Grab some popcorn for this: we’re going down the rabbit hole of ad fraud and Indian case studies. Depending on various sources, anywhere between 15%- 25% of ad spend gets wasted due to ad fraud.
Mobile ad fraud, bots, false attribution and bad actors in the ad value chain contribute to the majority of the fraud. Besides overpaying for the ad which leads to an increase in the CAC, there are far-reaching consequences in marketing strategy and tactics. Ad fraud messes up the funnel metrics leading to decisions being taken which are detrimental to the company’s growth.
All major verticals gave been a victim to ad fraud, with ecommerce majorly bearing the brunt of the attack.
Source: Tech Arc
According to MMA, a majority of marketers surveyed believe that ad fraud is only going to increase in the future. The survey showed that among the major types of ad fraud in India, cookie stuffing leads with 74%, along with adware traffic (65%), data fraud (61%), ad injection (54%) among other major frauds.
Marketers have been active in trying to address ad fraud. Even with real-time tracking and analysis, many marketers are aware that fraud is not completely eliminated. Here are some of the traps to avoid:
1. Thumb Rule Fallacy
In the absence of a fraud tech tool, marketers are dependent on fraudulent traffic thumb rules that agencies share.
A typical ad agency thumb rule – ‘Mark traffic as fraud if Click to Install Time Interval is greater than 1 day’
Click To Install Time (CTIT) is one tell a tale sign that campaigns and performance data is being compromised by ad fraud. It works on the basis that for each app install, there is an expected window of time in which a majority of clicks and installs occur. High volumes of traffic outside this norm point to fraud.
While, a high proportion of conversions with excessively short CTITs is an indication that fraudsters are attempting to claim attribution of installs through click injection, having a thumb rule for CTIT both for a short time interval and for the higher side has adverse effects.
Install time also depends on the app size. The user may have been in a bad network area or may have switched off his phone before a flight. Tagging it as fraud using a thumb rule may lead to false positives and the agency may end up shutting down a traffic source from which good users are entering the system.
It is better to have holistic rules rather than thumb rules. What is prevalent for one industry vertical may not apply to another vertical. Games that are over 100 Mb have a very high CTIT, applying the same CTIT rules to other industry verticals will impact campaigns.
Another case is a thumb rule being applied for post-install events. If the events are less for a user then the user is usually put into the fraud bucket. This does not mean that all users with lesser events are fraudulent users. Just like every user is not a power user, some users take time to use different features of the app.
A blanket thumb rule in such a case may impact the funnel campaigns and eventually lead to churn.
Lastly, thumb rules often work for agencies to hide bad traffic along with good traffic. Instead, it is better to have a holistic view of the events being fired. Use patterns or Machine Learning to identify installs to be put into the fraud bucket.
2. Polluted Data Observation
Fraudulent traffic pollutes post-install data if not tagged properly. An Indian mobile app wanted to grow fast and had spread the campaign across various ad networks. With increased competition, ad networks went as low as delivering an install at Rs. 2 (~3 cents).
The marketer was pretty happy with the pace of acquisition, however, the product team realised that the users are not using the hero feature of the app. This led to rounds of brainstorming to change the product flow and eventually spending valuable tech effort in reworking the product.
At Rs 2 per install no publisher can make any money, how can he, therefore, deliver such a huge volume of installs which are not fraudulent. If something is too good to be true, it probably is.
The funnel data that got polluted due to fraudulent installs led the company into a different direction.
This can be avoided by a good fraud detection tool in place. Work with a company which tags the fraud install as early as possible so that the events are not tracked as a part of the funnel analysis.
Also, ensure that channel tagging is done for all users so that during the funnel analysis deep dive it can be seen if users from any particular channel do not exhibit similar behaviour to the rest of the user base. Identify the source and cause of the fraud, plug it, tag users as fraudulent so that they do not appear in any further funnel analysis.
3. False Authentication Vortex
Often advertisers are confident about their authentication mechanism during signup and if payments are done to the agency or publisher on the basis of a post-install event, marketers tend to not rely on fraud detection tools. Here’s an interesting case study of a fintech startup.
