters is the slope and the DAU/MAU ratio.

Why DAU drives every infrastructure decision

Knowing DAU and concurrent peak lets us:

1. Right-size servers. A clear DAU forecast prevents the two failure modes — over-paying for idle capacity, or capacity-flooring under launch load. We map DAU to peak concurrent and back into vCPU / RAM / network for each tier.

2. Pick the SFU and CDN. 200 concurrent viewers run on one SFU; 5,000 across multiple regions need a managed SDK, multi-POP routing, or self-hosted scaled SFU clusters. Different products, different stacks.

3. Plan recording and storage. Recording a 30-min call at 720p is ~270 MB. At 5,000 daily calls, that’s ~1.35 TB a day, or ~40 TB a month. Kept for 30 days on Cloudflare R2 ($0.015 per GB-month, free egress, October 2026 list) that is about $600 a month; kept for a year (~480 TB) it is about $7,200 a month. Retention policy, not DAU alone, decides the storage bill.

4. Schedule load tests. A load test sized to 1,000 concurrent uses 5× less budget than one sized to 10,000. We schedule test events to match the realistic launch curve.

5. Decide where to push compute. Server-side video processing or client-side? Heavy server compute scales linearly with DAU; client compute does not. The DAU number tilts the trade.

How to forecast DAU before launch (and what to do if you cannot)

If your product is already live. Use Google Analytics, Mixpanel, Amplitude, or PostHog. Pull DAU, MAU, retention curves, and event funnels for the past 90 days. Trend the slope into your forecast.

If you have backend telemetry but no analytics. Pull request rates from your reverse proxy, unique sessions from your auth logs, or daily-active devices from the cloud provider console. Less precise; still actionable.

If you are pre-launch. Three options. (1) Use Similarweb or Sensor Tower (which absorbed data.ai, formerly App Annie, in March 2024) to estimate competitor DAU and apply a fraction. (2) Use the “1,000 active users in month one” default our experience supports for most pre-launch consumer apps. (3) Build to scale to 10,000 concurrent without re-architecture, so the actual launch number does not matter as much.

If you are pre-launch B2B. Sales pipeline maps to DAU more directly. 50 logos × ~25 seats × daily-use stickiness 30% ≈ 375 DAU at full ramp. Ramp to that figure over 6–12 months.

If your product is closed enterprise / corporate / government. The DAU is your seat count, period. Build for the seats; do not over-engineer for “maybe public someday.”

Concurrent peak vs DAU: the metric infrastructure actually needs

Servers do not care how many users you have today. They care how many are active right now. Translating DAU to peak concurrent depends on session length, time-zone spread, and product type.

Figure 3. Servers scale to concurrent peak. Solid bars show the low end of each range; lighter shows the high end.

A live commerce app at 1,000 DAU is heavier than a productivity SaaS at 10,000 DAU. The product type matters as much as the headline number.

What is DAU in system design? From DAU to requests per second

In system design, DAU is the first input of every back-of-the-envelope estimate: average requests per second (QPS) = DAU × requests per user per day ÷ 86,400, and peak QPS is that figure times a peak factor, which we take as 2–3× for a single-region product. Storage and bandwidth follow the same pattern: DAU × per-user volume.

Worked example: 50,000 DAU. Say each user makes 40 API calls a day: 50,000 × 40 = 2,000,000 requests a day, or about 23 QPS on average and 50–70 QPS at peak. Two photo uploads of 200 KB per user add about 20 GB of storage a day. If 10% of those users are on a video call at the busiest hour, that is 5,000 concurrent streams; at 1.5 Mbps each, the media layer must carry about 7.5 Gbps. The API tier for 50k DAU is small; the video tier is where the money and the engineering go.

A worked cost model: what infrastructure costs by DAU

Order-of-magnitude monthly infrastructure cost for a video-heavy app at three DAU scales, sized realistically and using a managed SDK.

Figure 4. Order-of-magnitude monthly cost at three DAU scales. The video-SDK line dominates as you grow.

