Episode 1 - Introduction to POIs (Points of Interest) with Jason Richman, CEO of Safegraph

Welcome to the Location Insights Podcast
This episode features an interview with Jason Richman, CEO of Safegraph, discussing the fascinating world of Location Data and POIs (Points of Interest).

Safegraph website: https://www.safegraph.com
Safegraph POI Data Documentation: https://docs.safegraph.com

Topics covered include:

• Jason's background and experience leading up to the location data industry.
• Anatomy of POI Data:
  • Simple definition of POI Data and its common business uses.
  • How Safegraph collects POI data globally.
  • Strategies for keeping POI data FRESH (Up-to-date).
  • Safegraph’s POI Verification process.
• Current changes and trends observed in the POI data industry.

Have questions or like to be on the show drop us a mail: [e-mail] 

Creators and Guests

Host

Kevin

Global Marketing Manager at unerry

Guest

Jason

Jason Richman is the CEO of Safegraph

What is Location Insights?

Welcome to the Location Insights Podcast
Where we bring you information, news, and insights from Partners, Clients, and Data Scientists about: Location marketing, Retail Media, POI, Data Analysis, and other interesting topics.

Kevin: Hi, and welcome to Location Insights Podcast. I'm your host, Kevin, the Global Marketing Manager at unerry Inc.

Kevin: unerry is a location analytics platform and data insights company based in Tokyo, Japan.

Kevin: And with me today is Jason. He's the CEO of SafeGraph. Hi, Jason. Great to have you on the show. How are you doing today?

Jason: I'm doing great. Thanks so much for having me. I just want to first say it's an honor to join this podcast and speak with a Japanese audience.

Jason: Japan has an incredibly advanced spatial ecosystem, and I personally have always admired the precision and innovation that Japanese companies bring to this space.

Jason: I'm really grateful for the opportunity to share what our perspective is today.

Kevin: Excellent. That sounds like a great way to start off.

Kevin: So a quick step back, maybe introduce yourself and your background.

Kevin: So you're based in the U.S., right?

Kevin: Yes.

Kevin: And SafeGraph is based out of the U.S.

Kevin: So how did you come into the POI market and maybe a little bit about?

Kevin: Yeah.

Jason: So I'll take a quick step back.

Jason: So my name's Jason Richmond.

Jason: I'm a CEO of SafeGraph.

Jason: For those who don't know, SafeGraph is a pure play data company that builds high precision

Jason: data about physical places globally.

Jason: And we basically use our data to power analytics and applications for some of the world's top

Jason: innovators.

Jason: I'm based here in Miami, Florida.

Jason: I was previously in New York, but the SafeGraph team is a remote first startup.

Jason: And the way that I kind of found myself in this business, it wasn't a linear path.

Jason: So I'll kind of start from the very beginning.

Jason: I basically spent a better part of my career at the intersection of data and technology,

Jason: helping startups commercialize complex B2B products and scale them into real businesses.

Jason: I've been at SafeGraph now for eight years.

Jason: But prior to SafeGraph, I worked at a company called Metamarkets,

Jason: which was a SaaS analytics startup that catered to the advertising technology ecosystem.

Jason: So for those who don't know, in advertising technology, it's very similar to how high-frequency

Jason: trading works in stock markets.

Jason: Ads are getting bought and sold in near real time using algorithms.

Jason: And so there's a huge amount of data flowing through the advertising ecosystem.

Jason: And Metamarkets was a tool that allowed these different technology platforms to visualize,

Jason: process, and store vast amounts of programmatic advertising data.

Jason: And so I was leading sales at Metamarkets up until the exit to Snap Inc, which owned Snapchat back in 2017.

Jason: And then through that process, I ended up meeting Auren Hoffman, who was the original founder of SafeGraph.

Jason: Now, Auren was a serial entrepreneur who also founded a company called LiveRamp.

Jason: And so I got really excited when he reached out to me as an investor of my prior business and basically recruited me to come join SafeGraph as one of the first commercial hires.

Jason: And so I started out as one of the first sales hires, eventually worked my way up into revenue leadership.

Jason: And then last year, I stepped into the role of a CEO.

Jason: And so I've been in the CEO seat for the last 14 months or so.

