How the leader track works
Six sessions, 45 minutes each, no code. The artifact you build across them is a host playbook you could hand to a new creator and defend to a finance director. Tonight is how to read a room's numbers. Session 2 is what the top streams share and why the first answer is a trap. Sessions 3 and 4 are the say and the show: what the script and the camera each add. Session 5 puts a nowcast in the producer's ear and lets you coach a replayed stream. Session 6 is how you prove a cue worked, and which urgency claims the platform and the regulator read as deception. The running example is Lumen Glow, a skincare brand selling one 29 dollar serum across 60 TikTok Shop LIVE streams, four human hosts and two AI avatars.
★ What you walk out with tonight. The vocabulary TikTok Shop uses in its LIVE analytics, so a report reads the same to you and to the platform. The funnel those numbers sit on, in the order the platform's own diagnosis page reads it. And one habit: before you praise or blame a host for a night's GMV, divide the traffic out. Across Lumen Glow's 60 streams, GMV tracks the size of the room at 0.96. GPM, GMV per thousand views, tracks the conversion rate at 0.86. One is mostly the platform's doing. The other is mostly the host's.
LiveWhy the biggest number on the dashboard is the wrong one to manage3 min▶
GMV is what the business is paid, so it is what the business looks at, and that is fine for a finance report. It is a poor number for managing a host, because the host controls perhaps a third of it. The platform decides how many people enter the room. The host decides what happens to them once they are in. Rewarding a host for a big GMV night rewards them for the platform's traffic decision; punishing them for a small one punishes them for a quiet Tuesday.
Every session in this track keeps two lines apart: what the room was given, and what the room did with it. The first is views, peak viewers, traffic source. The second is watch time, click rate, click-to-order rate, and GPM. A report that mixes them is not wrong so much as unmanageable, because nobody in it can tell what they are responsible for.
Two Lumen Glow streams, three weeks apart. Maya's Wednesday 21:00 (stream 17): 29,799 dollars, a room that peaked at 1,579. Priya's Monday 18:00 (stream 16): 11,217 dollars, a room that never passed 498. The review praised Maya. Divide by the traffic and Priya's GPM was 2,309 against Maya's 1,817, her click-to-order rate 14.0 percent against 11.6. She sold better to fewer people, and the review told her the opposite.
Self-studyThe vocabulary the rest of the track uses4 min▶
- Room - how many people are watching right now. TikTok Shop reports its peak (Peak Concurrent Viewers) and its average (Average Concurrent Viewers).
- Views and Viewers - views are entries into the room, viewers are unique people. Someone who leaves and comes back is two views and one viewer.
- Average watch duration - how long a viewer stays. TikTok's diagnosis page calls it average viewing duration per view. The room's verdict on whether the stream is worth staying for.
- Product impressions and clicks - how many times the pinned product card was shown, and how many times it was tapped. CTR is clicks over impressions.
- CTOR - click-to-order rate, orders over clicks. TikTok Shop's word for what the rest of e-commerce calls conversion. The number that says whether the pitch closed.
- GMV - gross merchandise value, what was paid before refunds. Watch GPM is GMV per thousand views: the traffic divided out.
- Segment - what the host is doing in a given minute: hook, product build, demo, price, CTA, urgency, Q&A, filler, dead air. Sessions 3 and 4 put a value on each.
Four stages, four questions
9 min liveTikTok Shop's own LIVE Diagnosis page reads a stream as four stages: did people tap in, did they stay, did they click the product, did they buy. Each stage has one number and one owner. Learn the order and a bad night stops being a mood and becomes a location.
