Dec 5, 2018 · 25m · mad
Knowledge Graphs To Understand The World // Hicham Oudghiri, Enigma (FirstMark's Data Driven NYC)
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this presentation and fireside chat at FirstMark's Data Driven NYC, Enigma Co-Founder and CEO Hicham Oudghiri explains how knowledge graphs and machine learning entity resolution transform vast public and enterprise datasets into actionable operational intelligence.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 4.5% of the talking time here. How this is scored →
Matt as informed peer 0.7 Guest teaching 0.3 Guest disagreement 1.0 Matt pushing back 0.2
speaking balance: gold is Matt, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 3:40 Deconstructing the enterprise data myth
Hicham pushes against the industry misconception that tech giant web-data solutions easily apply to enterprise real-world operational problems.
Hardest push from Matt ▶ 19:04 Challenging the data collection burdenMatt challenges Hicham on the operational friction and complexity of maintaining accurate small business data as businesses open and close.
Biggest teaching moment ▶ 6:00 Explaining terrorist list resolution mechanicsHicham educates the room on the stark differences between probabilistic web ad tracking and high-stakes bank compliance identity resolution.
Matt holds his own ▶ 19:04 Highlighting small business volatilityMatt demonstrates his domain expertise in startup data logistics by pointing out specific edge cases like barbershops and local SMB turn-over.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Matt as informed peer | Guest teaching | Guest disagreement | Matt pushing back | Why |
|---|---|---|---|---|---|---|
| The Proliferation of Data vs Insights | 0 | 0 | 1 | 0 | Hicham delivers an opening presentation monologue outlining Enigma's funding and mission to solve real-world data integration issues across institutions. The host does not speak or intervene during this section. | |
| Connecting the Dots in Real-World Data | 0 | 0 | 1 | 0 | Hicham explains the difference between internet bits data used by tech monopolies versus messy real-world atoms data. The host remains silent throughout the presentation segment. | |
| Challenges of Real-World Identity Resolution | 0 | 0 | 1 | 0 | Hicham detail how identity resolution works in financial compliance and defines the core metadata properties of knowledge graphs. The host does not participate in this segment. | |
| End-to-End Pipeline and Resolving SMB Entities | 0 | 0 | 1 | 0 | Hicham walks through SMB entity resolution challenges, such as identifying multiple 'Joe's Pizza' locations, and how machine learning assists matching. The host is not present in the dialogue. | |
| Augmenting Data Lakes into Knowledge Graphs | 0 | 0 | 1 | 0 | Hicham wraps up his presentation detailing how knowledge graphs surpass traditional data lakes for mission-critical banking and underwriting use cases. The host has not yet started the Q&A. | |
| Fireside Chat and Audience Q&A with Matt Turck | 4 | 2 | 1 | 1 | Matt Turck leads a friendly fireside chat asking about target clients, workflow placement, and the difficulty of SMB data collection before opening to audience questions. Both guest and host maintain an agreeable, highly collaborative tone throughout. |
Statements from this episode (15)
Enigma recently raised just under $100 million in funding
“We recently fundraised, ah, just shy of a hundred million dollars”
Oudghiri: Businesses have not done much to put data to work
“I still don't believe that businesses have done much. In terms of putting data to work.”
Oudghiri: Enterprise data value requires transforming internal customer data
“Half of the equation is not just about our data, but about the customer data and how the customer data needs to be transformed in such a way that you can actually take decisions on it.”
Enigma shifted from raw public datasets to building a unified knowledge graph
“Whereas we had exposed hundreds of thousands of data sets and told people, go at it, the reality of the matter was that, you know, we had to connect the dots in our data in the data ourselves, and we had to merge the data together to make it Into one data mode…”
Oudghiri: US state liquor licenses contain over 300 code categories
“I'd say upwards of 300 different code categories if you wanted to do a comparison as to who sells, who's allowed to sell, you know, wine and beer, and who's allowed to sell hard liquor.”
Oudghiri: Most enterprise data successes rely entirely on digital activity logs
“Most of the successes you hear about in, ah, the data space Enterprise or not. The data that they deal with is instrumented in bits, i.e., the data that they're collecting comes from a machine, most often user activity generated on the internet.”
Oudghiri: Banks deploy thousands of manual reviewers for sanctions screening
“When you're, you know, a bank issuing a credit card screening against a sanctions list of known terrorists and financial criminals in dozens of language with phonetic spellings and real-world spellings, and all you have is a, you know, set of names from the U.…”
Oudghiri: Real-world operational data lacks primary keys and consistency
“That's the reality that we deal with, ah, in the world of atoms, that data is inconsistent. There are no real keys.”
Oudghiri: Knowledge graphs depend on embedded metadata and graph traversal
“There's two key components. One, it's the ability to embed metadata in the data structure itself, i.e. The attributes are known. So if you're talking about a person, they kind of always look away. A company, they always kind of look away, and that helps you mo…”
Oudghiri: SMB corporate registration addresses do not match physical locations
“None of their addresses and their corporate registrations are actually the addresses that the locations are in.”
Oudghiri: Graph databases are not yet sufficiently mature for enterprise outputs
“Graph databases are not quite where they are yet”
Oudghiri: Data lakes centralize data, but entity models are required for connectivity
“The data lake journey basically just moves all of your stuff into one place, but if you really want to get a view of how your data is connected, you really need to think about making things much more about, you know, representing them in a much more entity spe…”
Oudghiri: Enigma's company graph covers 30 million SMBs and 1,000+ attributes
“The Enigma company graph is probably the one folks are familiar with the most at this point, which contains thirty million You know, different SMBs worth of data, and over a thousand attributes that we've been sourcing from public data, deriving.”
Enigma counts BlackRock, PayPal, AmEx, and BB&T as key clients
“BlackRock is a big customer. PayPal is a big customer. American Express is a big customer. BB&T is a big customer.”
Oudghiri: Entity resolution machine learning models cannot generalize across domains
“So it's definitely not generalizable, which is, you know, part and parcel of our IP, right? Being able to resolve companies and specifically small and medium businesses really well is different than being able to resolve people and being able to resolve, you k…”