The TikTok Shop Growth Doctrine

Every number in this document is measured from Euka unless it is marked as derived. The primary dataset is Brand A, an enterprise consumer brand, February 25 to August 25 2026, with Brand B, a national wellness brand, and Brand C, a consumer wellness brand, where noted. Where a claim cannot be supported by measurement, it is marked as a hypothesis or it is not here.

The identity

Growth on TikTok Shop is converting videos per sample. Everything else is a way of moving that number.

Sample Yield

Sample Yield = post rate × videos per posting creator × video conversion rate

Brand A over six months: 0.757 × 3.18 × 0.1103 = 0.265. Twenty one thousand nine hundred and twenty nine samples produced 5,815 converting videos. It reconciles exactly, which means the three terms on the right are the complete set of levers. There is no fourth thing.

Read the terms.

Post rate is not a lever. It runs 75.7 percent at Brand A and between 68 and 85 percent at every single creator tier. Brand B runs 91 to 98 percent across three different challenge structures. Getting a creator to post is a solved problem and every hour spent on it is an hour spent on the one term that does not move.

Videos per posting creator is a lever, and incentives move it.

Video conversion rate is a lever, and targeting moves it. It ranges from 8.90 percent for L1 and L2 creators to 28.59 percent for L3 and above, a 3.2x spread that is decided before a message is sent.

Community does not appear in this equation. Neither does Discord, engagement, sentiment, or roster size. That is not a rhetorical point. It is the reason this document reorders almost everything.

The second thing the identity does is expose a trap. Terms two and three are not independent. Paying a creator for video count buys volume and sells conversion rate, because effort per video is elastic and a creator optimising for a count will find the cheapest way to produce one. Brand A went from 4,592 creators at 19.3 percent video conversion to 14,251 creators at 11.0 percent. Volume tripled and quality fell 43 percent.

So the thesis in one sentence: growth is converting videos per sample, the two levers are targeting and volume, and the entire craft is raising volume without selling conversion rate to do it.

Part 1. Five definitions, in causal order

TikTok Shop

TikTok Shop is not a storefront. Listings do not sell. Content sells, and content only scales after it proves it converts organically and gets flighted into GMV Max where paid distribution amplifies what already works.

That makes it a distribution auction that runs on content supply. The constraint is not price, listing quality, or ad budget. It is converting content, and the only way to produce converting content is to put product in a creator's hands.

Every dollar is downstream of a sample. That is the physical law of the channel.

Creator recruitment

Recruitment is not outreach volume and it is not reply rate. It is getting product into the hands of a creator who has a reason to post, and it is not complete until both are true.

Recruitment is also, at Brand A, the binding constraint, and the previous version of this doctrine did not mention it.

L3 and above received 218,311 invites over six months and produced 1,952 sample requests. That is 0.894 percent. The tier that returns $397 per sample answers less than one time in a hundred.

The reason is not volume. Invites per creator by tier run L1 5.2, L2 13.5, L3 20.9, L4 24.3, L5 23.6, L6 25.1. Invite frequency rises with tier while response falls. An L4 creator received 24 invites in six months and ignored 99.15 percent of them. An L1 creator received five and responded seven times better. We are saturating the tier that matters and under-offering it.

Recruitment sets the ceiling. It does not raise output on its own, but nothing downstream can exceed it.

Creator retention

Retention is not a creator staying in a Discord and it is not a creator answering messages. Retention is a creator posting again.

It is governed by the speed and certainty with which effort becomes income. Fast and certain retains. Slow and probabilistic churns.

The measurement is unambiguous. Month one retention across six Brand A cohorts averages 26.4 percent. By month six it is 9.0 percent. Returning creators are 12.7 percent of the active base, which means the program is running almost entirely on new supply.

And the reason is arithmetic rather than sentiment. The mean converting creator produced $504 of GMV over six months. Excluding the top fifty, $257, which is roughly $8.57 a month in commission at a 20 percent rate. Nobody stays for $8.57.