How the numbers are built. The SDK column assumes about 2.5 HD participant-minutes per DAU per day (75,000 minutes a month at 1,000 DAU), priced at Agora’s October 2026 list of $3.99 per 1,000 HD minutes (audio $0.99, Full HD $8.99, first 10,000 minutes a month free). A product where every user spends 20 minutes on video daily will pay about eight times more. Server prices moved too: Hetzner raised prices on existing products from April 1, 2026 and again for new cloud orders from June 15, 2026, when some US cloud instances roughly tripled. Re-quote before you budget.

Self-hosted SFU on Hetzner brings the SDK line down sharply at 50k+ DAU; below that, managed SDKs win on operational simplicity. When you cross that line, our scalable video streaming development team moves the load to mediasoup or LiveKit SFU clusters sized to the same DAU forecast. The encoding choices behind those SDK minutes (codec, bitrate ladder, resolution) set your per-stream cost; our video encoding guide walks through the trade-offs.

Mini case: a DAU-driven launch plan that saved $40k / month

Situation. A consumer live streaming startup had budgeted “a million users on launch” into the architecture. The proposed cluster — multi-region Kubernetes, ten redundant SFUs, hot recording pipelines — quoted at $42k / month before launch.

What we changed. Looked at the marketing plan honestly. Realistic launch DAU was 500–2,000 in month one, growing to ~10,000 over six months. Re-architected to a single SFU on a Hetzner AX node, a managed SDK fallback for any spike above the floor, and a recording pipeline scaled with launch traffic. Built the path to multi-region but did not pay for it on day one.

Outcome. Month-one infrastructure cost dropped from a budgeted $42k to $1,800. Launched on schedule. Hit 8k DAU in month four; scaled the SFU to a 3-node cluster in week 18. The $40k / month difference funded a marketing sprint that drove most of the early growth.

We have run similar “right-size for the real DAU” exercises across dozens of projects. Honest forecasting almost always frees budget that can fund product work or marketing.

Server vs client compute: where DAU tilts the trade

For a video editing or filter app, two architectures are valid: server-side processing (cheap to ship, expensive to scale) and client-side processing (more code, cheaper to scale).

Server-side wins when: early DAU is small, devices are constrained (Africa, SEA, Latin America), platform breadth matters (web, iOS, Android shipped together), and time-to-market is the dominant constraint.

Client-side wins when: DAU is large, devices are modern, latency matters, offline-first is needed, and per-server cost is a real line in the P&L.

Hybrid usually wins. Frame effects on the client, format conversion on the server. Real-time inference on-device, batch jobs on the server. We design on-device inference against the same DAU forecast that drives server sizing.

Latency budgets shift with DAU and product type

Latency-sensitive products such as real-time video effects, music collaboration, DJ mixing need sub-1-second glass-to-glass at every DAU level. The architectural cost of meeting that bar at 1,000 DAU is small. At 100,000 DAU, it dominates infrastructure choice. See our deep dive on minimizing latency for mass streams.

Latency-tolerant products such as transactional banking, search, and content publishing can ship a simpler architecture that grows with DAU rather than ahead of it.

DAU patterns by vertical: what we see in real Fora Soft projects

Live commerce. Sprii-style live shopping shows extreme peak / DAU ratios — 70%+ of DAU concurrent during peak shows, 5%+ outside. Build for the show, scale down between shows.

Telemedicine. Like CirrusMED: appointments are scheduled, peaks are predictable, peak/DAU stays under 15%. The infra optimization is throughput, not burst.

E-learning. BrainCert-style virtual classrooms have hour-long sessions and academic-calendar peaks — September, January, April. Scale with the calendar.

Video interpretation. TransLinguist runs a marketplace of 30,000+ certified interpreters in 75+ languages and won the NHS UK national framework. The peak/DAU ratio is moderate, but cross-region routing matters more than burst capacity.

Closed enterprise / on-prem. Like Nucleus: DAU is the seat count, period. Build for the seats, never for “maybe public someday.”

Re-architecture triggers: knowing when to rebuild before it hurts

A clear DAU forecast lets you pre-define when to evolve the architecture, instead of scrambling when production starts smoking. If you already run in production, an independent architecture review checks your current stack against these triggers and says what to keep and what to change.

Trigger 1 — single SFU saturation. When peak concurrent passes ~500–1,000 (depending on workload), move from a single SFU to a clustered or managed setup.

