👋 Hey, it’s Sundar! Welcome to experiMENTAL: a weekly newsletter on B2C Marketing & data science how-to guides, frameworks, and stories from 15 years including early Uber.

This week is sponsored by Deel

Streamline IT management with 'The World at Work 2024: Deel IT

Discover how you can transform your IT operations, enhancing operational efficiency & compliance across borders. Our guide has essential strategies for managing a global workforce.

Download free guide

How to beat ad fraud

Ad fraud is a $100B+ problem that keeps growing every year. It’s estimated to represent 22% of all spend. Let’s let that sink in. It’s a massive drain of resources and disproportionately affects the Marketing poor that don’t have always have the tools to fight it.

There's also a reason ad fraud is huge on digital marketing. It's:

• Lucrative

• Easy to scale

• Hard to catch

The ultimate gold mine for digital thieves. But, like any other type of fraud, the first step is to understand what it is and then take the right steps to get rid of it. If you don’t, it just keeps growing and growing.

What is ad fraud?

Ad fraud simply is any attempt to deceive advertisers into paying for something that has no value. Here are a couple examples:

1. Bots generating clicks

2. Ads on fake websites

3. Ads shown to bots 

It’s across the entire funnel, but each type has it’s own nuance. 

Impression fraud

Impression fraud is when an advertiser takes credit for showing your ad without there being actual human eyeballs seeing them. This commonly happens in display and social channels.

Example 1: Ads are shown to non-humans (bots)

Example 2: Places where humans can't see them (like below the fold).

Example 3: The classic impression fraud example is "stacking." Imagine a website with 10 ads all stacked on top of each other. You can only see the top one, but all 10 register impressions. FRAUD!

Click fraud

Click fraud is when you pay for a click without there being an actual intention to see the next step. This commonly happens in CPC channels like Search

Example 1: Your competitor sits in a room clicking on your ads repeatedly to drain your budget. Damn, you must have really pissed off the competition.

Example 2: More commonly it’s when bots click on your ads automatically.

Although it’s not necessarily always fraud, but this happens a lot with web crawlers. I’ve seen examples of Google and Amazon generating hundreds of clicks at consistent times when their spiders crawl the web.

Now, I can already hear a voice at the back saying, "We don't pay per click; we pay per conversion." You ain’t special because it’s happening to you too.

Conversion fraud

Conversion fraud goes when step deeper and it’s when fraudsters simulate legitimate user actions that register as conversions.This commonly happens with affiliate / partnership models where you’re paying on a CPA (Cost per Action). This is obviously the most dangerous one.

Example 1: Fill out lead forms with fake information

Example 2: Install apps and immediately uninstall them

Example 3: Add items to a cart and even complete purchases with stolen credit cards

Bonus type: Domain spoofing

Domain spoofing is when the domain you’re advertising on is misrepresented. You want to be running an ad on NYTimes.com, but it’s actually on a sketchy ass website. This commonly happens with affiliate and programmatic channels.

It doesn’t always result in a cost (although I guess it’s a form of impression fraud) but it can damage your brand. Imagine your favorite brand advertising on a website that’s 100% counter to their and your values. It would make you reconsider buying from them.

Why should you care about ad fraud?

The most obvious reason is that you're wasting money. Ad fraud ruins your:

1. CAC (Customer Acquisition Cost) → Let’s assume a % of your conversions are fraudulent so your true CAC is significantly worse than you think:

2. Marketing attribution → If you’re using click based attribution, then obviously click fraud will skew and warp attribution. The interesting thing is even models like MMM / MTA don’t necessarily help here because they’re based on modeling that assumes impressions are true, clicks are true, and conversions are true.
   

3. Budget allocation → Again, depending on the sophistication and attribution you’re using, you might double down on channels that look like they're performing well.

How to detect ad fraud

Detecting ad fraud is half science, half art, half intuition. Yes, that math adds up. The biggest key is understanding your funnel but more importantly there are certain “laws” that are hard to beat. 

Let’s look at some standard benchmarks:

It’s REALLY hard to be outside these ranges. Those are reserved for the best of the best. That doesn’t mean if you’re above these benchmarks it’s ad fraud but it’s worth considering “Are my numbers really this good?”

The biggest thing to look out for are spikes. You’ll see things that don’t make sense in large clusters quickly. It’s almost like greedy bots could get away with it if they weren’t so greedy so quickly…

Here are some more specific things you should look out for: 

1. Unusual traffic patterns

Look for abnormal spikes in traffic that don't align with your marketing activities.

Example: If you see conversion rates spiking at 3 AM local time when your target audience is asleep, that's a red flag.

2. High bounce rates & low engagement

If traffic lands on your site but immediately leaves without engaging with your site, you might be looking at bot traffic. Humans typically don't click on an ad and immediately bounce.

Of course this assumes your website isn’t complete shit and that your ad and landing page are actually part of the same journey. That’s a huge assumption. But please don’t be that marketer that now uses ad fraud as the catch all like you’re using seasonality. 

Metrics to monitor: 

• Session duration

• Pages per session 

• Scroll depth 

3. IP analysis

Multiple conversions coming from the same IP address within a short timeframe is suspicious. Similarly, spikes in outdated browsers can signal bot activity.

4. Geo anomalies

If you're targeting users in the United States but suddenly see a surge of traffic from countries known for click farms like certain regions in Asia (cough cough China and India cough cough) then investigate!

How to eliminate ad fraud

There is only one method to eliminate ad fraud. Monitor + Incrementality testing. You need to set up a regular cadence of audits that check for a lot of the signs above. Schedule monthly or quarterly deep dives into your traffic quality then run incrementality tests on sections you think are worth testing on.

When you find fraud, act quickly:

• Implement fraud prevention tools

• Block suspicious IPs or traffic sources

• Adjust bidding on problematic placements

• Request refunds from publishers or ad networks

Work with reputable partners

Choose quality over quantity in your marketing partnerships:

• Vet publishers and networks before running ads

• Request transparency in programmatic buying

• Include fraud clauses in your media contracts

Use tools to help

Unsurprisingly a $100Bn a year problem attracts companies to try and solve them. There are tools out there that can help make it more effective. I’ve only ever worked with Lunio but there are a few other options:

1. Lunio

2. ClickCease

3. CHEQ

4. DoubleVerify and IAS (Integral Ad Science)

5. Built-in platform tools

Note: This list is AI generated so do your own research first

Major ad platforms are also trying to tackle this given the reputational risk but they’re not perfect yet. 

• Google Ads has invalid click protection

• Facebook has their own fraud detection systems

• LinkedIn filters some automated traffic

But fraud is always like a game of whack a mole.

Wrapping up

Ad fraud isn't going away. As digital ad spend continues to grow, so does the incentive for fraudsters to develop more sophisticated techniques.

The good news is that with the right approach, you can minimize its impact on your marketing performance.

Enjoyed reading this?

Share with your colleagues or on your LinkedIn . It helps the newsletter tremendously and is much appreciated!

Missed my last article?

Here it is: How to master Pre / Post analyses