Data Quality Meets Security – Jeff Jonas, Senzing

This is Textron TV. Well, the great pleasure being joined by Jeff Jonas Jeff is Joan SEO of senzing. Hopefully I said that company name right to get it right Jeff.

Yes, you did zenzing. It was The Zing. Excellent.

Watch us a little bit about yourself. Tell us a little bit about what sensing does. I've spent my life building software systems and over the years as a custom software developer.

We kept running into entity resolution problem, which is where you're trying to match records is Elizabeth the same person. And it just kept coming up over and over and over eventually we created a product so we didn't have to keep rebuilding it. I sold that product to IBM in 2005 and then while at IBM I said, you know for 50 million dollars.

I think you could build I could build a sixth one that we really really good and affordable to the world. So we started no nine and in 2012, we started shipping it and in 2016. I spun my team in the technology out of IBM.

So today I'm the founder and CEO of sensing and We're trying to make entity resolution record matching link detection. Fuzzy Duke what everyone call it? We're just trying to make it easy for the world.

Like why would you write your own spell check or grammar checker anymore? You just like plug it. That's what we're trying to do.

Well, and you know, I'm sure we all have multiple accounts on several, you know Services applications Etc, but think about that I mean Susan Sue s dot, you know just to start with on any one system more or less across multiple relationships that accompany might have with a customer Prospect or across multiple. It's it gives an idea of the size of the problem. Well, it really sits at the bottom of all customer relationship Management Systems all patient record systems banks have this trouble insurance companies.

They think you're three people you're really one I go check into the hotel. I got my ability card to go. No, they go.

Well, we just search it up. And they are which which Jeff Jonas are you and then they named three and I'm like, those are all me. This is a ubiquitous problem and it it makes analytics and correct it makes Machine learning and correct, you know if you think you have three customers, but it's really one.

Hmm and it's exceptionally problematic and fraud. You know, we've seen this at a big bank. They they terminated a customer for You know having bad money wondering type Behavior, but the same person turned back up in a different country with one letter off in their last name in a different passport.

They love their relationship with that bank. They just turned right back up Bank didn't know it like, oh great customer. Yeah, that that particular one is a fraudulent account.

But that one a knowingly is also fraudulent, but you think it's a different person, right? Yeah, just because there's like a letter off and so when do people call you in for help is this we've got a big systems integration project and conversion this we're building a new app or trying to Source data from our old systems. And what are some of the scenarios use cases where perfect time to bring in sensing?

Yeah. Well the main area right now that we've been focus on is there's plenty of people they might call it something different but they know what it is. They know that you have a problem with their matching and they're trying to figure out what to do next.

Like they realize they they're simplistic algorithms are working or they want to move it to the cloud or they want to be able to go to new GEOS and have a support, you know, Arabic or Mandarin and so, you know, that's what we being found the most right now are trying to make sure we can be founded by people that know what it is and know they need it. mmm, the biggest competitor is homegrown, but just people don't realize you can You can spend five ten million dollars and get something. That's only 70% good you can spend 30 million dollars.

I get something 90% good. It's still barely competitive. Well, and it's not a problem you solve once right at least you're finding all those to do if I can use that, you know entity mashing term.

Yeah, what happens is people get it is a little fascinating space. So you'd be surprised the number of people that just love. Tinkering over like oh, you know these records are a lot different.

But at the same, oh these records only have one letter off Junior and senior Jr. Sr. Wow one letter occupied to keep them apart, but then these records are so messy, but they're the same.

If you know fascinating and fun. So yeah a little bit about so I imagine someone in the audience might say oh this sounds like a perfect problem to solve with AI. Ml that audit that a lot of take care of it for us, you know, we say it a lot of things probably.

Well you better have a lot of data and then you better have a lot of patience to teach at all the spellings of Muhammad. And patience for matching super messy addresses the amount of labeling you have to do might be a universe lifetime. You know as we've combined.

Machine learning and off of large data sets and have models that are pre-built that are built in so it comes out of the box smart and then it sells Tunes in real time. An example is just comparing names. We've statistically learned off of 850 million names.

How to match names like you can look at the structure the name and go. Oh, that's Arabic. Well, it's an Arabic name culturally been in Hajj actually aren't part of the name.

But why should somebody have to go train their system for that label that like, how long does that gonna take? Hmm? So anyway, so what we've done is we've rebuilt in things like that with some models and then it as it's as it's ingesting it gets smarter.

an example of that by the way is let's say you're ingesting and one of the fields you have as a driver's license number because you're doing DMV data or voter data. And so driver's licenses are those are good identifiers, you know. But what if one of them is?

Does it turns out you know, like a year from now one driver's license like zero one zero one zero one zero one. And today there's one and then the next day there's four and the next day. There's 15 different people.

Pretty soon you have to learn that driver's license number is not a good driver's licensees are generally good but not that one. Hmm. And so what our engine does it learns that like in real time you and I also set for that's the right word from the first name last name can be first or last right Jeff and Jonas and mentioned actually, you know, my last name used to be Jonathan.