Trigger 2 — recording storage growth. When daily recording volume passes ~500 GB, evaluate cheaper storage tiers (Cloudflare R2 at $0.015 per GB-month with free egress, Backblaze B2 at $6.95 per TB-month, both October 2026 list prices) or aggressive lifecycle policies.

Trigger 3 — cross-region latency. When >30% of DAU is outside your primary region and P95 latency climbs above 250 ms, add a second region or move to a global SDK.

Trigger 4 — per-minute economics. When managed-SDK billing exceeds ~$15k / month consistently, evaluate self-host on Hetzner.

Trigger 5 — compliance scope change. New customer with HIPAA, FedRAMP, or EU residency. Plan the architecture upgrade alongside the contract.

Figure 5. Pre-define each threshold so a rebuild is a planned upgrade, not a production fire.

A decision framework: turn DAU into a build plan in five questions

1. What is your honest month-one DAU? Use telemetry, competitor analysis, or the “1,000” default. Do not pad. Pad later if growth surprises you.

2. What is your peak concurrent / DAU ratio? Use the table above as a starting point. Adjust for your time-zone spread.

3. Which path can scale 10× without re-architecture? The answer should be cheap on day one and survive a four-quarter growth surprise.

4. Where is your latency-critical surface? Real-time video, multiplayer, live audio. Spend infra budget here, not on the easy parts.

5. What triggers a re-architecture? Pre-define DAU thresholds (e.g., “at 25k DAU we switch to multi-region”). Knowing the trigger is half the work.

Five DAU pitfalls we see every quarter

1. “Build for millions on day one.” The most expensive mistake. Burns cash, delays launch, locks in over-architecture. Build for 10x your launch number, not 1000x.

2. Reluctance to forecast. “You’re developers, you know better.” No one knows your audience like you do. Honest forecasting is non-delegable.

3. Confusing DAU with vanity install counts. A 100,000-install app with 200 DAU is not a 100,000-user product. Investors and engineers care about the 200.

4. Quietly redefining DAU after launch. If “active” means “opens app” on day one and “completes a meaningful action” on day 30, your trend lines lie. Define once, instrument cleanly, never change.

5. Not load-testing the realistic peak. A load test at 100× expected concurrent burns budget; one at 1.5× expected validates the launch.

KPIs to track alongside DAU

Engagement KPIs. DAU/MAU stickiness inside the band for your category. Sessions per DAU per day >1.5 for a habit product. Average session length matching your product type. Retention curves stabilizing past day 30 (a flat tail is a sign of product-market fit).

Reliability KPIs. P99 server response time <500 ms in-region. Crash-free session rate >99.7%. Error rate per DAU <0.5%. Push notification delivery >95%.

Cost KPIs. Infrastructure cost per DAU <$0.50 / month for most consumer apps; <$5 for video-heavy products. Cost per active session <$0.05 for most categories. If your unit economics break at 100 DAU, no DAU growth will save you.

How to increase DAU (and DAU/MAU) without gaming the metric

You increase DAU by giving users a reason to come back every day, not by loosening the definition of “active”. The levers that work in the products we build are the same five every time:

1. Shorten time to first value. A user who completes the core action on day one (first call, first lesson, first live show) is the user who returns on day two. Cut sign-up steps and pre-call checks before you buy traffic.

2. Build a daily reason to return. Scheduled live sessions, daily content drops, streaks and progress. Duolingo, a product built around daily streaks, still grew DAU 23% year over year in Q2 2026 at a scale of 58.7M daily users.

3. Send notifications that carry content. “Your class starts in 10 minutes” brings users back; “We miss you” trains them to mute you. Cap frequency and measure opt-outs next to DAU.

4. Fix reliability before growth. A dropped call or a frozen stream on day one ends the habit. Crash-free sessions and call-join success are DAU metrics in disguise.

5. Watch DAU/MAU, not DAU alone. Paid installs lift DAU for a week while stickiness falls. Growth is real when DAU and DAU/MAU rise together.