Jason: But across everything I've done, the through line is pretty simple.

Jason: I love building companies and I enjoy the crazy, chaotic, fast-paced startups.

Kevin: Awesome. Awesome.

Kevin: That's a great way, I think, to start off with the introduction of yourself.

Kevin: Now, for SafeGraph, you're all about POIs, right?

Kevin: So maybe just for listeners, if you're not familiar, in simple terms, what is your definition of what POI data is?

Jason: Yeah, that's a great question.

Jason: So POI data describes a specific location that someone like you or I may visit.

Jason: It may tell you what the place is, where it is, and any key metadata about the location.

Jason: So an example could be your local restaurant, gas station, a Starbucks, a mall, a strip mall, even a doctor's office or an EV charging location.

Jason: It's basically anything where people interact with the physical environment.

Jason: You know, POI data is the backbone of almost every single location-based application,

Jason: from navigation to retail analytics platforms and fintech data enrichment APIs.

Jason: And so POIs are really, really exciting a space to be in.

Jason: You know, what originally pulled me into the location data industry, to answer your earlier question, it wasn't some long-term plan, right?

Jason: I didn't go to school.

Jason: For GIS, I actually went to undergrad business school and studied finance.

Jason: But for me, it was really about just intellectual curiosity.

Jason: When I started reading up on the location data industry, I was really fascinated by all the use cases that require accurate POI or location data, things such as retail, CPG, financial services, and mapping.

Jason: And then I had previously worked in software sales.

Jason: And so when I heard about a data as a service or data only business, I got really excited.

Jason: And then fast forward, I started talking to actual customers in the space.

Jason: And through that conversation, I really appreciated the opportunity in front of me when I realized that there was a real problem around, you know, building accurate data at scale.

Jason: And I recognized that SafeGraph was in a unique position to become a core player in this ecosystem.

Kevin: Awesome.

Kevin: I think that's a great introduction for listeners that are new to POI and want to have a better grasp in terms of what it is.

Kevin: So what are some of the typical use cases that a company would use POI data for?

Jason: Yeah, so when you're a data business, naturally you can be horizontal because you're a data ingredient or utility.

Jason: Now, we have definitely decided to focus our ICP or ideal customer profile here at SafeGraph,

Jason: but I will run through a couple of typical examples

Jason: that we see for commercial allocations.

Jason: So back to that advertising or ad tech use case,

Jason: there are companies within programmatic advertising

Jason: that build audience segments

Jason: and audience segments allow advertisers

Jason: to more efficiently target and reach their consumers

Jason: with ads at the right time.

Jason: And so there are audiences out there

Jason: that are built off of location data.

Jason: These are companies working with cell phone GPS data

Jason: coming off SDKs.

Jason: and these raw location things are tied to advertising IDs or mates.

Jason: And these audience builders need a really good way to enrich these raw location signals

Jason: so they understand the behavior of consumers for these audience segments.

Jason: And so SafeCrep comes in by providing a very accurate map of the visible world

Jason: by way of licensing POIs, and then more often than just giving POIs,

Jason: we're also giving polygon data.

Jason: We call that our geometry product, which gives really precise tenant level shapes or boundaries of POIs so that companies can do more accurate attribution of location data to places.

Jason: And then in return, build more accurate audience products for their advertisers and advertising technology partners.

Jason: So that's one example.

Jason: Another example might be working with a large multinational retailer who's trying to expand its brick and mortar footprint.

Jason: right so it's a classic real estate or retail site selection use case you know you're going

Jason: into a new market maybe you're looking around tokyo to open up a new lawson's and you want to

Jason: understand if this is a good location for every lawson's location uh and so you basically want

Jason: to look at a catchment area around a prospective site and it's really important to have full spatial

Jason: awareness of what are all the types of places co-located within that catchment area now those

Jason: might be competitor locations. They might be other types of retail generators. It might be transit

Jason: stops for schools, but POI data is a core ingredient or input into these site selection

Jason: models and workflows for these data mature retailers and QSR chains. And then last but not least,

Jason: another common use case might be in the financial services sector. You have private equity funds who

Jason: are trying to acquire companies that have a physical world or brick and mortar presence.

Jason: And so naturally they use POI data for target identification, doing pre diligence, looking

Jason: at sort of PAM analysis or white space analysis.