My grandpa was Swedish and they cut off the Sun and made it Jonah. So when I go to Sweden and check into a hotel and I go my name is Jeff Jonas, they go. Are it's like your name is Rick Sam or something, you know?

Yeah, you too. You didn't know it's a journey. Oh, wait, what's that?

You didn't know it's a joke in that country in that country. It was so how do you think's transpose is a very common problem and then we'll have twins born on the same day and name one Marianne and the other one and and Mary. Mm-hmm.

Interesting George Foreman in Golf Club of his boys George imagine that you get George Foreman with the home address in a home phone and you think you have a single human So essentially you've kind of looked at so many. It's such a large volume of data. Do you do this as a SAS service?

Is that how you kind of collect that data? Oh my train your ml or is this no implemented on a per side basis? I say that's a great question.

We've done a lot of work over the years in the Privacy space and building privacy features into our software and it let me have a point of view when we launch sending that I just didn't really want everybody's data. Hmm Crazy Chris work correction. It creates a big vulnerability if somebody were to steal it and you know, there's always this risk yet to worry about especially in the practice Community about collecting it for one reason and somehow unbeknowns to get being repurposed.

How's it being used? Yeah, so we send people our software and they run it on their cloud or on their premise. Okay, so my next questions, Yeah, then it solves the Privacy thing because there's a locally hosted API it sit behind their firewall and sits behind their login credentials and their security.

Yes. It's a link Library. It's like we've turned any resolution into three API calls.

I mean, there's more things there's other API called but you can now do any resolution but just you just compile it into Java python or see and it's like really it's willing as much as people get fast. Anyways, this should just be a mundane data quality tasks, like really only a better things to work on around here. That's the data validation process.

Yeah. Yeah into the matching. Interesting.

What are there? Even though the folks are running this in their own software in their own environments or in the cloud themselves? What are some of the privacy concerns, you know gdpr there's certain things that you have to be concerned about in your software or capabilities that you have to offer customers.

Yeah, you want to build a few things into these systems, you know, sometimes entity resolution systems will combine records like a loss e process though, they'll squish two records together, but forget where the parts and pieces came from the problem is an under gdpr or the California law CCPA or the other emerging products of us if you have to delete a record how you're going to delete the record. Yeah. Which one do you delete which one because it's been combined into one.

So one of the things that you have to do is pull attribution where you have to keep each record atomically pure to where it came from so that you could locate the record. No, you know, keep a current if it's changed. Imagine loading a watch list, by the way, then having something change in the watch list or record deleted from the watch list you better to get that out of there and then you build it to be able to do a gdpr.

Delete. And so you have to have full attribution you have to be able to atomically delete a record. And most of the Technologies out there.

If you go to delete a record, you just have to you know, you delete 50 records at the end of the month you reload everything. Well, our customers are 100 to millions of records of billions of Records, you'll enter. She always energy that users have to reload everything.

Mm-hmm. And so we allow you to transactionally delete, you know, so you can You can ingest you can stream in bulk data you can. Do real-time queries or updates based on onboarding new customers like for know your customer processes and then you can do individual gdpr deletes all at the same time.

You know to be able to demonstrate that you've done that that you've removed them or you know, how you've used their information or you've combined it or not or you know, so there's this kind of auditability of that accountability for what you're doing. Yeah. I it's a little randomly.

No, but since we're just chatting, I think there's a big thing to be said, I hope to blog about it someday, but these batch systems where you have to read boil the ocean every time a record changes. It's like a plus you get one new record and for 10 new or 10,000 you and you have to reboil the ocean the amount of compute that's being used for that. Mmmoil the ocean once a week.

Whether it's 10 million records or a billion. is so much more expensive than being able to transactionally just load 10,000 records like adding 10,000 records to a billion records in a streaming transaction engine like we have is like five dollars and you know two Watts or something, you know and making up numbers here, but it's just so and it's a lot of processes in the world that are batch and stuff and Anyway, that's expensive and unnecessary. I think well, very good.

It sounds like it sounds like a problem. Most of us are going to have we have customers. We're gonna duplicates.

A tell us a little bit about where folks can find out more and engage with you. com. We you can download our software for free.

We don't even ask for an email address is designed for developers. We don't build cars we build Transmissions. So if you're a job or a python programmer, or see we're gonna have a great great resources for you.

All of our technical docs are searchable online. Nothing's behind a account wall. Everything's just out there and we have a over 100 open source project.

So you can replicate the graph databases, um, you know, update and elasticsearch database off the back end. Input data from csbs more quickly blah blah blah. So anyway, we're just trying to make any resolution easy for the world is everyone can spend their energy on other stuff.

Mm-hmm. Fantastic. Well good stuff.

Thank you Jeff. Appreciate you stopping by sharing this with us, and I'm sure it's a problem that relates people can really relate to pretty easily. I don't like it.

Just trying to manage my own identities my own applications. I have where more or less the systems that are trying to manage me. So thank you for joining us and we hope to talk to you again soon.

Thanks, b****.