Digital advertising is designed around ecommerce; all ad platforms are built ‘ecom first.’ But few are going beyond the standard ROAS target and unlocking the full potential by using first-party data to train algorithms to ‘know’ what success looks like.
Let’s start by looking at why this is important.
Algorithms are Dumb
Today’s media buying is powered by AI-powered algorithms that decide everything; bids, audiences, even creatives. However, as “smart” as media buying appears, it is still rather simple and you still need to train the algorithm.
There are two ways to influence how an algorithm makes decisions:
- Limitations (target, targeting and budgets)
- Reward (conversion pixel)
Limitations can help you avoid bad behavior, like spending a month’s budget in two days or ensuring German ads don’t end up in Spain. But they don’t help with teaching good behaviour, such as targeting high-value customers. For this you need to reward the algorithm by changing the value it receives via the conversion signal every time it does something you want to encourage.
But why is this game-changing for your business?
Optimising Toward True Business Value
When looking down the P/L of any ecommerce store, it is obvious that revenue is not the goal itself. There are plenty of direct and hidden factors that influence profitability, and for media buying they can drastically change the ROI of your marketing investment. Let’s look at a simple example:
When accounting for just minor changes in cost, the ROI of the media investment can be vastly different. Most marketing teams are aware of this — few are able to act on it.
Product margins are a good straightforward example, but in our industry they are old news. There are much more exciting challenges to take on besides product margins:
- Returns that are your eating margins
- Advertising too much to existing customers
- Algorithms that do not prioritise low sellers
Challenge #1: Returns are eating your margins
Increasing consumer expectations put pressure on sellers to give more and more favourable return policies, which hurts your business’ profitability. It is tempting to calculate the average return rate for each item and adjust conversion values accordingly, but that can be dangerous.
To predict return rates, you need enough data to see a meaningful trend. Most ecommerce stores constantly change the product assortment, making product-level analysis of return rates unpredictable. Furthermore, if your customers are allowed to return products after 30 days, you need to allow that as a minimum before defining a return rate for a product. Instead, try defining the average return rate per market, product category and month (especially during Christmas), which will be a much more applicable adjustment to your conversion value.
Challenge #2: You advertise too much to existing customers
The hottest topic nowadays is new customer acquisition strategies and growing the total number of new customers rather than sales from any customer.
You could leverage the plethora of new customer-centric features being launched by ad platforms, or you can use more sophisticated methods such as taking more control of the conversion signal via custom pixel. The key to success is knowing WHY you do it.
When customers buy a product, their previous experience with you, not advertising, is the biggest factor in deciding whether to go with you or a competitor. This means that the incremental effect of every ad dollar spent is always less for existing customers than new customers. The algorithm doesn’t know that, so you need to either 1, bid lower for existing customers by having your campaigns split, or 2, lower the conversion signal you sent to the algorithm.
Option 2 is usually preferable since despite the mountains of data hoarded by the ad platforms, they still don’t have perfect visibility nor control of who YOUR customers are. Only you know that and can control the signals appropriately.
Challenge #3: Algorithms do not prioritise low sellers
It’s hard to teach an old dog new tricks — and that’s a common curse of algorithmic media buying. Once it’s found a product that drives sales, it will push it endlessly — even as your inventory is running low.
But whereas understock will lead to you spending ad dollars pushing items that would naturally sell out, overstock can really hurt your bottom line when you turn to price cuts to clear inventory.
The answer to reversing this frustrating trend is early detection of over- or understock. My go-to measure is ‘Days left of stock’, calculated as “number of items in stock” divided by “average sales per day.” If ‘Days left of stock’ is lower than your restocking lead time or days until next season’s drop, you are risking understock and vice versa.
You should curate what products you expose to potential buyers using this data. This is straightforward for ecommerce sites as digital ad solutions are built for you. Product feed ads, like Google’s Performance Max and Meta’s Dynamic Products Ads, allow easy product segmentation. Test moving high stock items into a separate campaign with more aggressive bidding and removing ad exposure for low stock items. That will save you and your purchasing department a lot of future headaches.
Data is Ecommerce’s Best Friend
These strategies are key because, in a world powered by AI, having a data advantage is how you get ahead of the competition. Ecommerce marketers must learn how to leverage first-party data effectively or risk being outmanoeuvred by those that do. To get started always remember:
- Something is better than nothing: Being overly worried about accuracy will stop you before you get started and getting started is the most important.
- Prioritise factors with wide distribution: You may have a problem with high return rates, but if all product categories have similar rates, reducing conversion value by a fixed percentage won’t change what products the algorithm picks. Therefore, prioritise including factors that have a high impact on your profits.
- Quick is better than perfect: Speed is more important than accuracy since bidding algorithms have a recency bias, e.g. Google only uses conversion data from three conversion cycles (usually three to nine days). It’s better to quickly send a signal rather than waiting to collect more information and update later.
Andreas Arentoft is a data and technology nerd currently working as the Head of Go-to-Market, Technology at Precis Digital. With a diverse background in designing advanced media buying strategies and custom data solutions, Arentoft focuses on bridging technology and business problems to find action-oriented solutions. Right now, he is obsessed with marketing evaluation and helping businesses make better decisions about their media buying. Through his work with Precis Digital, he has helped to create an award-winning, data-driven digital marketing agency with approximately 600 employees and 16 offices in Europe and North America.