How to Measure Influencer Campaigns in the Absence of Full Attribution

“Our CPI is WHAT?” I said.

We had just run our first influencer marketing campaign – and the CPIs I was seeing for these influencers were in the double digits, many times that of other sources for our app.

I was shocked – and embarrassed. Nearly all the influencers we had used had shown a similar performance. I had been very excited about testing influencers – because, wasn’t that what all the cool kids were doing & being successful with?

My then boss said by way of consolation – “oh well, at least those videos will stay forever on the internet. And lots of people might just view the video and download the app later.”

“Sure,” I said. Yep, that’s how Youtube worked. That wouldn’t make our numbers look any better though.

So I decided not to run any more influencer marketing campaigns for the product at the time.

I was SO stupid.

The big challenge with attributing influencer campaigns

It was only later, when I learnt to run influencer campaigns the right way did I realize that we’d measured the performance all wrong in my first attempt (in all fairness, I had no idea).

The big challenge with influencer campaigns is that they aren’t your directly attributable campaigns that you’d have from your other sources. Some of your users coming from influencer campaigns click on links, sure – but most of them will watch the video, and search for your app separately, and then install, thus showing up as ‘organic’.

After my first attempt, I learned that just because these campaigns aren’t directly attributable, didn’t mean we could do nothing about it. I learned that we could absolutely make intelligent assumptions to arrive at a very good estimate of influencers’ performance.

Here is the approach I learnt and put to use, and you can use too.

How to solve for incomplete attribution

Each influencer will have ‘directly attributed’ installs (i.e. – installs that are attributable directly to them via an attribution link) – in addition to organic installs that resulted from the influencer (i.e. – users who viewed the influencer’s creative and installed without clicking on the link).

While these ‘organic’ installs can’t be directly tracked, it absolutely is possible to get a very good estimate of these numbers.

In order to estimate organic installs from each influencer, here is a process you can follow:

Export hourly data from your MMP to identify a baseline organic number for your ‘measurement window’, a time frame when you expect most installs to come from an influencer after they go live. A baseline number is the number above which your organics go once the influencer goes live.I typically recommend a 4 hour time frame as a measurement window because that is when most direct installs from influencers come before the organic lift from them starts to subside into noise. For instance, if the influencer is going live at 4pm, identify baseline organic numbers from 4pm to 8pm for the same day of the week for at least the last 4 weeks). That said, if a different measurement window makes more sense for your biz – 2 hours or 6 hours or 1 day, you should feel free to use that.

After the influencer goes live, calculate the ratio between organic lift to directly attributed installs for the first 4 hours after install(let’s call this R – organic ratio).R = (Organics in 4 hours after going live – organic baseline)/(Direct installs in the first 4 hours after going live)

Report directly attributed installs for each influencer after whatever tracking window you’re measuring(we typically take 14 days).
Calculate your total attributable installs = (1 + R) * Directly attributed installs.

How it works in practice

Here is an example of how this calculation would work in practice. You can also refer to this handy spreadsheet which you can make a copy of and customize as needed.

Pre-campaign

Influencer A is scheduled to go live on Saturday 10/27/2018 at 7pm UTC
Our organic baseline calculation period is from 7pm UTC to 11pm UTC.
Your organic baseline (B) = Average of the sum of organics from 7pm UTC to 11pm UTC for the 4 preceding Saturdays. Let’s say this is 200 installs.

Campaign-day

It’s Saturday 10/27 at 7pm UTC. Influencer A goes live.
At 11pm UTC, you measure your:
- Direct installs attributed to the link used by influencer A(call this ‘D’). Say this is 150 installs.
- Total organic installs your app got during this 4 hour window(‘O’). Say this is 400 installs.

– Your organic lift(L) is calculated as ‘total organic installs in the first 4 hours’ – ‘organic baseline’(O – B above). This would be (400 – 200) above = 200 installs.

– Your organic lift ratio is calculated as R = Organic lift(L)/Direct installs(D) = 200/150 = 1.33

Post-campaign

Your campaign ends after your tracking window (typically 14 days after the influencer goes live) – so this would be on 11/10/2018.
On 11/10/2018, you track your final directly attributed installs. Call this D1, let this be say 600 installs.
Your total installs (directly attributed installs and their organic lift), is now calculated by simply assuming that all the installs you get in 14 days have the same organic lift as the ones you had in the first 4 hours. So total installs (T) = (1 + R)*D1In the case of our example, T = 600 * (1 + 1.33)
T = 1400 installs

Your actual eCPI = Cost/T
Assuming you spent $1400 on the campaign,
Your eCPI = $1

That gives you about as accurate a measure of the actual performance of your campaigns as possible given the uncertain & only partially quantifiable nature of influencer marketing.

And learning that process of course explained to me what I had missed when I had done it wrong. I had taken direct installs after the first day(rather than 14 days), and assumed no organic lift and used it to calculate CPIs.

In the above case, we might have had say 300 directly attributed installs after the first day, so we’d calculate a CPI of $4+ rather than the actual CPI of $1.

I messed up back then, but learnt from it. Thankfully.

I hope your learning process isn’t as hard as mine – and that you’re able to use the above methodology and spreadsheet to estimate influencer performance accurately.

Good luck.

[With much thanks to Adam Hadi for showing me many of the nuances of influencer marketing].