Hacker News new | past | comments | ask | show | jobs | submit login
Launch HN: Aura Vision (YC W19) – Google Analytics for Physical Stores
88 points by jblok 9 months ago | hide | past | web | favorite | 41 comments
We're Daniel, Jaime, and Jonathon - the founders of Aura Vision. (https://auravision.ai)

Aura Vision is like Google Analytics for physical retail stores. Our mission is to ensure that physical retailers can innovate and improve their stores with data, in the same way their eCommerce counterparts do, while protecting customer privacy.

Retail teams often know very little about what shoppers do in-store leading up to a purchase. To try to increase sales, they change layouts, products, and media in their shops based on anecdotal knowledge, and experience. That’s because it’s hard to get good quality data about what consumers actually do in their stores at the moment. Many retailers periodically place people in doorways with clipboards recording shopper demographics and behaviours, which of course is costly and not very scalable.

We use existing security cameras in stores to detect the demographics (age, gender, staff/customer) and behaviour of all visitors using our proprietary computer vision technology. This creates an anonymised feed of aggregated data for the retailer, giving them new tools to improve their stores. E.g.

- To increase footfall, retailers can A/B test window displays, selecting the one with the highest peel off rate (the ratio of entries to people walking by)

- To uncover why a product is underselling, retailers can learn about the movement and dwell times of different demographics around products.

- To increase sales, they can select products that are suited to the demographics in that store.

- To increase conversion rates, retailers can identify where customers spend most of their time in-store and locate staff accordingly.

We started out in the UK during the birth of GDPR, so we’re acutely aware of the need to protect customer privacy. Video is deleted as part of the processing, and never stored thereafter, and our system never identifies people, nor stores identities. All data is aggregated into 15 minute chunks, which fully anonymises the counts, so you are left with information on the behaviours that the camera observed in that period. Those chunks are supplied back to the retailer through our dashboard and API as heatmaps and counts. We don’t rely on facial recognition, instead taking in visual cues from all features across the body.

In contrast many other retail tracking solutions, like Bluetooth and WiFi, aren’t GDPR compliant as they store MAC addresses, or other phone IDs without consent, which count as personal data. This means they can re-identify you when you come back to the store, or another store on their network. While regulation will do a good job at getting rid of these tracking solutions, we want to help by showing retailers there’s an option that gives them more useful data anyway.

Daniel and Jaime studied under the same supervisor at the University of Southampton during their computer vision PhDs. They saw plenty of opportunities for using deep learning in people tracking. A key part of Daniel's PhD was estimating people's demographics from CCTV footage and this led to the end result we are running now. Myself and Daniel went to primary school together, and my background is in APIs and frontends.

Thanks for reading! We know the HN community has many people interested and knowledgeable in computer vision and deep learning, so we're looking forward to hearing your thoughts. If you or someone you know has experienced similar challenges in retail, please reach out! jonathon@auravision.ai

While at a major beverage brand conglomerate, I used to be interested in this kind of data.

We had various projects that helped us gain insights on consumers from hiring agents, observing camera footage, tracking SKU sales & out-of-stocks, and use of beacons.

Every major FMCG company I've worked with already uses some tech vendor for segmented footfall model to drive placements of branded displays and promotional space.

We did run into real challenges conducting A/B tests in the retail environment though. Unlike with web apps where A/B tests are easily deployed and measured with analytics, in the retail environment, its hard to effectively measure success of A/B tests and attribute shopper motivation. Say if an A/B test is conducted on whether a retail display will drive sales of SKU's we'd need to be able to attribute sales against those SKU's. The challenge is that a FMCG company will try to place new products everywhere around the store and only on paid-for displays so it'll be hard to assess the success. Can you also track where in the store a product was picked into a cart?

What I think is an interesting angle you guys have is that you integrate with existing hardware/ security cameras in stores. Solutions sold to me required us to install new cameras to gain insights on our brand's aisles. It may help your go-to-market to partner with major security camera companies (like Avigilon http://avigilon.com/) to sell your document as a part of a bundle.

Thanks Nathanlee, it’s great to hear about your experience in FMCG and some of the challenges you faced around A/B testing layouts and integrating sensors, I’d love to pick up a conversation with you about this.

We’re currently focused on speciality retail, especially where the store is more of a ‘show-room’ style such that product layouts and window displays can be more easily configured. We’re also considering partnerships with camera manufacturers as this is on our roadmap, but not mandatory at this stage as we integrate with any IP-camera system.

So many people talking about what a good idea this is, or how it has already been done or ...

Absolutely nobody talking about ... if we really, as consumers, want to be tracked everywhere all the time always.

Do you just assume that there's no way to stop it so might as well nerd out over it?

This is ... So weird, considering how HN usually treats privacy.

I hope this company, and every company trying to track me both on and offline are destroyed by strong data privacy legislation.

Enough is enough.

>We don’t rely on facial recognition, instead taking in visual cues from all features across the body.

