Why a live stream needs its own audience maths
A video on a shelf has one number that matters, the views. A live stream has several, and they answer different questions. How many people were in the room at the busiest moment? How many were there on average? How many different people came at all? How long did a typical viewer stay? If a brand or creator mixes these up, a stream can look like a triumph on a screenshot and a disappointment in the sales report, or the other way round.
This free lesson gives you the vocabulary and the three formulas that every later lesson in the course builds on: the average concurrent viewers, the average watch time per viewer and the peak-to-average ratio. You calculate them in a spreadsheet from three plain counts that any streaming platform's analytics page will show you in some form: the stream length, the unique viewers and the watch minutes.
The example is Wavelark Studio, a fictional baking-kit brand that runs a weekly Saturday show called the Bake-Along. All streams, names and numbers are invented for learning. They are illustrative and not official platform data, and real platforms name and count these metrics a little differently, so always read the definition in your own analytics before you compare numbers.
The four numbers on every live dashboard
Concurrent viewers (often written CCV) is the number of people watching at one moment. It changes every minute, so it only makes sense together with a time or a summary. Peak concurrent is the highest CCV during the stream, a single moment. Unique viewers counts every person once, however long they stayed, so it is always at least as large as the peak. Watch minutes (also called viewer-minutes) is the sum of the minutes every viewer spent watching the live show.
Think of a stream as a room with a door. The peak is the most people the room held at once. The unique viewers are everyone who walked through the door. The watch minutes are the total time that all visitors together spent inside. Stream S-01 of the example had 1,240 unique viewers, a peak of 312 and 11,400 watch minutes in a 60-minute show.
Average concurrent viewers: the fair size of a stream
The peak can be a lucky moment, such as a raid or a mention in a big channel. A fairer size of the show is the average concurrent viewers: watch minutes ÷ stream length in minutes. It answers the question "how many people were in the room on average?" For S-01 that is 11,400 ÷ 60 = 190 viewers, although the peak was 312.
The average concurrent is also the number that connects a live audience to an hour of attention. An average concurrent of 200 over a 60-minute stream equals 12,000 watch minutes, which is 200 viewer-hours of attention. That makes it easy to compare shows of different length: a 90-minute stream is not bigger just because it produced more watch minutes.
Average watch time per viewer: how long people stay
The second question is about loyalty within one show. The average watch time per viewer is watch minutes ÷ unique viewers. For S-01 that is 11,400 ÷ 1,240 = 9.2 minutes. A typical person stayed a little over nine minutes of a 60-minute show. That is not a failure by itself: many viewers drop in, see whether the topic interests them, and leave. What matters is how the number moves over weeks and with the format.
Watch time per viewer depends on the length of the show. Longer streams give people more time to stay, so a 90-minute stream will usually show a higher number of minutes than a 60-minute one. To judge fairly, compare it with the stream length as well, or compare streams of the same length only. We do this near the end of the worked example.
The peak-to-average ratio: spiky or steady?
Divide the peak by the average concurrent and you get the peak-to-average ratio. A ratio close to 1 means the room stayed about the same size all evening. A high ratio, say 2 or more, means a short spike that did not stay. The ratio does not say which is better, it says what kind of audience you had: a steady one that you can plan with, or a spiky one that arrived with an event.
In this example the ratio is between 1.26 (S-09) and 1.65 (S-08), a fairly steady channel. A ratio that suddenly jumps in one stream is worth a look: was there a raid, a mention, or a link posted on a busy page? Lesson 8 and lesson 17 come back to these questions.
