Influencer Marketing ROI: Why the Average Lies

Search this topic and within two results you will meet the same statistic. Brands make $5.78 back for every $1 spent on influencer marketing. Sometimes it is $5.20, sometimes $6.50, sourced to a survey of marketers reporting their own results. Every guide leads with it, and it is the least useful number in the discipline.

I look at creator programs from the data side, and the reason that benchmark fails is not that it is wrong. It is that an average is a terrible description of this data. Creator outcomes are not clustered around a middle. They are a long tail with a few enormous winners, which means the average is pulled miles above anything you should expect from a given campaign. Quoting it to your CFO sets a bar that most individual campaigns will structurally miss, even when the program as a whole is working beautifully.

So this guide covers the formula quickly, because it is easy, and then spends its time on the part that actually decides whether your number means anything: the shape of the returns, and the three ways brands systematically mismeasure them.

The formula, and what belongs in it

There is no trick here.

ROI = (revenue attributed to creators minus what you spent) divided by what you spent.

A program that returned $18,000 on $6,000 spent is at 200%, or 3x, depending on which convention you like. Marketers usually say "3x return" and mean revenue divided by spend, which is technically ROAS rather than ROI. Pick one and be consistent, because the two differ by exactly 1x and the confusion shows up in every deck I have ever been sent.

The part people get wrong is the denominator. Costs are usually undercounted, because the visible ones are only half:

Count product at cost, not at retail. Valuing a $60 gifted item at $60 inflates your spend by whatever your margin is and quietly craters the ROI of every seeding program ever measured. This single convention explains a lot of the "gifting doesn't work" conclusions floating around.

A note on the metric everyone reaches for next: earned media value. EMV takes your impressions, multiplies them by a CPM you chose, and reports the result as money. It is not money. Nobody paid it, nobody banked it, and the number moves entirely with the CPM assumption you picked. Use it as a rough reach comparison if you must, and never put it in the numerator of an ROI calculation you plan to defend.

The number that breaks the formula

Here is the data that changed how I think about this. We track creator posts continuously, so we can look at what an individual post actually earns in reach. Restricted to creators between 1,000 and 250,000 followers, which is the tier brands actually seed and hire, this is the distribution.

Read the top and bottom rows together. The median post in that set gets 762 views. The average post gets 268,203. Both numbers are correct, and they differ by a factor of 352, because the top 1% of posts hold 57.8% of all the views in the set. Nine in ten posts never reach 100,000 views. Roughly one in 26 clears a million.

This is a power law, not a bell curve, and it changes what a forecast means. In a normal distribution the average is the thing to expect. In this one the average is an artifact of a handful of outliers, and the outcome you should actually expect from any single post is much closer to the median. The $5.78 benchmark is the same species of number: an average across programs where a few enormous wins do most of the lifting.

Everything difficult about influencer ROI follows from that one fact.

What a power law does to your decisions

Four consequences, and they are the practical content of this whole subject.

1. Judge the program, not the campaign. A single campaign is a sample of maybe five to fifteen posts from a distribution where most of the mass sits in the top few percent. It will usually come back below average, and occasionally come back at 20x. Neither result tells you much. The program-level number over a quarter or two is the one with enough draws in it to mean something.
2. Volume is a strategy, not a compromise. If outcomes are long-tailed, your odds of catching a winner scale with how many independent shots you take. This is the real argument for seeding wide and cheap rather than betting a quarter's budget on three expensive creators. More draws, same money.
3. Never kill a creator on one post. The same creator can post a 400-view video and a 900,000-view video in the same month, and the difference is mostly the algorithm's mood. One post is not a measurement of a person. Two or three is a signal worth acting on.
4. Expect a fat zero column and do not flinch at it. Most rows in a healthy creator program return roughly nothing. That is what the distribution looks like when it is working. The failure mode is not the zeros, it is having no winner in the set, and you can only tell those two situations apart if you are counting every row.

That last one leads directly to the first way brands get the number wrong.

Mismeasurement 1: the numerator is missing most of its posts

ROI is a fraction, and the fraction is only as good as its coverage. Most brands compute the numerator from the posts they know about, and the posts they know about are the ones that tagged them.

The mechanic is mundane. Someone buys or is gifted the product, films with it, says the brand name out loud or spells it without the @, and posts. Your mentions folder stays empty. The post is real, the audience is real, and any orders it drove land in your store as unattributed direct traffic.

Now put that next to the distribution above. If most of the return lives in a small number of outlier posts, and you can only see a small fraction of your posts, then the probability that your biggest winner is one of the posts you cannot see is uncomfortably high. That is not a rounding error in your ROI. It is the difference between a program you renew and a program you cancel.

The fix is not a better spreadsheet. It is watching creator handles and brand terms directly rather than waiting for a notification, which is what creator discovery does, and it is the same reason a seeding round has to be scored by detection rather than by tags.

Mismeasurement 2: you stopped counting too early

The second one is a timing error, and it is the reason so many programs read as break-even.

Campaign ROI is almost always calculated at the end of the campaign window, usually 30 days. Creator content does not respect that window. We snapshot tracked posts daily, so we can watch what happens to a post long after everyone stopped looking at it.

