We faced a similar issue with GDPR at Uber. We will definitely need to be careful, but many merchants already have customer opt ins (e.g. the cookie consent pop up) for data processing that we ideally should be able to piggy-back off of.
Yes, that's mostly right. We also vary the discount value, so it's less a binary discount/no discount and more a range. There is often a cutoff though. Merchants can input a hard cutoff e.g. if they want to ensure everyone gets a discount (great if they also have marketing assets for a sale), or if they want to avoid making their sites feel too 'sales-y'. Otherwise the cutoff is defined by conversion prediction, inventory levels, and a few other inputs.
There's actually a lot more we could do to make this cutoff more intelligent though - e.g. at Uber the cutoff was set to exhaust a certain promotional budget. Or we could target a specific ROI if we eventually have good enough predictions.
Thanks for the reply. Do you use Bayesian models for this? Btw, Pete Fader[1] has done so much work in customer valuation where estimating the probability of purchase is a crucial aspect. Maybe you already use them.
Yes. We're going through Shopify, so merchants agree to terms when they install the app.
There's user activity data, but also contextual data and shop data that we use. 'Contextual' data refers to things like device type, traffic source, time of day, day of week (there have been some interesting trends with corporate vs. non-corporate customers in this one).
Shop data includes things like product profit margin and product conversion rate. Obviously we can go deeper with our discounts on products that are very profitable, and it's typically more efficient to give a discount on products with lower conversion. Merchants also like boosting items that haven't been selling well.
We use historical purchase data, as well as view history, traffic source, device type, etc.
Traffic source a lot of times is the most impactful. People coming from ads are often more in a browsing mindset, vs. people typing in the url directly have a higher purchase intent.
We don't have abandoned cart rate as a feature in our model, but actually might be something worth looking into adding.
We probably haven't done a great job communicating this to smaller merchants yet.
Re personalization - We'd plan to make it an optional field for merchants to use, so if they are worried they don't necessarily need to enable personalization. But I personally think a fair bit can be explained with on-site copy like "flash sale".
The 1% was meant to already bake in some flexibility for smaller merchants haha. But if we get that feedback we might consider more of a tiered model.
1) What we're really doing with this first product is predicting how price impacts conversion rate. That's been a relatively simple thing to measure in my experiences. It's more difficult to do things like predict a customer's probability of buying based on their order history.
2) Yes so we don't have a case study comparing us to manually setting discounts, but the task gets pretty time consuming quickly if you want to update the discount daily (or more frequently) and personalize the discount (which is one of the features we're planning on adding).
In my experience there's also a ton of B2B demand for AI right now. I've been working with AI in this space for the past 4 years and the desire from merchants to use AI tools has really ramped up. Supply follows demand to a certain extent.
Right now the updates occur daily, we're planning on building a bit more intelligence into the update cadence over time (e.g. once we see we have a stat sig read on how the price change impacted conversion).
Retailers don't approve the discount changes, but they do provide guardrails like maximum discount value to avoid us carving into their margins too much. They can also log in and review / update discounts at any time in our app.
1) Yes, personalization might have a bit of an experience tradeoff. We can try to mitigate this with messaging like "flash discount" or "just for you". But we also want to make it optional for merchants. In my experience there's still a lot of improvements from other things too like dynamically adjusting discounts and varying the discounts across products
2) One of the takeaways from my time at Uber was that certain predictors of discount efficiency held pretty constant across markets. A couple were conversion rate (if more ppl were going to convert without the discount, it's less efficient to give the discount) and profit margin. We're betting that we can train a model to generalize these trends across stores to create a bump in performance.
3) We'll be kicking off our first case study with a customer in a couple weeks. At Uber, just varying the discount across merchants on Uber Eats improved the profitability of the discount by 40% (mostly because we were able to take advantage of differences in commission rates across merchants).
