If you’ve ever been followed around the internet by a laughably inaccurate targeted ad, you’ve seen firsthand how difficult personalized marketing can be. You buy your friend a baby shower gift, and some database somewhere decides that you must have a baby yourself, and now this poor diaper brand is desperately trying to sell you diapers for a kid that doesn’t exist.
It’s a little funny and annoying to you, and a real waste of time and money for that brand. But it wasn’t supposed to be this way—the promise of data-based targeted advertising was that brands would be able to find and reach their best customers directly. People would be served ads and offers for things they’re actually interested in, and brands wouldn’t waste their budgets marketing to consumers who aren’t in their target audience. In theory, it should be a win-win.
But finding the right audience for an ad has never been easy. For years, marketers relied on third-party demographic and contextual data when doing their ad targeting. But as people began to rightfully demand more control over their personal information, the industry pivoted toward more privacy-first practices. While it’s a shift that was necessary and long overdue, it also made it harder for marketers to have confidence that they were getting their ads in front of the right people.
That’s why so many retailers are getting into the ad network business—first- and zero-party purchase data is the new gold standard when it comes to ad targeting. But even so, divining consumer behavior is still no easy feat; knowing what someone buys at one store in a vacuum isn’t terribly illuminating. When marketers are stuck working with partial, inaccurate information like that, a thoughtful gift purchase suddenly gets you aggressively pitched on pull-ups.
Maybe the problem is that this whole consumer data ecosystem leaves a pretty important participant out of the equation: consumers. By trying to figure out what people are buying and what brands they like from afar, marketers can never get the full picture. What if they just asked instead? And what if they made it worth people’s while to tell them?
Rewards app Fetch is betting that by bringing consumers into the marketing data loop, they can get that full picture of shopper behavior. By rewarding people for sharing their shopping history by uploading their receipts, they’re working to create a database of consumer behavior that cuts across all retailers, online and off. It’s a goal that many have had over the years, from credit card issuers to major retailers to market researchers, but Fetch is looking to do so on a scale no one else has been able to achieve.
Their pitch to consumers is straightforward: Upload your receipts and receive points towards gift cards in exchange. All of those paper receipts that end up stuffed into wallets or wadded up in the bottom of jeans pockets are a previously inaccessible look at real shopping behavior, an invaluable source of information.
Turning those slips of paper into parsable data isn’t easy, however. But this turns out to be a specific use case that AI is perfect for. While most of what we see these days in the overhyped world of AI are search results of questionable accuracy or creepy images straight from the uncanny valley, sorting through mountains of data from disparate sources and making sense of it is right in AI’s sweet spot.
Fetch has developed a machine learning system that can read and parse receipts from virtually any store. By using a proprietary transformer model built with human-in-the-loop training, they’re able to turn raw images of any and all receipts into a comprehensive, accurate database of consumer habits.
By their nature, receipts are non-standardized and messy. Each retailer has its own format, abbreviations, and product codes, so a receipt can be difficult to interpret without knowing exactly how a store lists a product in their inventory system. To overcome this hurdle, Fetch’s team spent two years manually cataloging products and receipts from stores across the country.
Now Fetch’s AI can look at one retailer’s “GRN BN 16OZ” and identify it as a pound of green beans, or another’s “MLK 2% OG 128OZ” as a gallon of a specific brand’s organic milk. Even handwritten receipts from restaurants and small businesses are able to be accurately parsed.
It’s a system that keeps getting more accurate over time, as every new piece of data helps to further hone it. Fetch’s AI processed more than 3 billion physical receipts and 360 million e-receipts just last year, with users submitting upwards of 85 percent of their purchases of all kinds, from groceries to restaurants to retail. And the more receipts the system ingests, the better it becomes.
With that comprehensive, consistently up-to-date data in hand, marketers can then personalize specific offers within the Fetch app like point boosts to consumers they know for a fact are buying their products. This helps users earn points towards gift cards, creating a virtuous cycle of brands connecting with their best customers and those consumers being rewarded for buying the products they already love.
Of course, talk about personal data collection can give people pause. How can they feel secure that the data they’re sharing is being stored and shared responsibly? Explicitly built to ensure privacy, Fetch uses data clean rooms: protected, cloud-based environments that anonymize first-party and zero-party data. It’s a system that ensures that data is used only by the companies that consumers provide it to, and in a way that’s not connected to any personally identifiable information.
When people opt in to sharing their shopping data in a secure way like this, it brings the physical world into the digital. It takes the ideas behind tools like browser cookies that have delivered imperfect results in the past and finally creates a complete picture of consumer behavior. And it does so out in the open rather than in the shadows.
Fetch is making the same argument to both brands and consumers: rewards are a better way to advertise. Rewards are opt-in, always-on advertising that puts consumers in control, guarantees relevance and makes brands the center of joy.
With this approach, the hope is that rewards bring both marketers and consumers a bit closer to the initial promise of targeted advertising: getting offers in front of the people who actually want to see them. And then maybe you’ll finally stop getting those diaper ads served to you until you actually need them.
