Why E-Commerce Conversion Rate is Stuck at 3%

18 years ago (gulp) I helped bring one of the first Martech SaaS products to market. Without fail it increased conversion rate. Often doubling or tripling it. Today, if you use Adobe Target you still use the same “m-box” (“m” for “Marketing”) which was core to the A/B Testing, Targeting and Personalization that Offermatica pioneered. It’s gone on to define the category and helped Adobe maintain an amazing 15 year foothold as the enterprise standard for marketing software.

The Martech SaaS category has grown by an astounding 7,258% over the past decade. The company Segment was purchased for $3.2B in 2020. This is because the value in increasing conversion rates is massive. But where is that value today? It has been painful for me to watch conversion rate stasis. U.S. e-commerce conversion rate has been stuck at 3% for the past 10 years! Why? 

After all, we have had a myriad of advances in data collection and storage as well as billions of VC dollars invested in SaaS tools for testing, targeting and personalization. This includes the rise of customer data platforms (CDP) over the past 5 years from stand-alone vendor solutions to composable systems integrated into cloud data warehouses and data lakes so analytic and data teams can grok this customer data. 

The start of NRF next week is a good time to note that an increase of overall U.S. e-commerce conversion rate from 3% to 4% of customers grows U.S. e-commerce by $334B a year. A brand that doubles its conversion rate cuts their customer acquisition cost (CAC) in HALF creating exponential ROAS. Whether it is a purchase, a sign-up, or some other transaction there is no metric more important in the digital age than conversion rate. Conversion rate is what spins the flywheel.

Yet here we are stuck and looking ugly. E-commerce is growing but it’s only because of demand being driven by the convenience of Amazon and the targeting of Google and Meta (more on them later). It’s not growing through the efficiency of marketing dollars. The lack of conversion rate growth proves brands and retailers are not providing better digital customer experiences. 

Every website looks almost entirely the same. Product pages follow a similar template everywhere. E-commerce has become gentrified and commoditized. Andrew Chen of A16Z calls the copycatting of marketing leading to stasis and poor results the “Forever War.” This sameness certainly plays a big role in the conversion rate stasis.

The biggest problem today in our ever crowded channels of marketing clutter and sameness is lack of relevance. Relevance always cuts through the clutter. Relevance stands out. Relevant marketing always gets noticed by you. It is how your brain functions. Smart marketers have always known, it is the relevance of the product, the offer, the timing or the creative/messaging that lifts conversion rate. 

Lifting conversion rate is thus a matching problem. Understanding the needs and desires of people and showing them what is interesting to them is what gets someone’s attention and ultimately their purchase. Yet despite more data and more measurement and more tools than ever marketers are not moving the needle matching people with products. Why?

The question of why conversion rates have remained the same over such a long period began to nag me coming out of COVID and over the past year and a half I’ve thought and researched a lot about why matching sucks and why marketing is not relevant in owned and earned media. Why is conversion rate stuck at 3%?! Why are ads and messages irrelevant outside the walled garden media platforms? What makes the last decade plus of data, tools and practices different from the early days? Why despite the ubiquity of customer behavioral data, transaction data and tools for both have we not moved the needle?

The answers to all these marketing performance questions are of course related to data. But it’s more than just data. Despite all the data and all the tools it became very clear to me in my research that the practices of using data in conversion rate optimization have changed significantly. 

Let’s take a standard A/B test on a homepage as an example.

When we brought A/B testing to market with Offermatica we would run a test to everyone that hit the homepage. Behavioral data from every visitor would be collected in the results. While the aggregate results were interesting, more interesting and useful for optimization was slicing the results through lenses of different segments. Did morning visitors to our homepage react differently than evening visitors? Were results different from Search referrers than from Direct? Did first time visitors respond differently than return? What about people from Texas vs New York? On and on we would “segment” the results to learn about our customer and what was relevant to whom.

Understanding how different segments of customers behave compared to each other is the key to generating large scale lifts in conversion rate. Where there might be an overall lift of 5% in Version A vs Version B, as you explored segment level results you began to learn about your different customers. Some segments would achieve lifts higher than 5%. Other segments might have no lift at all. Some thin slices of segments might show tremendous lifts over 100%. Some segments might have performed better with the version that lost overall. 

This is where the practice of matching segments to relevance comes in. By optimizing your matching of version A and B based on segment level performance and then adjusting your marketing efforts based on segment level insights you can drive better performance. And the segments that showed little to no lift at all? They get a new iteration of the test based on our learnings. 