Since Aadhar and PAN were being used to authenticate the user during the signup, the company was not too concerned about fraudulent installs thinking it would not impact their funnel. The company was very happy with the paid user acquisition setup; however, the marketing team was unaware of the fact that organic stealing was at play here.
One of the publishers that the ad agency worked it was hijacking organic traffic and tagging it as theirs instead. The company was in fact paying for the users that were coming aboard due to the brand awareness and SEO/ ASO efforts. In this case, the company was a victim of click hijacking fraud.
Click hijacking is one of the most common types of attribution fraud. Click hijacking is when a fake click is sent to the attribution platform directly after the installation has begun. This tricks attribution tool into attributing that install, to the fraudulent click, as it was the last click received (typically most advertisers follow last-click attribution). Get hold of a fraud detection tool that gets rid of click hijacking. It is quite easy to detect with the right tool.
4. Lengthy Attribution Catalyst
Quick question before we start this one.
What is the default attribution window for Google Analytics?
Google Analytics uses a last, non-direct click attribution model throughout the bulk of the interface, assigning 100% of the credit to the last known source/medium, looking back over the last 6 months
How long a direct session is attributed to the first original indirect session, depends on the setting of the “Campaign Time Out” in Google Analytics. By default, this is set to 6 months. To see the settings, navigate to the Google Analytics Admin area, select your property and navigate to “Session Settings” in the “Tracking Info” section.
Specifically, a leading fashion startup did not know that for the past 2 years it was running campaigns with the default setting. While it was not subject to any malicious fraud, it ended up paying for direct traffic to multiple ad agencies who first brought in the users 6 months ago.
Interesting fact, most eCommerce sites are running on GA for attribution. Many are still on 6 months attribution. This also opens up these companies to click stuffing fraud.
Make use of Multi-Channel Funnel reports and model comparison to sort out things. More importantly, check the default setting in GA right away.
5. Re-Engagement Inactivity Time Conundrum
Besides Google and Facebook, ad networks are also pushed into service to get these users back. All good till here.
Another Indian fashion startup found out that the ad network was click spamming to get the transaction attribution of even active users to the reactivation campaign.
This is pretty simple to execute if the advertiser does not have the right attribution partner. In case the attribution partner does not support reactivation settings (like the way Adjust does) a fraud detection tool would be needed.
In the case of retargeting campaigns, payment is done on CPC basis. Organic stealing is prevalent here and many advertisers are unaware.
The measurement for reactivation from a growth team’s perspective and the ad buying team’s perspective is different. An ideal case would be to have a single data pipe but that’s mostly not the case in many startups.
6. Brand Bidding Acceleration
Bidding on their own brand name allows advertisers to show more listings on the brand search result page. That means much more real estate for the brand to showcase the value proposition. It also keeps the competitors’ ads from being shown above their brand’s listings.
Bidding on brand keywords is not as costly as most other high-performance keywords. Also, CAC for such keywords is lower since the conversion rate is high.
Some affiliates end up bidding on brand keywords even when they should not as per the advertiser’s guidelines.
This, in turn, increases the bid costs for the advertiser since multiple parties are bidding on the same keyword. Additionally, if the affiliate wins the bid, the advertisers end up paying much more for the conversion.
An eCommerce company’s bid rate spiked by 40% during their flagship sale day due to affiliate bidding on the same keywords.
Till a few years ago even I was guilty as an advertiser, by trying to address this by hiring interns to look for affiliates bidding on the keywords. Manual checking (which is the go-to solution for most of the companies ) falls flat when the ad is shown across regions. Affiliates will avoid bidding on brand keywords in the HQ region where most of the marketing effort is concentrated.
A brand bidding protection system is required in this case. Partner with a company which has integrated with Google to provide such data.
Ad fraud just does not increase the direct cost but that many implications on company strategy and marketing tactics which ends up costing much more in the longer term.
Having seen mobile fraud over the years, there are many more case studies that can be discussed.
Attribution platforms (Appsflyer, Tune, Kochava etc) also have a fraud detection suite which can be used to weed out some of the frauds. Specialised fraud detection tools (Mfilterit, Performcb, Interceptd, etc) are an integral part of the marketing tech stack these days.