Retention is higher leverage than recruitment per creator, and recruitment sets the ceiling retention works under. L3+ retains at 49.4 percent against 32.6 percent overall, at month six 25.7 percent against 9.0, and produces $680.64 per creator against $51.15 for everyone else. That is 13.3x, and it compounds because they stay.

Incentives

An incentive is a price paid to buy a specific behaviour. Every incentive should be able to name the behaviour it buys.

Three rules.

Pay cash for effort and percentage for outcome. Never percentage for effort. A creator who posts ten videos and sells nothing, earning a higher percentage of nothing, has not been incentivised.

Never pay more per video than the video produces. Ten dollars a video is 103.8 percent of what an L1 video generates, 29.5 percent at L2, 15.5 percent at L3, 7.5 percent at L4, and 5.2 percent at L5. The same offer is value-destructive at one end of the ladder and nearly free at the other. Run this test before designing anything.

Gate the effort or the incentive eats the conversion rate. This is the rule the previous version of this doctrine did not have and it is the most important one in the document. A three second product shot over trending audio and an hour of crafted storytelling both count as one video and both clear a 95 percent post rate. Only one converts. A per-video price with no floor manufactures the first kind.

Euka enforces a minimum GMV floor on posting challenges and forbids setting it to zero or a dollar, with ranges from $20 to $300 depending on shop size. That is not an anti-fraud nuisance. It is the effort gate, and it is the mechanism that stops the incentive from destroying the third term of the identity.

The brief serves the same function. Its first job is to hold effort per video up. Making content flightable to GMV Max and licensable to Meta is what pays for that effort, which is why the brief and the multiplier stack are one mechanism, not two.

Community

Community is not a Discord, though it can be hosted in one. It is not relationships, though it requires them. It is not campaigns.

The previous version of this doctrine said community is the mechanism to retain the affiliates you recruit. That is wrong and the measurement is not close.

Discord reaches 4.37 percent of sampled creators at Brand A, 3.23 percent at Brand B, and 2.53 percent at Brand C. Discord-linked creators account for between 3.1 and 6.3 percent of all sale-driving creators. A channel touching one sampled creator in thirty cannot be the retention mechanism for a program.

More precisely, Community Lift by funnel stage at Brand A runs 1.85 at contact to request, 1.28 at request to receive, 0.94 at receive to post, and 1.30 at post to sale. Discord creators convert better at entry and better at sale, and they do not post more. The stage where community was supposed to produce content is the one stage where it produces nothing. That pattern holds on all three brands, at 0.94, 1.00, and 0.96.

End to end the lift is 2.90x, 4.49x, and 2.71x. It cannot be shown to be causal. The Euka filter returns creators linked through Discord at any point with no join date, so a creator pulled into an ambassador channel after driving $20,000 sits in that cohort and inflates it. At 3 percent penetration, selection is the more likely explanation.

Money is the retention mechanism. Community is a distribution channel to a small, high-value subset, and it belongs in this doctrine as an ascension venue rather than a program-wide system.

One related correction. The previous version of this doctrine argued that one-to-many communication gives the team leverage. Reply rates say the opposite. Brand A's monthly challenge broadcasts reply at 0.12 to 0.68 percent. Triggered sample-approval messages reply at 3.9 to 8.8 percent. Recovery messages reply at 13.6 percent. The leverage is in the trigger, not the breadth, and one-to-many is 10 to 40x worse than the moment.

Part 2. Leverage, defined three ways

Three separate accounts that have to move together.

Creator leverage is earnings per unit of creator effort. A creator gains leverage when the same video earns more, not when they are asked to make more videos.

Brand leverage is converting videos per sample. Sample Yield, measured at 0.265 at Brand A, meaning 3.8 samples per converting video.

Team leverage is GMV per hour of management.

The binding rule: a creator only unlocks leverage by producing brand leverage. Tiers are purchased with outcomes. They are never granted for tenure, engagement, or attendance. When multipliers get handed out for loyalty, we pay more for the same result and the ladder stops meaning anything.

The test every tier benefit must pass: each rung pays more for identical effort. That is now measurable rather than asserted.

The four multipliers on one video

One brief-compliant video can be paid four times. On the post, through challenge cash tied to output. On organic conversion, through affiliate commission. Again when it converts well enough to be flighted into GMV Max, where the creator earns on ad-driven sales. And a fourth time when the asset is licensed off platform and run on Meta.

The fourth is the largest. A brand spending $100,000 a month on Meta and $10,000 on TikTok ads is deploying the same asset against ten times the media weight off platform. The creator's ceiling on that video is set by total media budget, not TikTok GMV.

Pay the Meta share as a percentage of attributable revenue, never as a percentage of spend. Percent of spend pays the creator more when the ad performs worse and caps the client's willingness to scale.

This is also how the effort problem gets solved in the right direction. A creator will put more into one video if that video has four ways to pay. The incentive and the outcome point the same way.

One honest limit. Flight rate is not measurable at Brand A. Every contest campaign returns zero attributed videos and zero revenue, so the share of converting content actually reaching GMV Max and Meta is unknown. The multiplier stack is structurally sound and empirically unverified. Treat it as the highest-priority instrumentation gap in the program.

Part 3. The tier model

The tier is the unit of analysis, not the creator. Every number that matters moves by 3x to 202x across the ladder, and any strategy that treats creators as interchangeable will be wrong at both ends.

Two ladders, kept separate

The platform ladder is Euka's L1 through L7, set by a creator's rolling 30-day GMV across all of TikTok Shop. It determines conversion rate, sample value, and therefore what we can afford to pay. We do not control it and we cannot move a creator up it alone.

The brand ladder is what a creator climbs with us: sampled, active, ambassador, VIP. We control it entirely.

The previous version of this doctrine conflated these. The $500 threshold it used as a VIP gate is brand GMV, and at Brand A the mean converting creator produced $504 over six months, so $500 describes a median converter rather than a VIP.

The measured economics

L6 had 13 delivered samples and L7 had zero videos. Those are raw counts and they do not support a rate.

A sample sent to an L6 creator returns 202 times what a sample sent to an L1 returns.

Three findings that follow

Allocation is inverted against value. L1 absorbs 76.5 percent of samples and produces 29.5 percent of GMV. L4 through L6 take 2.7 percent of samples and produce 32.5 percent.

Approval effort is inverted against value. We approve 39.1 percent of L1 requests, 78.1 percent of L3, and 77.9 percent of L4. We filter hardest where a sample is worth $22 and barely filter where it is worth $499. A rejected L1 request costs almost nothing. A rejected L4 request costs a $499 opportunity.

Concentration is more extreme than the rule of thumb. The previous version of this doctrine said the top 5 percent drive most GMV. Brand A's top fifty creators are 0.35 percent of 14,249 active creators and produced $653,924 of $1,308,098. Half the revenue from one creator in three hundred. L3 and above are 6.5 percent of creators and 47.9 percent of GMV.

Part 4. Activation

Three parties mean three different things by this word, and the confusion has cost us.

To the creator, activation is the first dollar. Nothing before it is real.

To the brand, activation is the first converting video. The moment a creator starts paying back the sample.

To the team, activation must be something we control. Per-video conversion is 11.03 percent at Brand A, so whether any specific video converts is not something a strategist influences.

So: activation is a sampled creator reaching that tier's conversion threshold inside a defined window. The team KPI is the percentage of sampled creators reaching threshold, because it is the only one of the three fully in our hands and it mechanically produces the other two.

The threshold is a computed number, not a round one

Conversion threshold

Conversion threshold = 1 ÷ per-video conversion rate.

L1 and L2 convert at 8.90 percent, so 11.24 videos. L3 and above convert at 28.59 percent, so 3.50 videos. Euka's tier filter is binary, so no finer split is measured, and any per-tier breakdown below this is derived rather than measured.