Jason: Again, these are funds that are specifically going after businesses that have that physical

Jason: world presence, but the POI data is typically layered on with other types of data sets to

Jason: inform some type of investment decision.

Kevin: Very, very interesting.

Kevin: And from what I understand, SafeGraph actually has POI data, not only from the United States

Kevin: where you're based, but also like globally.

Kevin: So countries all over the world.

Kevin: So like around how many countries do you cover and how do you get the data from so many different

Kevin: places?

Jason: Yeah.

Jason: So before I get into the specifics there, I think it's important to step back and highlight

Jason: that SafeGraph is very unique in this space and that we are one of the only true pure play data

Jason: companies and what I mean by that is unlike some of our peers we only license a flat file of POI

Jason: data we don't provide any software we don't do insights or analytics we are just focused on

Jason: being this high quality weighted ingredient so that our customers can rely on facts about the

Jason: physical world and build their sort of derivative software analytics and services on top of data.

Jason: Now that maniacal focus allows us to really scale the global POI product with really high degrees

Jason: of quality. So the core SafeRef product today is called Places. And to answer your earlier question,

Jason: we maintain a global database of over 80 million POI across 200 plus countries and territories.

Jason: Again, as I said earlier, these places may include things like a Starbucks, a Lawson's,

Jason: and McDonald's. It could also include an Amazon warehouse, a Tesla supercharger, a park, a school,

Jason: or your local mom-and-pop grocer. All of this data today is provided to our partners as a flat file,

Jason: typically through cloud storage on S3, Snowflake, or the equivalent. And then we make sure that we're

Jason: updating this data on a monthly cadence so that our customers and partners are getting the freshest

Jason: data possible for their downstream analytics and applications. In addition to providing obviously

Jason: the location, you know here is a Starbucks on 555 Main Street, SafeRamp has also built a very robust

Jason: library of rich metadata or attributes about these locations. So we offer very flexible

Jason: licensing configurations where somebody can say okay I know the name, the address, the category,

Jason: in the phone number, I also would like a building footprint or the square footage or the hours of

Jason: operation or I'm looking at restaurants and I want to understand the types of amenities or even the

Jason: cuisine that is offered at a location. And so it's not just about getting the rows correct,

Jason: but it's also about bringing in really rich, accurate metadata about these locations

Jason: to inform various use cases for the customers.

Kevin: yeah that's that's really interesting and a huge huge number of pois there you said over

Kevin: 8 million 80 million 80 million well yeah that's kind of um so like what what continents if you

Kevin: sort of break it down by continents is is that covering yeah so if you if anyone who's listening

Jason: if you go to our uh our doc site at docs.safegraph.com we provide the most transparent technical

Jason: documentation around the Safe Graph Places product. And on there, you could see a stack ranked list

Jason: of all the geos that we support and the various coverages across these geographies. But we,

Jason: you know, sort of look at it in terms of tiers. So we're a US-based company. We have very strong

Jason: footprints in, you know, North America, Western Europe, parts of APAC, such as Australia and Japan,

Jason: for example, and that in LATAM, you know, we have a pretty big database in Brazil.

Jason: We may not have as many rows of data as a headline number compared to some of our peers,

Jason: but that is actually by design because more data doesn't necessarily mean always good data, right?

Jason: What we do is we're a little bit more systematic of how we go about building data

Jason: because we're constantly thinking about recall versus precision.

Jason: And what we want to ensure is that we're not introducing false positives, duplicates, or permanently closed locations that ultimately would end up in our customers' pipelines and products and create sort of data cleanup challenges for the customer, right?

Jason: So that's how we design the product.

Jason: But we also do strategic data sourcing.

Jason: So somebody can approach us and tell us, hey, I really want to do a little bit deeper R&D in this specific market.

Jason: And we have very scalable infrastructure that I'll highlight a little bit later in this podcast that allows us to go and build up those countries at scale for specific use cases.

Jason: I think the key takeaway for the audience to remember is that as a data company, our value proposition to our partners is we just want to curate, validate, and transform places data.

Jason: so that our customers can really save time and money and focus on the derivative,

Jason: which is building out their analytics, their models, and their applications for their customers.

