During the past few months install fraud has become a hot topic in the mobile marketing industry. Third-party trackers and media partners are announcing measures to prevent it as if it was something new. Fraud, however, has always been here. It was there during the desktop era and it is here now in the mobile marketing world. Actually, it has become more intense and more sophisticated over the past 2 years.
When talking about fraud in mobile advertising, I’m talking not only about IP fraud, proxies, country mismatch, and abnormal user patterns… but also about bad practices such as delivering incentivised installs for non-incent offers.
A recent article on VentureBeat estimated that mobile advertisers will lose $100 million to fraud this year. So what can we, as mobile marketers, do about it? Can we prevent it? How can we spot it? Here are some tips for spotting and preventing mobile ad fraud:
1) If it looks too good to be true, it probably is
Check for partners or sub-sources with extremely good performance or abnormally high event rates. Dig up and investigate these installs, they probably came from bots and fake users.
2) Keep all your KPIs present
When running UA campaigns, it is good to simplify the KPIs you measure so you don’t lose focus. This is definitely a good practice but don’t lose sight of the rest of your KPIs either. If possible, create automated alerts when some KPIs increase or decrease by a significant percentage compared to their averages. For instance, if your app offers sign-ups with Facebook, Google and email, check for an abnormal rate of users logging in via email as this is much easier to fake.
I also like to keep some secondary KPIs private and don’t share them with my partners. If all KPIs look good except one, this might be an indicator of more sophisticated fraud happening.
3) Retention rates are key
One of the easiest ways to spot fraud and incentivised installs is by looking at retention rates. Look at them with as much granularity as possible and compare it to a media channel you know is legitimate.
Extremely low retention rates (many times starting on day 1) are an indicator of bots and incentivised installs. Extremely high retention rates or strange-shaped retention curves may also be caused by fraudsters.
4) Check for very low install rates
Install rates below 0.1% can be indicators of click fraud. Another article from earlier this year suggested that an increasingly deployed fraudulent tactic involves playing with fingerprinting algorithms. In summary, fraudsters send lists of millions of devices as clicks hoping to match an organic install, and therefore attributing them as theirs. This practice has a bigger impact on popular apps with lots of organic traffic.
Another way to check on this is by looking at the delay between attributed click time and install time. A regular user usually downloads an app just a few seconds or minutes after clicking on an ad. However, if there is click fraud involved, this gap can be hours or even days.
5) Audit different samples of users periodically
If you suspect a specific source or sub-source is delivering fraudulent installs, run a check of a sample of 10 or 15 individual installs. Although time consuming, this will probably confirm your suspicions.
6) Finally, look at fraud-detection tools
Have a conversation with your third-party tracker about which fraud prevention measures they take. The best measures are those taken before the install is attributed.
Look at fraud detection tools that allow you to detect install fraud after it’s attributed. Some of these products include: Forensiq, 24metrics or Datavisor. All of them run pretty advanced algorithms that detect “high risk” installs with lots of accuracy.
Mobile fraud is an industry’s disease. As with any other disease, the best solution is prevention. Working with trusted partners with direct traffic and direct sources seems like the ideal solution. However, the lack of transparency of the industry introduces a very thick layer of uncertainty, therefore we must keep our eyes open and stay vigilant of the signs data shows us.