At 11.24 videos, a creator posting ten has a 68.9 percent chance of at least one conversion. Posting three has 29.6 percent.

Post rate is not activation, and mistaking them hides everything

Post rate runs 68 to 85 percent at every tier. It looks flat because posting once is easy. Set it against the threshold and the picture inverts.

The average L1 or L2 creator posts less than a third of what they need to convert. The average L3+ creator posts almost double. Estimated share reaching threshold is roughly 3 percent against 65 percent, a twentyfold gap that post rate conceals completely.

Three gates

Ignition is sample delivered to first video. Owned by the brief and the challenge launch.

Threshold is first video to Nth video. This is the gate ZeroTo1 owns, because it is the only one fully in our control and it determines the third.

Conversion is the first dollar. Downstream of Threshold and largely automatic once it is cleared.

Time is a lever inside Threshold and should be priced rather than gated. Ten videos in seven days pays more than ten in fourteen, which pays more than ten in thirty. Everyone who does the work gets paid and speed earns a premium.

Part 5. The incentive architecture

Two gates before any design

Affordability. Per-video payment divided by GMV per video for that tier. Above roughly 10 percent the offer is expensive. Above 100 percent it destroys value.

Effort. A minimum GMV floor on every output-based incentive, scaled with the tier. Without it the incentive buys checkbox content and sells conversion rate.

Offer economics by tier

Threshold is that tier's conversion threshold. Expected GMV is threshold times conversion rate times dollars per converting video. These are derived from the measured L3+ aggregate split across L3, L4, and L5, so treat them as estimates.

You can pay an L4 creator $33 a video and return 2.8 times what a $10 per video L1 challenge returns.

The threshold evidence

Brand B launched three commission-ladder campaigns on the same day, January 13 2026. It is the only near-controlled threshold comparison in the portfolio.

Two readings. Acceptance did not fall as the threshold rose, 17.6 percent at three videos and 17.8 percent at ten, which kills the assumption that a harder ask suppresses takers. And output per sample rose tenfold, from 4.13 to 41.12 videos per unit of product cost.

The segments differ, so selection is likely and the 10-video cohort was probably higher quality to begin with. The sample is 26 units. Do not build a forecast on it. Do treat it as the strongest available evidence that high thresholds outperform low ones.

Part 6. Two journeys

Activation is not what separates the tiers. Everyone posts. What differs is where each population breaks, and it breaks in two different places, so there are two journeys rather than one journey with tiered numbers.

Journey A. L1 and L2, who break at volume

They post 3.53 videos against a threshold of 11.24. They fail a bar they were never shown.

Sample, with the threshold and the money stated in the same message that confirms the sample. Ten videos, paid per video, with a minimum GMV floor scaled to the tier. Speed priced in tiers rather than gated. Brief attached, because it is the effort floor. Payment on completion, honoured without exception.

Constraint: their videos are worth $9.63, so the per-video price has to sit near or below a dollar to pass the affordability gate, and the money has to come mostly from the completion bonus rather than the per-video rate. This tier cannot carry an expensive challenge and the honest fix for L1 is sample allocation, not a cheaper incentive.

Journey B. L3 and above, who break at the request

They post 178 percent of their threshold once product is in hand. Nothing after the sample is broken for them.

The entire intervention is the invite. A guaranteed dollar figure in the target collab message, not a commission percentage, because a creator doing $100,000 a month does not read a percentage as an offer. Auto-approval, because a seven-day review buys nothing at 78 percent approval and is disqualifying at that tier. Then the challenge that the invite promised, at that tier's threshold of three to six videos rather than ten, honoured in full.

Using the offer to break the request constraint and then not delivering it would spend the only thing this tier has that L1 does not, which is a reason to open the next invite.

Part 7. Campaigns

A campaign is the delivery vehicle for the incentive architecture. It is not a parallel initiative. Everything above is inert until a campaign carries it to creators.

What a campaign buys. Concentration of content in a narrow window, which is what the algorithm rewards. A competitive dynamic at no marginal cost when structured as a leaderboard. Public visibility on the Euka creator app, which makes it an acquisition channel and not only a re-engagement tool. And an operational moment where a threshold and an effort gate can actually be enforced.

The leverage points, in order of size

Audience composition, and it is currently unmanaged. Brand A's July Content Challenge reached 13,707 L1 creators, 1,969 L2, 565 L3, 288 L4, 67 L5, and 8 L6. 94.4 percent of the audience sat in the two tiers where a video is worth $9.63 and $33.92. No tier structure inside the campaign fixes an audience allocated that way. Who receives a campaign matters more than what it offers.

Threshold setting. The campaign is where the conversion threshold becomes a number a creator can see. See the Brand B ladder above.

The effort gate. The minimum GMV floor is set here or it is not set at all.

Ascension. A campaign can move creators between tiers rather than only extract output where they sit. Split the budget explicitly. Performance allocation to the tiers with the lowest cost per converting video. Ascension allocation to creators near the top of their current tier, where the distance to the next rung is shortest.

Sequencing. Launch the campaign first, then run outreach into it. Outreach is the promotion engine and there has to be something to promote.

Capacity

Campaign capacity is retained creators multiplied by participation rate. Euka's measured opt-in rate is 0.7 to 5 percent of eligible creators, with 2 percent as the working estimate. At Brand A that is roughly 73 creators a month against 3,650 samples shipped.

Euka contest budgets at a shop Brand A's size run $1,000 to $5,000, which buys dozens of creators. A campaign is a re-engagement instrument. A compensation restructure across thousands of creators is a different instrument. The contest engine will run a pilot well and will not carry a rollout, and any proposal has to say plainly which of the two it is.

What a campaign costs that is not money

Euka enforces a five-message cap per creator. Every broadcast spends a slot.

Brand B spends two of five slots on shipping notifications that reply at 0.8 and 1.3 percent, while their re-engagement message replies at 12.4 percent and their showcasing follow-up at 10.9. Forty percent of the finite communication budget with a creator is spent on messages that ask for nothing. A VIP Discord invite sent to $500 GMV creators returned zero replies on 67 sends.

Message slots are the scarcest resource in the program and they are currently the least managed.

Two hard limits

The claim that campaigns are 60 percent or more of growth on the platform is unverified. It is a hypothesis and this document will not repeat it as fact.

Campaign effectiveness has never been measurable at Brand A. Every contest returns zero attributed videos and zero revenue. Until that is fixed, measure by matched before-and-after windows at 45 and 90 days, name the confound when campaigns overlap, and understand that a lift which reverts by day 90 pulled output forward rather than creating it.

Part 8. The order of operations

The previous version of this doctrine implied recruit, retain, build community, run campaigns. The evidence supports close to the inverse in emphasis.

One. Decide who gets a sample. Targeting sets video conversion rate, one of only two real levers, and it is decided before any message goes out. This is the highest-leverage single decision in the program and it is currently being made by default.

Two. Get the high-value tiers to accept. A guaranteed dollar figure in the invite. This is the binding constraint at 0.894 percent.

Three. Remove approval delay for the tiers you chose. Seven days buys nothing at 78 percent approval.

Four. Set the threshold from that tier's measured conversion rate. Not a round number. One divided by the conversion rate.

1. Five. Pay certainly at the threshold, with an effort gate, and honour what the invite promised.

Six. Stack the multipliers on the same asset. Flight converters into GMV Max within 72 hours and tell the creator. License the strongest to Meta on revenue share.

1. Seven. Fund ascension separately from performance.
2. Eight. Invite the retained subset into community, last.

Steps one and two are where almost all the available leverage sits, and they are the two steps that received the least attention in the previous version of this doctrine.

Part 9. The metrics

Retire as goals: community size, Discord member count, creators recruited, videos posted. None are causally connected to new customers acquired.

Run on these.

Sample Yield, converting videos per sample. The master metric. Brand A 0.265.