Jason: So the analogy I like to say to people is I'm not building a cake,

Jason: but I'm providing really high-quality flour and sugar so that people can go build their own cake.

Kevin: Right, right. Yeah, I think that's a great analogy for the type of data and the data set itself

Kevin: that your company provides.

Kevin: So I think you touched a little bit, but maybe if you can sort of give an overview.

Kevin: So how do you keep the POI data fresh, right?

Kevin: There's new stores opening, new stores where some stores aren't doing well.

Kevin: They shut down, they close, you know, restaurants, especially coffee shops.

Kevin: There's very fast turnover rate.

Kevin: So how do you update all these POIs?

Kevin: Yeah, it's a really hard problem, right?

Jason: Which is why we decided to only focus on this problem and just be a pure play data provider,

Jason: because it requires that level of attention so as you correctly called out

Jason: places are constantly changing to be clear I don't think anybody out there

Jason: who's in this space is going to have a data set that is a hundred percent

Jason: perfect because data is a moving target right stores are constantly opening

Jason: stores are closing stores are rebranding open hours are changing right cuisine

Jason: types on the menu might be changing. So to sum it up, it is a hard problem. Now, we have a phenomenal

Jason: engineering and product team here at SafeGraph that I've been working with for many years.

Jason: And we have a very robust data sourcing and pipeline process. And so it starts with sourcing

Jason: or gathering data. So we go out and we crawl the web. We go out and we crawl every single week,

Jason: thousands and thousands of public websites where we bring in different pieces of metadata about a

Jason: place right and really we do it from thousands of sources because we don't want to be reliant on one

Jason: single source of data instead we want to bring in as much data as possible and cross-requence

Jason: different components of data from different sources to basically understand is this place

Jason: open or is it a real place right um once we bring in that data from thousands of sources

Jason: we then run some preliminary sort of etl on that data right so we're cleaning we're parsing the

Jason: data from input sources uh we're enriching some of the input data right we have these like

Jason: data classification models where we go in and we tag this is a pharmacy or this is a gas station

Jason: and then we push these inputs downstream into our data pipeline for further sort of enrichment and

Jason: cleaning right but really what's happening is there's all the sourcing and data coming in

Jason: and then there's just a lot of data conflation that happens to determine whether or not a place

Jason: has actually existed in the physical world and you know it's really important that we do these things

Jason: because for many enterprises out there bad poi data specifically can cost 10 to 100 times more

Jason: than high quality data costs a license right imagine for example you use poi data in a

Jason: navigation system like an av system right and that data is incorrect and you are routing people to

Jason: the wrong location right well that person could potentially be put into an unsafe condition or

Jason: they can churn from your app and no longer become a customer because of a poor customer experience.

Jason: And so I could go on and on around, you know, the implications of bad data, but this is really

Kevin: why we spend so much time on keeping this data fresh and today. Yeah, definitely. It's a huge

Kevin: challenge considering the scale and the number of POIs, but also like the, you know, the sort of

Kevin: ad hoc frequency that stores may open or close, depending on how their business goes. So like,

Kevin: You touched a little bit on it, but do you want to add anything on SafeGraph's POI verification model or process or any kind of metrics that you use when validating?

Kevin: Yeah, so it is like a multi-layer process, to be clear.

Jason: So I touched on, you know, the sourcing process where we do a lot of web crawling.

Jason: There's also anomaly detection, entity resolution, classification, deduplication, as well as AI and human QA efforts.

Jason: But essentially, our verification model blends machine learning, rules-based QA, and human oversight to eliminate false positives and maximize accuracy.

Jason: So, for example, our customers, if you take one of our columns, we have category information about a POI.

Jason: So, we'll tell you this is the NAICS code of a location, right?

Jason: And then here's the top level and the subcategory of the location.

Jason: It's really important for a lot of our customers that we don't mislabel places, right?

Jason: An example would be like an advertising technology.

Jason: Companies use our data as like a blacklist

Jason: or suppression list to make sure

Jason: they're not geo-targeting sensitive locations, right?

Jason: So the stakes are really high.

Jason: And so to do best in class industry categorization on POIs,

Jason: we do multilingual LLN based models, right?

