[HN Gopher] Launch HN: Promi (YC S24) - AI-powered ecommerce dis...
___________________________________________________________________
Launch HN: Promi (YC S24) - AI-powered ecommerce discounts
We're Peter and Jiaxin, and we're building Promi
(https://www.usepromi.com). Promi uses AI to optimize retail
ecommerce discounts across products and customers (think new
customer discounts, clearance sales, holiday sales, etc.). Here's
a quick video overview: https://www.youtube.com/watch?v=SHTw9VH8bCw
Discounts have traditionally been a bit of the 'wild west' of
pricing. Optimization techniques at even the largest merchants are
heavily manual. Product level discounting decisions are distributed
among operations or category managers, and store-wide discounts are
often set by marketers. Typically they rely on order history, look
at competitor discounts, or find prior discount performance to
inform their decision. But there's not a lot of science behind
choosing the discount, and teams don't have the time or means to
optimize it at a granular level - i.e. by product or customer
groups. We believe AI can better solve this problem by setting the
most appropriate discount value, varying that discount across
products, and personalizing that discount across users in order to
achieve various goals. AI models can leverage more data (e.g. item
conversion rate, profit margin, customer referral URL, device type)
and update more frequently than is realistically possible to do
manually. Our approach will also allow us to generate discounts
for relatively small merchants. We use models (layering traditional
NLP models and custom LLMs) to build a large-scale knowledge graph
to gather similar products across merchants in order to build
profiles around different clusters of products. Those profiles help
us build solutions catering towards subscale shops which
traditionally do not have an optimal pricing strategy. Our first
AI product focuses on liquidating inventory, and uses a store's
historic transaction and sales data to jump start training our
model and generating discounts. Our model predicts the discount
required to increase conversion rate by the proper amount to
liquidate the inventory by the desired timeframe. We then monitor
the conversion rate and (if we have statistical significance) make
frequent adjustments as needed. We've got a full roadmap of new
model approaches, including personalization and new objective
functions (e.g. profit instead of liquidation) to fit more discount
use cases. How we got here: Jiaxin and I are coming from Uber,
where I led product for the discount team across Eats and Rides. We
launched several analogous AI features at Uber and saw just how
impactful they can be for structuring discounts. For example, we
had issues with deploying our ML models for automated discounts in
smaller markets because of the quantity of data required to train
those models. We pivoted to a 'global model' that used data across
countries to significantly reduce the amount of data required in
any one country. That model performed even better than country-
specific models, showing us that there were very reproduce-able
trends in improving discount performance. If you run or know
someone who runs a Shopify store, you can download and play around
with our app here: https://apps.shopify.com/promi-discounts We'd
love feedback, thoughts on other use cases for discounts + AI,
questions, etc. Looking forward to hearing from the community!
Author : pmoot
Score : 21 points
Date : 2024-08-14 15:28 UTC (7 hours ago)
| evgwugegwhwgeg wrote:
| [flagged]
| dang wrote:
| YC hasn't changed. This kind of thing happens with every major
| tech wave, and this is the majorest one in a long time.
|
| Right now the waters are all muddy with hype. The sediment will
| eventually settle and we'll find out which lasting companies
| have emerged - same as dotcom eventually turned into Amazon
| etc.
|
| (I/we didn't flag your post btw)
| meiraleal wrote:
| > same as dotcom eventually turned into Amazon etc.
|
| With a big bubble burst in between.
| dang wrote:
| Yes.
| pmoot wrote:
| 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.
| rd wrote:
| An actually cool use case of AI. Congrats on the launch.
|
| Some questions:
|
| 1) I'm assuming by "personalizing discount across users", you
| mean personalized one-time coupon codes? I wonder if the UX of
| seeing one price in regular Chrome and one in incognito would be
| upsetting. I also don't know how price discrimination works but
| seems relevant?
|
| 2) I'd love to understand more about how for smaller retailers
| there'll be enough data to make meaningful discount programs for
| a limited set of consumers? Will data from similar/multiple
| retailers be bucketed?
|
| 3) Any numbers/data on effectiveness so far?
| pmoot wrote:
| Thank you!
|
| 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).
| rd wrote:
| Very cool. Saw you were H '10 - go crimson! Best of luck!
| bx376 wrote:
| This is a perfect example of automated price management.
|
| I think you need to better communicate the causality between
| applying your model and increased profits to justify your 1%
| commission.
| pmoot wrote:
| Good feedback. Ideally we have a few case studies actually
| showing the impact we generate, we just have to complete a few
| of those first.
|
| We've been open to working with customers on a trial basis
| also. This price is mostly based on industry comps for other
| optimization tools.
| algo_trader wrote:
| What is the feedback time delay on these pricings?
|
| Do retails get/pass on to you hourly/second-ly aggregate data
| or what?
| pmoot wrote:
| 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.
| r2sk5t wrote:
| Not necessarily relevant for you, but expect legislation for
| algorithmic price collusion:
|
| https://www.theatlantic.com/ideas/archive/2024/08/ai-price-a...
|
| https://www.propublica.org/article/yieldstar-rent-increase-r...
| bryant wrote:
| Looks very relevant. If the same service (e.g same instance of
| the model, or even different instances of the same model) is
| provided to more than one client, I'd guess a prosecutor might
| reach for it.
|
| Easiest but most costly way this could be avoided is by
| creating new models for each client using the client's own
| specific data and keeping the data and models fully isolated
| for each client.
|
| If derivative insights are gathered across all models, it'd
| have to be one-way informing e.g business decisions for the
| overall company rather than informing how the models themselves
| operate.
|
| ---
|
| edit: "We pivoted to a 'global model' that used data across
| countries to significantly reduce the amount of data required
| in any one country."
|
| This might paint a bullseye on their back, but I'm a security
| and risk person, not a lawyer.
| nilirl wrote:
| I have a few dumb questions for people who really understand how
| this stuff works:
|
| - Is prediction based on historic transaction and sales data
| effective? I always assumed transaction and sales data in
| isolation didn't contain enough information to be effective
| predictors of buying behavior. Is that wrong?
|
| - How much more effective is it than a human intuitively setting
| a discount? I can see large retailers saving time on having to
| set discounts for a large number of products. Just wondering if
| small merchants would just be better off doing it themselves.
| altdataseller wrote:
| Re: small merchants. Thats what I dont get. The companies that
| need this are the big ones not the small merchants with
| relatively few products they wish to discount. They're better
| off investing time and effort to making more sales not
| nickeling and diming with discounts
| RussianCow wrote:
| > They're better off investing time and effort to making more
| sales not nickeling and diming with discounts
|
| But discounts are one way to get more sales! There are plenty
| of mom-and-pop merchants that have to compete with large
| retailers and would benefit from a sophisticated discount
| system to drive higher volumes, but they just don't have the
| resources or the experience to implement it themselves.
|
| Source: I know the owners of a brick and mortar garden store
| that struggles with this.
| pmoot wrote:
| 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).
| pupumeme wrote:
| Your global model approach from Uber is clever. Have you
| considered how to communicate this benefit to smaller merchants
| who might be skeptical about having enough data?
|
| I'm curious about the personalization aspect. How do you plan to
| balance the potential uplift with the risk of customer backlash
| if they feel manipulated?
|
| The 1% commission seems steep for smaller merchants. Have you
| considered a tiered pricing model based on revenue or order
| volume?
___________________________________________________________________
(page generated 2024-08-14 23:01 UTC)