When we brought A/B testing to market the optimization was done ex-post facto. We found the segments that worked best for the KPIs we wanted to improve. That knowledge is invaluable for optimization. Take the whole, divide it up and then match what performed best to whom. Simple. Fast. Effective. It’s how you take a 5% lift and turn into a 25% lift. This is the practice of conversion rate optimization we used and it worked across every use case.

Yet somehow in the last decade conversion rate optimization has moved away from the practice of using as much data as possible across as many people as possible to discover and match. Marketers became armed with more and more Marketing SaaS tools. Each one a point/channel solution. Each with their own data and their own segments. Instead of discovery across customers, channels and “all the data,” these tools relied on Marketers cognitive judgment ahead of time to solve the matching problem. And since your customer IDs and your behavioral data are in their database how do you segment them?

You have to do a SQL query. 

My research found that the practice of conversion rate optimization has predominantly moved from marketers using all of the available data on customers to using very little of it. This immediately reminded me of the Vint Cerf quote “Google didn’t win search because it had better algorithms. It won because it had more data.” 

But it’s worse than just using limited data. The creation of segments – the actual SQL query – is reliant on someone’s ideas and judgment of what measured behavior or events are most important to create the segment. And I can tell you with high certainty and experience – the people doing these queries are smart but they are not Behavioral Psychology PHDs. 

Below is a typical UI for SQL based segmentation. I’m not looking to pick on this product. Every one of 1000 Martech products including CDPs and Data Warehouses has a UI that looks like this. 

Here segmentation works by a marketer continually reducing the data it has on its customers using cognitive judgment and bias at each step to make those determinations. 

This is why we see things like segments for email campaigns that are “added to cart” + “last 30 days” + “no purchase.’  As a segment, most marketers will send all these  “added to cart” + “last 30 days” + “no purchase” customers the same message. Or maybe do a random A/B test and wait weeks to see how this very thin slice of customers react. It’s no wonder why it is not moving the needle on conversion performance. There is no matching optimization.

This is the classic “when you’re a hammer everything is a nail.” SQL queries are hammers on data. They don’t make marketers smarter about customers and their preferences. Learning can’t be applied across channels. Channel specific segments bump into each other in different tests at different times from different teams. It makes learning and optimization across the user experience an impossible task. There is no iterative learning from SQL based segmentation. Every segment is different every time. This has nothing to do with optimization.

The newest thing is using “AI” to turn a natural language question into a SQL query using GPT. Cool right? But still SQL. SQL queries are also being used for machine learning models and scoring which again limits data being used and relies on cognitive judgment and bias.

The use of SQL for segmentation across disparate datasets and even in aggregated datasets explains to me why conversion rates are stuck. Brands are learning nothing from all that data. It has nothing to do with who their customers are and what is relevant to them!

Last but not least, back to the platforms. 

As I’ve said for years, marketers can learn so much by copying what they do. Financial results of Google and Meta are reliant on optimizing relevance. Showing you the right message at the right time is their entire business. It is a matching problem. The better they do it the more brands will bid to ensure their product is the one being shown to you. And as a cost-per-click (CPC) business they only make money when you click and even more money when you purchase. 

In fact the platforms share a bit about how they do their matching. And guess what? They don’t use SQL. They employ multiple ways to collect and aggregate that data and use algorithms over their entire “customer” database to find segments of people to match against based on the goals/KPIs of the marketer. It’s the way we used to do it on the owned side but on steroids called “deep learning.” Meta uses a “holistic model architecture” that is founded on deep learning. Google uses a Deep Neural Network architecture. If all this deep learning sounds like something out of the reach of most marketers that used to be true. However, advances in AI are changing that. 

My research into the struggles in conversion rate improvements and my thesis that limited data and SQL based segmentation were holding back conversion rates turned into a pilot project last Spring. Based on early success that turned into a company last Summer. And over the past 6 months through private pilots with great brands across multiple verticals it has turned into an AI deep learning product that radically optimizes the work of marketing, product, innovation, martech and data/analytics teams.

The applied AI we’ve built improves conversion rates and any other marketing KPI you use through advanced segmentation and surfacing matching intelligence. Just like the platforms but with your data. Whatever data you have the AI looks at all of it and makes it instantly more valuable. But my written words and my research don’t do it justice. All AI has to be seen to be believed. So let us know if you want to see Neuralift AI and let’s start improving conversion rates once again.