Dec 6, 2017 · 25m · 20vc
20VC: The Value Chain of Machine Learning, Is There Really An Incumbency Advantage in ML & Will The Rise In Cyber Remain For the Long Term with Jake Flomenberg, Partner @ Accel
0:00 / 0:00
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of The 20 Minute VC, host Harry Stebbings interviews Jake Flomenberg, Partner at Accel, about evaluating artificial intelligence investments, building defensible data moats, and navigating the cybersecurity startup landscape. Drawing from his operational experience at Cloudera and Splunk, Flomenberg provides frameworks for assessing venture opportunities and avoiding common founder pitfalls.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Harry holds 29.4% of the talking time here. How this is scored →
Harry as informed peer 2.3 Guest teaching 4.2 Guest disagreement 1.2 Harry pushing back 0.7
speaking balance: gold is Harry, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 18:27 Calling out overhyped ML anomaly detection
Jake explicitly dismisses industry hype surrounding standalone ML anomaly detection in cybersecurity, noting CISOs often find uncontextualized anomaly alerts useless.
Hardest push from Harry ▶ 11:26 Challenging data set value with counter-expert quoteHarry directly pushes back on Jake's thesis on data capture by citing Aaron VanDevender's claim that large data set value is largely overplayed.
Biggest teaching moment ▶ 6:03 Explaining the AI investment triangle frameworkJake educates Harry on why algorithms are rarely durably differentiable and why workflow capture is the true foundation of sustainable AI software moats.
Harry holds his own ▶ 11:26 Citing industry peer to test guest thesisHarry demonstrates his industry fluency by invoking Founders Fund partner Aaron VanDevender's counter-perspective on large data set inflection points.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Harry as informed peer | Guest teaching | Guest disagreement | Harry pushing back | Why |
|---|---|---|---|---|---|---|
| Jake Flomenberg's Career Background and Operating Insights | 1 | 3 | 0 | 0 | Harry asks standard career background questions to open the interview. Jake provides detailed operating history across Lockheed, Cloudera, and Splunk, highlighting how scale creates inefficiency and citing Clay Christensen's disruption theory. | |
| The Three-Legged Stool Framework for AI Investments | 2 | 5 | 1 | 0 | Harry references Jake's quote about making AI non-BS and prompts him to explain his investment framework. Jake educates the host on his three-legged stool model, explaining why algorithms alone lack durable differentiation compared to workflow and data capture. | |
| Data Capture Strategy and the Demisto Case Study | 5 | 5 | 2 | 3 | Harry demonstrates technical fluency by citing Aaron VanDevender of Founders Fund to challenge the assumption that large data sets are always valuable. Jake politely nuances the argument by discussing cold-start problems and workflow-driven feedback loops using Demisto as a case study. | |
| Derivative Data, Synthetic Training, and Big Tech Moats | 3 | 4 | 1 | 1 | Harry introduces advanced topics including derivative data via supervised learning and synthetic data simulation. Jake unpacks the limits of simulated training data in high-stakes environments like self-driving cars compared to video games. | |
| Feature vs. Product Dynamics and Cybersecurity Opportunities | 2 | 6 | 2 | 0 | Harry prompts discussion on feature versus product dynamics and cybersecurity trends. Jake pushes back against cybersecurity market hype, explaining why standalone machine learning anomaly detection is overplayed because CISOs cannot act on uncontextualized alerts. | |
| Quickfire Round with Jake Flomenberg | 1 | 2 | 1 | 0 | Harry conducts a standard quickfire round. Jake criticizes common venture capital advice to raise as much as the market will bear and shares operational insights on startup talent acquisition. |
Statements from this episode (12)
Flomenberg: Large organizations suffer from inherent inefficiencies of scale
“There's an inherent inefficiency of scale at any company, and it's no disrespect to Lockheed Martin. Once an organization reaches a certain size, the amount of overhead and planning makes it very, very hard to move nimbly, and quite frankly, the incumbent natu…”
Flomenberg: Startups must instill self-disruption values while young
“If we need to go back to the drawing board, if we need to give up this line of revenue stream for something that's going to be more exciting in the future, I think that gets harder and harder for you to do as the company grows. You really need to instill that …”
Flomenberg: AI startups cannot build durable differentiation solely through algorithms
“A lot of the companies that I meet, they come up to me and they picked a spot that's very close to that algorithm dot on the top, and my concern with that is it's very hard to imagine an algorithm that's durably differentiable.”
Flomenberg: AI startups must build workflow software before building models
“It comes back to this notion of workflow first, data capture, followed by model building and iterations.”
Flomenberg: Prime vertical SaaS opportunities require zero initial data
“That's really where I think the bulk of the opportunity is going to be going forward, particularly in areas like vertical SaaS, where I can improve the current experience without any data, and any data that I can collect to improve your life is just gravy on t…”
Flomenberg: Data feedback loops provide stronger moats than initial dataset size
“That's where I think a lot of this feedback loop and sort of supervised learning around the data is Actually becomes a little bit more of an advantage than the original data set itself.”
Flomenberg: Synthetic data gets self-driving cars 99.9% there but misses edge cases
“In the instance of self-driving cars, It could make a lot of sense, and I think it can get you 99.9% of the way there, but the question is, how fully reflective is the underlying data set of the real world, and in the instance where it's not 100% fully reflect…”
Flomenberg: Google will compete with ad-data startups but unlikely cybersecurity ones
“If you want to use some Google-like or Google-specific dataset for an advertising product, it's a really uncomfortable situation because at any point in time, Google may very well decide to build that same product. Now, for instance, in security, could Google …”
Flomenberg: Independent public companies require at least a $1B TAM
“If there's not a billion dollar TAM for this thing, it's not possibly the basis for a product that could support an independent publicly traded company.”
Flomenberg: Enterprise focus on cybersecurity is permanent due to nation-state threats
“Look, I think the focus and emphasis on security is here to stay. What we've witnessed over the past years, particularly from nation state actors, is reasonably new. And nation state actors, unlike your average hacker or scammer, they're not going after the lo…”
Flomenberg: Standalone ML anomaly detection in cybersecurity is overhyped
“Getting back to our conversation around AI and ML, one area that I think is unfortunately slightly overplayed is this machine learning anomaly detection to rule them all off by itself.”
Flomenberg: Advising startups to raise whatever market allows drives risky overspending
“Raise what the market will bear. A lot of startups dig themselves into a lot of trouble by raising, and more importantly, behaving like They don't have to worry about the next round, and overspending can put you in very tricky situations.”