Jason: And then we do this AI agent post-processing to fix errors

Jason: when we do make a mistake before it goes to the customer, right?

Jason: That's just one example of verification specific to categories.

Jason: The other thing too is conflation, right?

Jason: As I said earlier, not relying on a single source,

Jason: but taking in thousands of sources

Jason: and then using an ML model for identifying groups of input POI

Jason: that correspond to the same place

Jason: and then merging those different sources together, right?

Jason: We also have built out address parsing capabilities.

Jason: So we have a suite of NLP models for parsing addresses

Jason: into constituent components, right?

Jason: Or we're building out this data.

Jason: And then last but not least,

Jason: we have something internally called a reality model,

Jason: which is an upstream ML model for scoring raw POI data

Jason: and a post-conflation model for identifying any fake POIs

Jason: that have sort of made its way through the pipeline.

Jason: And then it allows us to kind of go in and course correct

Jason: before we give that data out to our partners, right?

Jason: And so really the takeaway here is, you know, we don't just crawl data.

Jason: Like, you know, we're not a crawl company.

Jason: We are a true data product, right?

Jason: So there's so much work that goes into the product every single month before it's ready

Jason: to go and it's delivered to our partners on that monthly cadence.

Kevin: Yeah, yeah.

Kevin: I think that's a great sort of high-level overview in terms of what is involved with

Kevin: handling this data, checking it, verifying it.

Kevin: There's a lot that goes into it.

Jason: Exactly. And I'll say this, like, we've always been very sophisticated. You kind of have to be when you're just focused on this data problem. But I'll say that the tailwinds of AI have given us even greater leverage in technologies to continue pushing the pace of innovation here at SafeGraph, right, and within this specific industry.

Jason: And so we've definitely been dabbling. You know, we've always been using machine learning and neural networks and large language models, as I mentioned earlier.

Jason: But we definitely have made an effort internally to build out a genetic workflow.

Jason: So we've created a couple of AI agents ourselves that sit in different parts of our pipeline from crawl to classification to verification.

Jason: And so, again, we feel like we're at the forefront here of this space in terms of delivering the best product possible for our partners.

Jason: And again, it all comes back to data veracity, and we take that very seriously.

Kevin: Yeah, yeah, I think that's a great way to describe it.

Kevin: I think one of the kind of standouts or things that the SafeGraph has that other of your peers don't have.

Kevin: you touched on a little bit at the beginning, I think, but I wanted to sort of highlight it,

Kevin: is that you create polygons for the POIs themselves across the United States. Is that correct?

Jason: Correct. So not only across the United States, but also in some of our global markets as well.

Jason: I don't have it in front of me. I think we have somewhere close to 40 million global polygons,

Jason: with 23 million being in the U.S., and we refer to that as the SafeGraph Geometry product.

Jason: Now, the way we ended up building that product was many, many, many years ago, we used to

Jason: work with some ability data products and point data was simply not good enough to take that

Jason: GPS data that I was describing earlier and accurately joining into a place, right?

Jason: Because GPS data has something called horizontal accuracy, which is the estimate of uncertainty

Jason: under location ping. And so instead of doing like a point radius approach to calculating store

Jason: visitation, we recognized early on that people really need to work with polygon boundaries

Jason: and not just building footprints. We're very unique in that we took it a step further. We

Jason: actually developed what we call tenant split polygons. So an example would be taking a

Jason: strip center. Safe Graph would have a correct polygon wicket or weld in text shape of the

Jason: actual strip center itself.

Jason: And then through rooftop segmentation models, we would slice up that building into individual

Jason: tenant polygon shapes.

Jason: And then we also provide our customers with other metadata about the polygon.

Jason: Like is this polygon hand drawn?

Jason: Is it synthetic?

Jason: Is it shared by more than one tenant?

Jason: Or is this polygon co-located or located inside of a larger structure?

Jason: And that's really important.

Jason: And I'm happy you called that out because it's one of the best products for us in some of the verticals that I mentioned earlier, where companies are really working with very advanced, large-scale telemetry data.

Jason: And again, really want to accurately resolve that telemetry data to a place with low-false positives.

Jason: And so, yeah, I'm happy you called that out.

Jason: And for anyone who's listening, not to pitch it too hard, but if you'd love to get a sample and check it out, feel free to reach out to me.

