Cashback Attribution Model
Struggling with click fraud? How precise attribution mitigates risk and increases satisfaction.
Overview:
Attribution is the backbone of any affiliate marketing system. Given its importance, the system must be accurate and fair to work. If either of these two elements falter, a cascade of negative consequences begins to rear its ugly head. If left to fester, it often manifests itself as wasted resources, lost revenue, and negatively impacted trust.
Our client was facing some obstacles when it came to awarding cashback for purchases. It wasn’t an issue with cashback being awarded. Rather, it was an accuracy issue not in their favor due to a loophole found by some unscrupulous individuals.
The project was a success and delivered the following benefits:
- Implemented a fair and accurate attribution system.
- Fraud risk mitigation.
- Reduced operating costs, increasing profit.
Read on to see how ThunderSteed mitigated fraud, increased revenue by implementing a fair and accurate attribution model.
Issues:
Their system was being circumvented by these bad actors. After digging in, we found the issue growing rampant and had been occurring longer than initially expected. Not only was this activity jeopardizing relationships with affiliate networks, it increased expenses. The amount of time support was spending to find, research, and take corrective action was rising fast.
The vast majority of cashback sites have a time limit from click-to-purchase. However, our client had the most gracious policy in the business - there was no limit. Unfortunately, some accounts found a loophole and started taking advantage of this generosity.
- Accounts violating TOS.
- Cashback awarded without visting site.
- Rising support costs.
- Relationship strain on strategic partnerships.
During a routine fraud investigation, someone stumbled on a startling pattern. Accounts with hundreds of purchases over the course of a few weeks-to-months. The problem? These accounts had not clicked on the merchant’s storefront from the site's affiliate link before making the purchase. The site requires an active login to use merchant links; otherwise the user is redirected to a login page before continuing. After further analysis, it was determined the users had found a way to bypass that system. In doing so, they were in clear violation of the site’s TOS and how-to guides for using the site.
Further analysis revealed that it had been going on longer than initially expected. The process involved identifying accounts participating in illegal activities. Next, those accounts received a temporary ban, pending an investigation. This impacted support in communicating with the effected accounts to get complete their research.
However, despite their best efforts, it was becoming a real time sink. As fast as support could find and suspending these accounts, new accounts were springing up; and doing the same activities. This resulted in support’s time being misdirected from servicing legitimate inquiries. Manually policing this issue was not an effective solution, much less feasible. The "fix" was a band-aid at best and limited pursuing only the most egregious offenders; there simply were not enough staff or hours in a day to fully solve. This activity was robbing the both the site and its members in multiple ways.
First, affiliate networks take fraud prevention very serious. Accounts circumventing safeguards can impact trust with the merchant and the affiliate network. The penalties for having fraudulent users range from the merchant issuing a suspension to a permanent ban. It is even more devastating if an affiliate network issues a penalty because it impacts all merchants on that platform.
Secondly, fraudulent activity impacts all members using the service. The same for those abiding by the rules. By not visiting the site first before making a purchase on an affiliate site, the site is robbed of traffic. There is a direct correlation between the amount of traffic to the size of offer. Brands reserve better terms on promotions for sites with more traffic. It's just the nature of the game.
Third, the business saw a sharp increase in operating expenses needed by support to tackle the issue. And worst of all? There was no end in site!
Solutions:
After debriefing on the situation and sample data reviewed, the root cause became clear. No eligibility window + insufficient attribution measures = easy exploitation by bots and fraudsters. It was clear that to address this issue, there had to be a change in attribution. This problem required an accurate attribution model to curb abuse. To put a stop to this type of fraud, better business rules were required to determine eligibility.
Summary of improvements ThunderSteed implemented with this project follows:
- Attribution model developed
- Added preview/preheader text
- Optimized sending time and campaign schedule
- Created custom sending and tracking domain
- Brought best practices and CANN-Spam guidelines up to date (By adding alt-text, address to the footer, adding body text)
- Performed a re-engagement and list clean-up campaign to segment by activity
While click history was being recorded in the database, that data point was not utilized. Along with the transaction timestamp reported by the merchant, the click time was used to calculate elapsed time. Based on this duration, a decision could be made as on cashback eligibility.
One specific challenge was that of time zones, which were not standardized. We reviewed samples from the affiliate reporting feeds. By comparing the time difference, we discovered what time zone the affiliate used in their reporting. These time zone were captured and stored in the database for use by the attribution model.
The model must have a one-to-one relationship between order and click. A single click could be attributed to only one order. For example, if a visitor clicked on a merchant's link once from the site, only the first order would be eligible for cashback. Any additional purchases without registering a click from the site were not eligible for cashback
Being a new process, the threshold limit needed to be accessible to decision makers. This was necessary to adjust the value while they worked to find the "sweet spot." This needed to be accessible to authorized personnel without involving IT. An enhancement made to their dashboard for roles allowed to make adjustments. Auditing requirements were satisfied while recording, details on the user making changes.
Results:
- Fraud risk mitigation
- Decreased operating costs and increased profit
- Accurate and fair attribution
- Analytics increased visibility
The model made an immediate impact. First, it had the desired effect on deterring fraud. By removing the reward, the number of occurrences soon dried up. In addition, the site was able to recover lost revenue and reallocate support. Both lead to improved customer satisfaction. Capturing elapsed time between click-to-purchase allowed the support group to be more proactive. Accounts with a transaction exceeding the threshold limit were prime for a how-to use reminder.