When DAU is not the right metric to optimize

For low-frequency-by-design products such as tax filing, real-estate transactions, and certain B2B tools, DAU is structurally low. There, optimize for completion rate per session and revenue per active month, not DAU. For internal corporate tools, the seat count is the population; DAU just tells you whether the rollout stuck.

For most consumer and B2C products, however, DAU is non-negotiable. If it is not the headline metric on your dashboard, it should be.

Build vs buy: tools to instrument DAU

For most early-stage products, PostHog (open-source) or GA4 (free) gets you to actionable DAU within a sprint. Move to Mixpanel or Amplitude when the team needs deeper cohort analysis.

Free tiers as of October 2026: PostHog gives the first 1M product-analytics events a month free, Mixpanel 1M events a month, Amplitude 2M events a month, and GA4 is free. At 1,000 DAU and 30 tracked events per user per day you generate about 900,000 events a month, so all three paid tools stay free until roughly that scale.

FAQ

What is the formula for daily active users?

DAU = the number of distinct user IDs with at least one qualifying action in one calendar day, in a fixed time zone. Count people, not devices or sessions, and exclude bots and background syncs.

What is average daily active users?

Average DAU is the sum of each day’s DAU over a period divided by the number of days in it. Public companies such as Snap report quarterly average DAU this way.

What is DAU in system design?

It is the starting input for capacity estimates: average QPS = DAU × requests per user per day ÷ 86,400, then multiply by a 2–3× peak factor. 50,000 DAU at 40 requests each is about 23 QPS average and 50–70 QPS at peak.

How do you increase DAU?

Shorten time to first value, give users a daily reason to return (live sessions, streaks, fresh content), send notifications that carry content, and fix reliability. Track DAU/MAU next to DAU so paid installs do not hide falling stickiness.

What counts as “active”?

Whatever your product calls a meaningful action: opening the app, sending a message, joining a call, completing a workout. The discipline is to define it once, instrument it cleanly, and never quietly change the definition.

What is a good DAU/MAU ratio?

Above 20% is a habit product in our experience; below 10% is transactional or occasional use. For a public reference, Duolingo reported 58.7M DAU and 140.6M MAU in Q2 2026, about 42%. Compare to your own category and trend, not to an absolute target.

How do you forecast DAU before launch?

Three options: pull competitor estimates from Similarweb or Sensor Tower and apply a fraction; use the “1,000 active users in month one” default for consumer apps; or build to scale to 10,000 concurrent so the precise launch number matters less.

Why does DAU forecasting save money?

Every infrastructure decision — servers, video SDK, recording, CDN, load tests — keys off concurrent peak, which keys off DAU. Wrong forecast by 10× means either burning cash on idle capacity or missing your launch under load. Honest forecasts free budget.

Should you build for millions on day one?

Almost never. The marginal cost of over-engineering is real cash, real delay, and real complexity. Build for 10× your honest launch number, plus a clear scaling trigger. Re-architect on a known threshold, not a hope.

How does DAU affect my app store retention metrics?

App stores look at install-to-D1, D7, D30 retention more than absolute DAU. But healthy DAU growth without churn looks like exactly that — rising retention curves. Stores reward apps that keep users engaged.

Which tool should you instrument first?

For pre-PMF products: PostHog (open-source, free for low volume) or Google Analytics 4. For mid-market growth teams: Mixpanel or Amplitude. For late-stage: pipe events to a warehouse (BigQuery, Snowflake) and slice with SQL.

Does Fora Soft help with DAU-driven scaling plans?

Yes. We bake the DAU forecast into every scoping call, build sized infrastructure on day one, and pre-define re-architecture triggers. The result is right-sized cost on launch and a clear path to scale when the curve goes up.

What to Read Next

Ready to size your build to your real DAU?

Daily active users is the metric that turns a product idea into an infrastructure plan. Honest DAU forecasting saves money on launch, fits your architecture to growth, and frees the budget that funds product-market fit. Bad DAU forecasting either over-engineers a product nobody is using yet or under-builds one that hits the front page.

Fora Soft has been right-sizing infrastructure to DAU since 2005. We size every project at scoping, document the re-architecture triggers, and ship the minimum architecture that survives a 10× growth surprise. If you want a sober DAU-driven plan for your next launch, let us run the numbers with you.