Kevin: yeah yeah definitely it's one of the the things that i think sets your company safegraph apart

Kevin: from the other poi vendors um is that polyon product that for companies that really want

Kevin: that idea of the shape of the building and buildings come in all different shapes and sizes

Kevin: it all depends on the available retail space and real estate space um you provide that it's it's a

Kevin: fantastic idea and it's a great sort of data set solution so before we wrap are there some

Kevin: changes or trends that you can share from your perspective within the poi location data perspective

Kevin: i think you touched on llms is definitely something that's new and you know has been

Kevin: growing certainly in the last two three years or so um but what what other sort of trends

Kevin: if you can share from your perspective around POI and location data in general.

Jason: Yeah, I'll run through a couple.

Jason: So as you correctly called out, AI is raising the bar for accuracy and data standardization.

Jason: As you probably know, in the audience knows, AI systems are only as good as the data that

Jason: they're trained on.

Jason: So for large language models, bad data means hallucinations.

Jason: For real-world systems, such as an in-vehicle navigation system, it means incorrect routing

Jason: or failed deliveries, right?

Jason: I think many of us have heard the term over the years garbage in garbage out and that's why data

Jason: accuracy matters so much in location data. A misplaced store or a mislabeled category may seem small

Jason: but for an autonomous vehicle or an LLM agent it could actually cascade into major downstream

Jason: issues and this is why SafeGraph has chosen to put all of our focus into this data problem of

Jason: maintaining accurate places at scale so that's definitely one of the key trends or i guess

Jason: tailwinds that i've been seeing another interesting insight and i don't even know if this is just

Jason: location specific but this might be data specific right i think this actually impacts all data

Jason: businesses is that these large enterprises in the fortune 1000 you know they have a ton of first

Jason: party data right if you take a large public publicly traded retailer they have crm data on

Jason: customers they have point of sale data but what's happening is these companies are realizing that

Jason: first party data alone is not enough right to give them a competitive advantage they need

Jason: external data and so that's why companies like SafeGraph exists for poi you've got credit card

Jason: data companies, you've got satellite data companies, you've got SKU level product data companies,

Jason: you know, they have data on their own customers, their own point of sale, but in that sort of use

Jason: case around retail, you know, we can show them data on all their competitors locations, right?

Jason: That's external data that adds incremental value to a site selection use case,

Jason: a catchment analysis, or any sort of competitive intelligence model, right? So I think that's a

Jason: really good tailwind and I think it's only going to continue to grow and expand for the

Jason: data industry as a whole, especially as these companies become more data mature, right? And

Jason: they get hungrier for more and more external data. The other thing I would like to call

Jason: that as a unique insight is that I think the mode around these types of data businesses

Jason: is increasingly around data verification. And what I mean by that is, you know, there are

Jason: companies out there that just provide data crawl services right and you see a couple of these like

Jason: ai crawl services but anyone can go out there and write a crummel script right and call the website

Jason: but very few can really really do data verification right like detect false positives

Jason: do entity resolution handle duplicates avoid category classification hallucinations and so

Jason: So I think for anyone who's out there building a data business, you know, crawl and sourcing

Jason: of data is very important, but I think the mode outside of distribution is around, you

Jason: know, data verification and being really, really good at that.

Jason: So those are just a couple of the insights.

Jason: And I hope some of those things are useful for your audience.

Kevin: Yeah, definitely.

Kevin: I think it's been a very insightful conversation.

Kevin: I think we covered a lot in terms of an introduction to POIs and to Safegraph in particular.

Kevin: So if listeners that want to find out more, where should they go to?

Jason: Please should go to SafeGraph, our website:

Jason: So www.safegraph.com.

Jason: And then for the more technical audience, please check out our doc site at docs.safegraph.com

Jason: And then people could always just reach out to me, Jason at safegraph.com, if you want to shoot me an email.

Jason: And I'm happy to respond and engage with the audience.

Kevin: That's great.

Kevin: So thanks for everybody who's listening again at home.

Kevin: This has been Kevin, the Global Marketing Manager at unerry.

Kevin: Subscribe to this location Insights podcast feed and you'll get episodes automatically in your favorite podcast app. Thanks for listening.