Paid Population Rate, the share of sample recipients who receive any money. 18.2 percent under sale-triggered pay, 98.4 under post-triggered.

Activation rate, the share of sampled creators reaching that tier's threshold inside the window. Not post rate.

Time to first dollar, from sample delivery to money in hand.

GMV per sample by tier, which is the allocation scorecard.

Video conversion rate by tier, which is the early warning on whether an incentive is buying checkbox content.

Cost per converting video by tier, which ranks challenge budget.

Second post rate, the share of first-post creators who post again within 30 days.

Flight rate, converting videos reaching GMV Max divided by converting videos. Currently unmeasurable and the top instrumentation priority.

Reply rate per message slot, because slots are scarce and currently unmanaged.

New customers and GMV remain the client outcome. They are poor diagnostics, because by the time they move the decisions that caused it were made 60 days earlier.

Part 10. The case against this doctrine

The strongest counter-argument is about L1, and it deserves a real hearing.

L1 produced $376,028 of GMV over six months. That is 29.5 percent of the total and it would not exist if those samples went elsewhere. More importantly, L1 is where L3s come from. Every creator in L4 today was in L1 once. A doctrine that reallocates samples toward the top may harvest the existing top while starving the pipeline that replenishes it, and the effect would not show up for two or three quarters.

The counter to the counter is that we do not choose the current mix. L1 requests at 6.48 percent and L3+ at 0.894, so the 76.5 percent allocation is what happens when only one tier answers. Fixing the L3+ response rate adds samples at the top rather than subtracting them from the bottom. But that is an argument for a test, not a proof, and until the test runs the pipeline objection stands.

Two more honest weaknesses. The L3+ request rate may be a hard ceiling rather than an offer problem, because those creators are courted by every brand on the platform and 218,311 invites at 0.894 percent is a large sample to be wrong about. And the Brand B threshold evidence rests on 26 delivered samples in the ten-video arm with likely segment selection, which is thin ground for the most consequential design choice in the document.

Part 11. What would falsify this

If receive-to-post falls below 90 percent at any brand, then effort is a real friction point, the two-journey split is wrong, and the effort-gate argument needs rebuilding. Test by running the outreach funnel on five more brands.

If a guaranteed-offer invite does not lift the L3+ request rate, then recruitment is not the binding constraint, the current sample allocation is rational, and the order of operations inverts back toward activation. Test with a two-arm invite split on 12,000 L3+ creators, break-even at a 1.27 percent request rate against today's 0.894.

If video conversion rate holds steady when added volume comes from retained creators rather than new ones, then the volume-quality trade-off is an acquisition artifact, the effort gate is unnecessary, and per-video pay is safe without a floor. Test by splitting conversion rate by new against returning creators.

Part 12. What this changes

Sample allocation becomes a managed decision. It is currently an outcome of who answers. Every account should have a target mix and report against it.

The offer moves into the invite for L3 and above. A guaranteed figure, not a commission percentage.

L3+ sample requests get auto-approved. The review is approving 78 percent anyway.

Every incentive gets an affordability test and an effort gate before it launches. No exceptions, and any tier failing affordability is removed from the audience rather than given a smaller tier.

Discord moves behind the first payment and stops being reported as a growth metric. It is an ascension venue for a small subset.

Message slots get audited. Five per creator, and any slot replying under 2 percent has to justify itself.

Contest attribution gets fixed before another contest is funded. Zero attributed videos across every campaign means we have never known whether any of this works.

What remains unmeasured

Converting videos per individual tier, because Euka's filter is binary at L3+.

Flight rate into GMV Max and Meta, so the multiplier stack is structurally sound and empirically unverified.

Whether Discord lift is causal, which needs join dates relative to first sale.

Effort per video, which is unobservable and only proxied by GMV per video blended with creator quality.

Whether contests work at all, which needs attribution before it needs a budget.

Post-conversion retention, meaning what happens to a creator after their first dollar, which needs creator-level event data that the current tooling does not expose.