Worked example: twelve Saturday streams
Here are the twelve fictional streams of Wavelark Studio's Bake-Along from April to June 2026, with the three counts you would export from your analytics and the three measures you calculate in this lesson's lab. The sheet in the lab above has exactly these first six columns.
| Stream | Date | Length | Unique viewers | Peak | Watch minutes | Average concurrent | Watch time per viewer | Peak ÷ average |
|---|---|---|---|---|---|---|---|---|
| S-01 | 4 Apr | 60 | 1,240 | 312 | 11,400 | 190.0 | 9.2 | 1.64 |
| S-02 | 11 Apr | 60 | 1,385 | 340 | 12,950 | 215.8 | 9.4 | 1.58 |
| S-03 | 18 Apr | 75 | 1,510 | 366 | 17,250 | 230.0 | 11.4 | 1.59 |
| S-04 | 25 Apr | 60 | 1,290 | 301 | 11,200 | 186.7 | 8.7 | 1.61 |
| S-05 | 2 May | 90 | 1,760 | 402 | 22,450 | 249.4 | 12.8 | 1.61 |
| S-06 | 9 May | 60 | 1,475 | 355 | 13,700 | 228.3 | 9.3 | 1.55 |
| S-07 | 16 May | 60 | 1,630 | 391 | 16,150 | 269.2 | 9.9 | 1.45 |
| S-08 | 23 May | 75 | 1,520 | 348 | 15,800 | 210.7 | 10.4 | 1.65 |
| S-09 | 30 May | 60 | 1,845 | 440 | 20,900 | 348.3 | 11.3 | 1.26 |
| S-10 | 6 Jun | 90 | 2,010 | 468 | 27,300 | 303.3 | 13.6 | 1.54 |
| S-11 | 13 Jun | 60 | 1,720 | 417 | 17,400 | 290.0 | 10.1 | 1.44 |
| S-12 | 20 Jun | 75 | 2,180 | 505 | 25,100 | 334.7 | 11.5 | 1.51 |
| Channel | 825 | 19,565 | 211,600 | 256.5 | 10.8 |
Three things stand out. First, the stream with the highest peak, S-12 with 505, is not the stream with the highest average: that is S-09 with 348.3 average concurrent viewers, against 334.7 for S-12. Second, the stream with the most watch minutes, S-10 with 27,300, is a 90-minute show, so part of its size is simply its length. Third, the lowest average concurrent is S-04 with 186.7.
Step by step in the spreadsheet
Open the lab above and follow these steps. Each step is checked by the Coach as soon as you type the formula.
- Click G4 and type
=F4/C4: watch minutes divided by the stream length gives the average concurrent viewers. Copy the formula down to G15. - Click H4 and type
=F4/D4: watch minutes divided by the unique viewers gives the average watch time per viewer in minutes. Copy it down to H15. - Click I4 and type
=E4/(F4/C4)(or=E4/G4): the peak divided by the average concurrent. Copy it down to I15. - In D17 type
=SUM(F4:F15)/SUM(D4:D15)for the channel's pooled average watch time per viewer. - In D18 find the stream with the highest average concurrent with
=INDEX(A4:A15,MATCH(MAX(F4:F15/C4:C15),F4:F15/C4:C15,0)).
| Cell | Formula | What it returns |
|---|---|---|
| G4 | =F4/C4 | Average concurrent viewers of a stream |
| H4 | =F4/D4 | Average watch time per viewer in minutes |
| I4 | =E4/(F4/C4) | Peak-to-average ratio |
| D17 | =SUM(F4:F15)/SUM(D4:D15) | Pooled channel watch time per viewer (10.82 minutes) |
| D18 | =INDEX(A4:A15,MATCH(MAX(F4:F15/C4:C15),F4:F15/C4:C15,0)) | Stream with the highest average concurrent (S-09) |
Pool the totals, do not average the ratios
For the whole channel, the average watch time per viewer is the total watch minutes divided by the total unique viewers: 211,600 ÷ 19,565 = 10.82 minutes. A tempting shortcut is to average the twelve values in column H. That gives 10.63 minutes, a slightly different number, because it treats the stream with 1,240 viewers exactly like the stream with 2,180 viewers.
The same rule applies to every ratio in this course: add the top and the bottom first, then divide. The pooled channel average concurrent is 211,600 ÷ 825 stream minutes = 256.5. The mistake is not dramatic with twelve similar streams, but it grows quickly when the streams differ a lot in size, as they do in a real channel with an occasional big collaboration.
Does a longer stream keep viewers longer?
Group the streams by length and pool each group. The seven 60-minute streams give 9.8 minutes of watch time per viewer, the three 75-minute streams 11.2 and the two 90-minute streams 13.2. Longer shows do hold people longer in minutes. But as a share of the show it goes down: 16.3% of a 60-minute stream, 14.9% of a 75-minute stream and 14.7% of a 90-minute stream.
This is a typical pattern, not a rule. With only two 90-minute streams you cannot conclude that 90 minutes is better; you can conclude that you should keep the length constant when you compare other changes, such as a new topic or a new time slot. Lesson 7 and lesson 12 give you the tools to compare slots and formats properly.
Common mistakes
- Quoting the peak as the audience. The peak is one moment. Report the average concurrent next to it, and the unique viewers when you talk about reach.
- Dividing by the wrong base. Watch minutes ÷ stream length gives concurrent viewers; watch minutes ÷ unique viewers gives watch time per viewer. The two formulas differ only in the divisor, which makes them easy to swap by accident.
- Averaging ratios. A simple average of per-stream ratios gives a small stream the same weight as a large one. Pool the totals.
- Comparing streams of different length by watch minutes. A longer stream produces more watch minutes automatically. Use the average concurrent.
- Mixing live and replay. Live watch minutes and replay views are different quantities. Lesson 6 treats them separately.
- Changing definitions mid-quarter. If the platform changes how it counts a viewer, mark the date in the sheet and do not compare across it.
How to read the result
For Wavelark Studio the picture is a channel that grew over the quarter. The average concurrent rose from 190 in the first stream to 335 in the last, and the pooled watch time per viewer is 10.8 minutes. The steadiest stream is S-09 (ratio 1.26), which means that the audience of S-09 stayed in the room better than the audience of any other show.
Resist the urge to turn this into a verdict on content. Twelve streams are a small sample, and the lab data do not tell you why one stream was larger. What they give you is a reliable baseline: the numbers you will compare against when you change the schedule, the format or the promotion in later lessons.
What the rest of the course covers
The remaining nineteen labs build on these three formulas. You will work through the following topics, each with a different fictional data set from the same company:
- Lessons 2 to 4: the minute-by-minute retention curve and the trapezoid estimate of watch minutes, the chat rate and the chatter share, and schedule consistency with gaps and late starts.
- Lessons 5 to 8: follower conversion per stream, live versus replay views, time-slot grids with AVERAGEIFS and raids with retention of the arriving viewers.
- Lessons 9 to 12: collab versus solo streams, a rolling baseline with lift signals, a growth trend with SLOPE and FORECAST.LINEAR, and a content-format scorecard.
- Lessons 13 to 16: viewer cohorts, stream quality and early drop-off, pre-stream promotion channels and live shopping conversion.
- Lessons 17 to 20: view quality checks, outliers and a t-test, a quarter planning grid and a final scorecard with charts.
Measure honestly
Audience numbers feed decisions about money, partnerships and people's time, so keep them honest. Do not buy views or chat activity, do not hide that a stream is sponsored, and give viewers clear information about offers and returns when a stream sells something. Segment audiences only by behaviour, such as how often they come and how long they stay, never by protected characteristics. Ask for consent before you message people about your next show.
A short checklist before you publish a number
- Which definition of viewer and of watch minute does the platform use, and is it the same as last month?
- Is the number a peak, an average or a total? Is the unit minutes or people?
- Are streams of different length compared with a length-independent measure?
- Did you pool totals instead of averaging ratios?
- Is there a baseline from your own earlier streams to compare with?
- Could one unusual stream (a raid, a collab, an outage) explain the change?
Now go back to the lab above, fill the sheet, and let the Coach check each formula. The next lesson zooms into a single stream and shows how the audience builds up and fades minute by minute.
Frequently asked questions
What is the difference between peak viewers and average concurrent viewers?
Peak viewers is the highest number of people watching at one moment. Average concurrent viewers is watch minutes divided by the stream length, so it shows how many people were in the room on average. The peak is easy to boost with a short spike, while the average describes the whole show.
How do I calculate average watch time per viewer in Excel?
Divide the total watch minutes of the stream by the number of unique viewers, for example =F4/D4, and copy the formula down. For the whole channel, divide the sum of watch minutes by the sum of unique viewers instead of averaging the individual results.
How do I get average concurrent viewers from watch minutes?
Divide the watch minutes by the stream length in minutes, for example =F4/C4. If 11,400 watch minutes were collected during a 60-minute show, the average concurrent audience was 190 viewers.
Why should I pool totals instead of averaging ratios?
An average of per-stream ratios counts a small stream as much as a large one. Adding the numerators and the denominators first and dividing once gives each viewer the same weight and is the correct channel-level figure.
What is a good peak-to-average ratio for a live stream?
There is no universal benchmark. A ratio near 1 means a steady audience, and a ratio of 2 or more means a short spike. Compare the ratio with your own earlier streams and look into the reason when it changes suddenly.
Where do I find unique viewers and watch minutes?
Most streaming platforms show them in the analytics section of the channel or in a downloadable report, often under names such as unique viewers, viewer minutes or minutes watched. Definitions differ, so read the help text and keep the same definition over time.
Keep going: Live Streaming & Audience Growth: Concurrent Viewers, Watch Time, Chat Rate, Schedule Consistency, Raids & Collabs, Follower Conversion and Live vs Replay
This free lab is lesson 1 of 20. Unlock the other 19 hands-on labs, the final exam and your certificate for a one-time $9.90.
- The Retention Curve: Viewers per Minute, Share of Peak, Drop-Off Steps and Watch Minutes by the Trapezoid Rule
- Chat Rate and Participation: Human Messages, Messages per 100 Viewers per Minute and the Chatter Share
- Schedule Consistency: Gaps Between Streams, Late Starts and Whether Regularity Pays Off
- Follower Conversion per Stream: New and Net Followers, Followers per 1,000 Viewers and Pooled Conversion
- Live versus Replay Views: Total Views, Live Share of Views and Watch Minutes, and Replay per Live Viewer by Format
- Time Slot Performance: Average Concurrent by Weekday and Slot with AVERAGEIFS, Stream Counts and Evening Uplift
- Raids and Hosts: Raid Retention, Share of the Room at Arrival, Followers per 100 Raid Viewers and Large versus Small Raids
- Collab versus Solo Streams: Lift in Average Concurrent, Follower Conversion Gain and New Viewers from a Guest
- Baseline and Lift: Rolling Four-Stream Baseline, Lift vs Baseline, Spike and Dip Signals with IFS
- Audience Growth Trend: Weekly Change, SLOPE, FORECAST.LINEAR and the Compound Weekly Growth Rate
- Content Format Scorecard: AVERAGEIFS by Format, Weighted Index Score with SUMPRODUCT, RANK.EQ and the Best Format
- Viewer Cohorts: One-Time, Returning and Loyal Viewers, Segment Limits with COUNTIFS and SUMIFS and the Share of Watch Time
- Stream Quality and Early Drop-Off: Buffering Share, CORREL, SLOPE, FORECAST.LINEAR and a Fix-First Rule
- Pre-Stream Promotion: Join Rate, Cost per Live Join, Share of Joins by Channel and the Blended Cost
- Live Shopping: Order Conversion, Average Order Value, Net Revenue After Returns and the Share of Discount-Code Orders
- View Quality Checks: Chatters per 100 Peak Viewers, Review Flags with AND, Plausible Peak and a Conditional Format
- Is It Real? Z-Scores and Outliers, CONFIDENCE.T Margin of Error and T.TEST for a Schedule Change
- Quarter Planning: Scenario Grid with Mixed References, Audience Fatigue, Cost per Net Follower and a Worth-It Verdict
- Final Project: A Quarterly Live Streaming Scorecard with Pooled Conversion, Weighted Audience, Chat Rate, Replay Share and a Verdict