Nearly two thirds of the posts we observed were still accumulating views during the window, including 62% of those already more than three months past publishing. The typical gain is a slow drip rather than a second spike, and I would rather say that plainly than oversell it. But the direction is what matters for a measurement decision: a post is not finished when your campaign report is.

There is a bigger version of the same problem on the revenue side. A creator video seeds consideration that converts weeks later through search, through a saved cart, or through a friend who was shown the video. A 7-day attribution window catches almost none of that, and a 30-day window catches some. The shorter your window, the worse influencer marketing will look relative to paid search, every single time, which is roughly the whole story of why performance teams distrust it.

Two practical rules. Report an initial number at 30 days and a settled number at 90, and expect the second to be meaningfully higher. And keep the post-level view open indefinitely, because an asset that is still earning is an asset worth re-licensing or putting budget behind.

Mismeasurement 3: the return is attached to the campaign, not the post

The third error is structural, and it is the one that makes the first two unfixable when it is present.

Most creator programs measure at campaign grain. Spend goes in at the top, revenue comes out at the bottom, and the two are reconciled monthly. That tells you whether the month worked. It cannot tell you which creator, which post, or which piece of content earned it, so it cannot tell you what to do next.

The grain that works is one row per post, with the money on the row. Zeros included, because with a long-tailed distribution the zeros are most of the table and deleting them is how you convince yourself the average is the expectation. Once returns sit on individual posts, the decisions get easy and mostly mechanical: re-seed the creators who posted, ask for usage rights on the ones that converted, and put paid budget behind the proven winner rather than the one you liked.

That last move is where the math gets genuinely fun, because a post that already converted organically is the lowest-risk creative you can buy:

You are not gambling on whether the content works. You already watched it work, in the feed, with a real audience, and you have the orders to prove it. Boosting a proven post converts a long-tail outcome into something closer to a repeatable one, which is the only genuine way I know to beat the distribution instead of just enduring it.

What to actually report

Given all of the above, here is the reporting stack I would defend in front of a finance team. Four numbers, in this order.

1. Attributed revenue, at post grain, with the method named. Code redemptions and tracked links are hard signals; a post you matched to a sales lift is a soft one. Report them separately and add them to a total. A number whose composition is visible survives scrutiny; a single blended figure does not.
2. Cost per detected post. Total program spend divided by the number of posts you actually found. This is the single most useful operating metric in seeding, it is stable enough to forecast with, and it improves as your detection improves rather than as your luck improves.
3. Hit rate, defined up front. The share of posts clearing whatever bar you care about, whether that is 10,000 views or one attributed order. With a long-tailed distribution, hit rate plus median tells you far more than the mean ever will.
4. Repeat rate. How many creators posted more than once without being paid again.

That last one is the closest thing to a compounding asset in this channel, and it is measurable:

About one in six creators in our index has posted more than once. Every repeat post arrives at zero marginal cost, which means the true ROI of a seeding program is not the ROI of the round you just ran. It is the ROI of the round plus everything the roster posts afterwards, and a program measured campaign by campaign will never see that number at all.

How to calculate yours this week

You do not need software for the first version, and I would tell you to start without it.

1. Build one sheet, one row per post. Creator, platform, post date, link, views, and a column for orders. Not one row per campaign.
2. Add every cost, with product at COGS. Fees, product cost, shipping, rights, and an honest estimate of team hours. Undercounting spend produces a number nobody believes twice.
3. Give every creator a unique code and link. It is the cheapest hard attribution that exists, it is worth the discount margin purely as measurement, and without it you are inferring everything.
4. Search your own brand name weekly. Manually, on TikTok and Instagram, on the brand name plus obvious misspellings. Add every post you find to the sheet even when it is not tagged. Tedious, and it will surface posts you did not know existed.
5. Compute three numbers monthly: total spend, total attributed revenue, and cost per detected post. Recompute the previous month too, because it will have moved.
6. Review at 90 days, not 30. Look at the median row and the top row separately. Your decision about the program lives in the median; your decision about what to boost lives in the top.

This holds up to roughly 30 creators. It breaks at the weekly manual search first, which is the step that quietly stops happening in month three and takes your numerator down with it. That is the point where software that watches handles daily and lands orders on the post that earned them pays for itself, and yes, I work on one, so weigh that as you like. The method above does not change. Automation only changes whether the sheet is still true when nobody is maintaining it.

The short version

Influencer marketing ROI is revenue attributed to creators minus spend, over spend, with product counted at cost and earned media value kept out of it. That part is arithmetic.

The part that matters is that creator returns follow a power law. In our index the median post gets 762 views while the average gets 268,203, so the average is a description of the outliers rather than a forecast for your next campaign. Budget for many cheap shots instead of a few expensive ones, judge the program over a quarter rather than the campaign over a month, and stop reporting a mean without the median beside it.

Then fix the measurement itself. Count every post, including the 96% that never tag you, because with a long tail your biggest winner is disproportionately likely to be one you cannot see. Wait 90 days, not 30. And keep the money on the post rather than on the campaign, because the post is the unit that repeats, gets re-licensed, and gets boosted. Do those three things and the number you report stops being an industry benchmark you borrowed and starts being something you can act on.