This is somehow better? No, it just lets you add the misleading marketing point that you don't use facial recognition. Get it guys? IT's not facial, because it's seeing ... the whole body!

What the actual heck?

I don't mean to be negative, but I hope they have some large sign outside the stores using this technology, so I can choose not to shop there.

All of our online activity is already tracked with great accuracy. Our phones are being used as beacons tracking our every move, listening in on conversations, basically invading our privacy in every way possible.

I'm absolutely not a fan of all this tracking / analytics to increase conversion rates hype. That is what got us the sites with a million popups and psychological tactics that trick us into hitting that buy or subscribe button.

This is a real life version of that. No no no no no! Just NO!

I am sorry for crapping on something you guys have doubtlessly worked hard on, it's a shitty thing to do.

I'm just still kind of curious why anyone thought this would be a good look though? Like, the writing has been on the wall since before GDPR. The anti-facebook backlash alone has gotten more and more shrill over the last 9 months.

And you seriously thought that now is the time to launch a product that brings this same dehumanizing technology to the real world?

We've known for 15+ years that anonymized data ... isn't, from AOL search, from Netflix, from ... And so you include that as part of your pitch?

Nothing about why we should trust you, just it's totally anonymous, we promise.

And I realize this is being marketed to lizard people in suits who view humans as numbers to be optimized, but just, why is this even within the possibility space of new companies at this point?

While I agree with you; I believe this is non-reversible. If it's not this company, 10 others will be launching their products in the next few months. Humans have become hackable and unfortunately our strength to innovate is the one that's becoming our bane. As Yuval Harari points out; in the short future companies like Google, FB will know more about us than we do.

Yes this is going to inevitably lead to your phone / face / way you walk being a "cookie" that follows you offline and online. They say 'while maintaining privacy', but once a FAANG buys them out, that is probably going to be over, let alone governments tapping in.

When you enter a private business, they have every right to track you as they wish.

So why don't they send a person to follow me around, noting down specifics like what clothing I look at and how many seconds I take to poop? Because this would be creepy

But since they're doing it with invisible technology that most people don't know about, it's okay, right?


Furthermore, just as when the government does it, there is a qualitative, not merely quantitative difference between individualized data collection and total aggregation. It's legal for the cops to tail a suspect. IT's not legal for the cops to put a tracking device on everybody's car just in case.

And finally, a business may legally be able to track me in their own store, but where is my assurance that those data are not shared with other businesses, ones that I may not have entered?

Well.. if they notified me upon entry exactly how I would be tracked, I wouldn't even enter. But they don't do that, do they?

They also have the right to kick me out, but I hope they don't do either.

Michaelvoz is right, the expectation of privacy does not exist any more. Thats why I carry signal jammers to every bathroom out there!


There are DOZENS of companies after this space with solutions ranging from point-of-sale, to beacons (facebook was aggressively rolling into this space, have heard less after privacy missteps), to WIFI range tracking, to some combo of the above.

Aura is a "smack self on forehead" level of ingenuity / execution - I love it, think you guys are on to something big. Look at the institutional market too - literally an analytics dashboard / workflow that provides forward looking reporting on traffic / trending would be hugely valuable to REIT scale funds.

There are DOZENS of companies after this space with solutions ranging from point-of-sale, to beacons (facebook was aggressively rolling into this space, have heard less after privacy missteps), to WIFI range tracking, to some combo of the above

To add to your point, Walgreens is already using facial recognition to profile customers.


Thanks. Have to avoid them from now on.

Thanks! We've certainly explored that avenue in the past, working with city centres & BIDs to get street footfall data.

We're also talking with malls, who are trying to get footfall traffic data in myriad of different ways already, at differing levels of quality. They may not want to share their information with 3rd parties, but would be able to do trend analysis for their own purposes.

This is cool! There was a very similar startup, Prayas Analytics, that also went through YC (in 2015) and ended up shutting down: https://medium.com/@pranshum/10-lessons-on-3-years-with-pray...

Curious to hear your thoughts on what's different this time around?

Interesting questions - we haven't actually come across them before so I'm just going on what I learnt from that blog post, and I may not have the greatest appreciation for their specifics.

- Our strategy is to not start with large grocery retailers for the exact reasons they mention. They're slow to implement new tech and slow to change things in stores. Our primary target is mid to large-size speciality retailers, with low conversion rates, where a small change in conversion rate can have a big effect on sales.

- We've honed our camera integration and have made installation easy and fast with existing cameras. Our clients have even done the install themselves before.

- The post talks about the stores needing to actually make use of the data and our plan for this is to help them do the things they change up semi-regularly anyway (product layouts, visual merchandising, window displays)

- With data transfer, we try not to use the stores WiFi if possible, and use 3G dongles instead. We've managed to use new compression algos so we don't use a huge amount of bandwidth

Meta: I'm fascinated you've gotten as far as launching without anyone at YC mentioning a company doing something so similar three years ago. Seems like a failure on YC's part; maybe they thought there was nothing to learn from the other company but still seems surprising.