LiveWhat each stage tells you, and what it cannot3 min▶
Each stage is a ratio, and each ratio answers exactly one question. Reading one as if it answered another is how a good host gets blamed for a slow feed.
| Stage | Number | Answers | Cannot tell you |
|---|---|---|---|
| Enter | Tap-through rate, views, peak room | Whether the feed and the cover pulled people in | Anything about the selling. A giveaway opener inflates this stage on purpose |
| Stay | Average viewing duration | Whether the room was worth staying in | Whether anyone bought. A stretched urgency pitch holds people and sells nothing |
| Click | CTR on the product card | Whether the pitch made the product interesting | Whether the price closed it. Also falls as watch time rises, because the card is shown every minute |
| Buy | CTOR, orders, GMV | Whether the price, proof and ask closed | How many people saw the offer. GMV is this stage times all the others |
The second column of "cannot tell you" is the whole leader track in miniature. Session 2 is the giveaway row. Session 3 is the urgency row. Session 4 is the click row. Session 5 is what you do about it while the stream is still running.
LiveThe platform's diagnosis labels, and how to read them3 min▶
TikTok Shop's LIVE Diagnosis stamps each stream with one of four labels. They are worth knowing because your hosts will see them before you do.
- Risk - the session has violation information attached. Not a performance label at all. It goes to the top of the list regardless of GMV, because violations cost traffic on future streams.
- Need attention - core metrics are lower than your own past performance. A drop against yourself.
- Can improve - core metrics are lower than the same-category benchmark. A gap against the market.
- Normal - no anomaly signals.
The useful reading is that "Need attention" and "Can improve" are different diagnoses. A host who has fallen against herself has changed something; a host who is below category has never had the thing the category has. The first is a conversation about last week. The second is a conversation about the script.
Ask for the stage, not the label. When a stream comes back "Can improve", the first question is which of the four stages is below benchmark. The platform's page shows the stages; the label alone sends everyone to work on the wrong thing.
Self-studyThe rest of the dashboard, briefly4 min▶
Beyond the four stages, TikTok Shop's LIVE analytics reports engagement (comments, likes, shares, new followers), a traffic-source split (Feed, Promoted, Follow), per-product impressions, clicks and sold quantity, and two rate-per-hour figures, GMV per hour and impressions per hour, for comparing streams of different length. The exact formula behind each number sits in a hover tooltip inside Seller Center rather than on any public page, so a report that quotes a definition should quote the tooltip.
Two readings worth having. Traffic source tells you how much of the entry stage was bought (Promoted), earned from the feed, or owed to existing followers, which is the only way to know whether a big room was a good cover or a paid campaign. New followers is the one engagement number with a future in it: a room that follows will be back for the next stream at no cost. Comments and likes matter mostly as what the platform reads early to decide whether to push the stream, which is exactly why session 2 exists.
Divide the traffic out
8 min liveGMV equals views times GPM over a thousand. That is not a metaphor, it is the arithmetic. Views are the platform's contribution and GPM is the room's. Once you see GMV as a product of the two, a stream review changes shape: you stop asking whether the night was good and start asking which factor moved.
LiveWhat GPM is good for, and what it is not4 min▶
GPM is GMV per thousand views, and TikTok Shop reports it as Watch GPM. It answers one question well: given the people who entered, how much did this room sell. That makes it the right number for comparing hosts, scripts and formats, because those things act on people already in the room.
| Question | Use | Why |
|---|---|---|
| Which host sells best? | GPM, watch time, CTOR | All three divide the traffic out |
| Which night made the most money? | GMV | Finance is paid in GMV, not in rates |
| Did the promotion pay? | Both, side by side | A promotion can raise views and lower GPM at once. Session 2 |
| Is the feed treating us well? | Views, traffic source, tap-through | The entry stage is the platform's |
| Should we run more streams? | GMV per hour | Length differs; the platform reports this one for that reason |
What GPM cannot do is tell you whether the traffic was good traffic. A room full of people who came for a free gift will show a lower GPM than the same host with the same script in front of people who came for the serum. GPM measures the room as it was given. Whether the giveaway that filled it was worth it is a separate question, and session 2 is where it gets its own answer.
LiveTwo grain traps that show up in leadership decks4 min▶
You will not compute these yourself, but you will be shown them, and you should recognise both on sight.
- "Total viewers" that is really viewer-minutes. Someone adds up the concurrent-viewer curve, sampled every minute, and reports the sum as the audience. It overstates the audience by a factor of the average watch time, which for Lumen Glow is about nine. A deck once claimed 74,000 viewers for a stream whose dashboard showed 8,200 views. Nobody lied; somebody summed a state.
- A rate averaged over minutes instead of computed over the stream. A minute with 3 clicks and 2 orders has a 67 percent click-to-order rate. Averaging those gives a number that swings with the quiet minutes. The dashboard computes orders over clicks for the whole stream, and so should any report that sits next to it.
Lumen Glow's agency report and TikTok's own analytics disagreed on watch time by 40 percent for a month. The agency had averaged the room-size curve, which is average room size, and labelled it average watch time. Same data, different denominator. It was found when a host asked why she was "losing viewers" on a night her CTOR had risen.
Self-studyImpressions are bigger than views, and that is fine3 min▶
The average Lumen Glow stream shows 9,049 views and 60,000 product impressions. The card is shown to each viewer roughly once a minute while they stay, so impressions scale with watch time, not with people. Two consequences for reading a report. First, CTR is a per-exposure rate and it falls as watch time rises even when nothing about the pitch changed, so a host who keeps people longer will look slightly worse on CTR for a good reason. Second, comparing CTR across hosts with very different watch times is unfair; the builder track uses orders per thousand viewer-minutes instead, which puts the room in the denominator. When you see CTR move, ask what watch time did first.
Your one-page room report
★ 20 min · everyone buildsTake the last five streams you are responsible for, open the seller dashboard, and fill one table. Two lines per stream: what the room was given, and what it did with it. Twenty minutes, no export, no spreadsheet formulas beyond one division.
List the last five streams with the numbers TikTok Shop reports. Views, peak concurrent viewers, average watch duration, CTR, CTOR, GMV, Watch GPM. Copy the platform's names exactly, so the report reads the same to you and to whoever checks it against Seller Center.
Rank the five by GMV, then rank them by GPM. Write both rankings side by side. If they match, the traffic was similar across the five nights and GMV happened to be a fair judge. If they do not, the nights that moved up on GPM are the ones where the host outperformed the room they were given.
For every stream, name the stage that was weakest. Enter, stay, click or buy, using the platform's four-stage diagnosis. Compare each stage against the host's own last month, not against the best night ever. A stage below the host's own norm is a conversation about what changed. A stage below the category benchmark is a conversation about the script.
Write one sentence per stream, in the two-line shape. "The platform sent X; the room did Y with it." Say which of the two moved. This is the sentence that goes to the host, and it is the first one that is fair to them.
Flag anything stamped Risk. A violation label costs future traffic regardless of how the night sold. It goes to the top of the report, above GMV, every time. Session 6 covers which claims earn the label.
Keep the report to one page and two lines. Everything the leader track adds later, the giveaway effect, the script values, the camera values, the live cues, fits into the "what the room did with it" line. The shape does not change; the line gets more precise.
Before session 2
25 min- Finish the room report for ten streams instead of five. Note which hosts move up when you rank by GPM instead of GMV. Bring the names.
- For your ten streams, write down whether each one opened with a giveaway, a flash sale, or neither. Do not analyse it yet. Session 2 does, and it will be more fun if you have guessed first.
- Ask one host what number they look at during a stream. Most say the viewer count. Write down what they said they do when it drops.
- Find the hover tooltip for Watch GPM in Seller Center and copy the exact definition into your report's footnote. Public pages name the metric; the tooltip holds the formula.
Three questions
Maya's stream made 29,799 dollars on 16,404 views. Priya's made 11,217 on 4,858 views. Who sold better to the room they were given?
GMV is views times GPM. Maya had over three times the traffic; Priya converted hers better, 14.0 percent click-to-order against 11.6. The dashboard's biggest number ranks the nights, not the hosts.
A report says a stream had 74,000 viewers. The seller dashboard shows 8,200 views. Which explanation fits?
Room size is a state. Adding it across minutes gives viewer-minutes, which is the audience times the average watch time, here about nine.
TikTok Shop's LIVE Diagnosis stamps a stream "Can improve". What is the right first question?
"Can improve" means below the same-category benchmark, as opposed to "Need attention", which means below the host's own past. Either way the label is a summary; the stage is the address.