That makes sense. Thanks for taking the time to elaborate + best of luck!

Your explanation on how you plan to keep the data anonymised sounds good. I also love that you’re using existing cameras, I recently observed as a large it-consultancy piloted their multi-camera solution with a large Finnish grocery store chain and got rather poor results, mainly due to cost and amount work being too high when compared to the results.

This also instantly reminded me of the time I used to work in large thrift shop. Everyday the owner would count the daily visitors and compare it and the daily sales number to factors such as the amount of tables, changes in floor plans, amount of new items etc. He would have loved something like this.

Good luck with your product!

Thanks for the kind words plahteenlahti - it would be great to hear about your experience with the groccery store, what kind of data were they experience poor results from, was this because the data was poor quality / inaccurate, or simply due to not getting return once acted on?

First, the negatives. This space is enormously difficult. You have low margin, low technology clients with physical infrastructure. You have to get approval from operations, IT, and legal. Once you're up and running the data becomes stale fast and you have to produce insights like move this shelf here. Then you have to wait for them to do that, if they ever do, and demonstrate that the client is now seeing a ROI from that change. And prove it was you and not the new ad campaign they just ran or the better staff they have in place now that is making them slightly more money. I've been working in this area for years, also with a computer vision research background, also read Prayas's post several years ago and dismissed it, also started with specialty retail and focused on ease of install, also dismissed RetailNext for using their own hardware, etc.

Phew. Now, the positives! I do believe that someone will crack this at some point. There are several promising trends including a massive jump in the share of IP camera installs, massive drop in the price/performance ratio of GPUs, and steady increase in commercial broadband. And everybody loves data. Those trends, plus a pivot to optimizing physical business processes, are why I'm still working on camera software at my company, Perceive (www.perceiveinc.com). We got an interview at YC several years ago for this idea but were rejected.

So Jonathon and team, despite how this post started, I really do wish you guys the best. Your tech looks really good. I would love to talk.

"Our mission is to ensure that physical retailers can innovate and improve their stores with data, in the same way their eCommerce counterparts do, while protecting customer privacy."

Unfortunately "the same way their eCommerce counterparts do" is by violating customer privacy.

The only reliable way to protect customer privacy is simply to not collect the video and other data in the first place.

I gotta be honest, this doesn't look to great to me. Most stores, mid-to-low grocery stores, dont even track inventory completely automatically, let alone something as sophisticated at this.

How are you going to beat amazon that has expertise and actual stores to try this? Where is the tech break-through going to happen without that instantaneous feedback loop?

But going back to the first point..have you talked to target stores? I can't imagine them wanting to sign up on something that has very undefined value, potentially large backlash. They have much bigger problems on their day to day.

Amazon are currently more focussed on supplying services to developers through AWS, e.g. Rekognition that do one small, technical part of what we do. That and Amazon Go which is a cool concept that works really well for groceries/convenience stores. We're going after everything else and don't see Amazon making any moves to enter this space in the near future.

We've talked to lots of target stores and are live with a number of clients already. The story we hear most frequently is that they've previously tried a number of things to get this type of data but none have been good enough in terms of precision or depth of insight.

Good luck!

I joined a company with exactly this pitch in 2012. In 2016, I acquired a company with this pitch as part of a major international Telco. I have built a prototype of this technology and run a pilot with a major international conference venue.

The stores aren't interested in your aggregated data, and they wouldn't know how to use it even if they were. They're interested in insights which you can't have, because you don't know their business. You're not delivering enough value to make a business. Sorry.

I'd love to hear more about your reasoning. Obviously metrics and dashboards are shiny CTO-bait, but what would people be looking for in terms of business outcomes?

Like, could you do this on a consulting basis, where you run something like Aura Vision in store for a few tests, then come up with a business report?

I found this idea really interesting and I completely agree with you that stores want insights / actionable data instead of aggregated metrics.

Maybe start by focusing on one specific field (grocery store? fashion?), or become the technology provider to enterprise clients?

What if you used the data collected in stores to serve better targets ads online, paid stores to incorporate your tech and made money through more traditional online ad revenue?

That really makes sense.

the real value to be captured isn't footfall and such because the real efficiency of e-commerce isn't a-b t testing; it's retargeting. Amazon go and their competitors are solving this problem https://m.youtube.com/watch?v=yeS8TJwBAFs. I'm sure you're aiming at the same thing though.

Time to go back to my neighborhood grocery store where when i need some product he doesn't carry i talk to him and he will have it a few days later. I used to love technology and that's why i choose this field but now i feel it asphyxiating and that makes me sad, perhaps i'm just getting old.

There is also one company called V-Count(https://v-count.com/)


your privacy.

How do you process the video? On prem in a box or through the cloud? How do you deal with bandwidth concerns of retailers?


We've already asked you not to post shallow dismissals. Please stop.


Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact