# ML Unicorn Start-up Investor Tells-IT-All

George Mathew, Insight Partners | MLOps Coffee Sessions | Episode 126 | 51:01

Source: https://www.youtube.com/watch?v=u0Dyb1lpfg0
Channel: MLOps Community, now AAIF Live (https://www.youtube.com/@AAIFLive-x1r). Summarised by MLOps Talks.
Page: https://mlopstalks.com/talks/ml-unicorn-start-up-investor-tells-it-all
Published: 2022-10-04
Tags: build-vs-buy, data-engineering, platform-teams

## TL;DR
- George Mathew sees infrastructure automation moving toward large language models that can manage distributed systems at scale.
- He evaluates MLOps companies across data preparation, model development, and model governance and security, while also separating application-layer tools from algorithmic-layer tools.
- For early-stage software investments, George prefers simple deal structures, careful diligence, and founders prepared to build for a decade.

## Summary
George Mathew explains how he views the MLOps investment space from his role as a managing director at Insight Partners and from his earlier work at Salesforce, SAP, Alteryx, and Kespry. He expects infrastructure management to become increasingly automatable, with Shoreline Data addressing operational problems such as running out of disk in distributed systems. His investment framework separates MLOps into data preparation, model development, and model governance and security, then distinguishes application-layer tools from algorithmic-layer tools. He also discusses the differences between structured and unstructured data, arguing that a single MLOps stack will not fit every use case. George describes what he looks for in open-source communities, commercial software metrics, founder longevity, and efficient growth. He prefers Series A and later investments, usually with simple terms and a 1x liquidation preference at earlier stages.

## Key ideas
### Infrastructure management is moving toward promptable automation
[03:18](https://www.youtube.com/watch?v=u0Dyb1lpfg0&t=198s)
George says infrastructure automation is getting closer to the idea of prompting systems into existence. He points to Shoreline Data, an Insight Partners investment, which helps automate the work of standing up and managing infrastructure at scale. One example is detecting that a system is about to run out of disk and proactively deploying AI bots to handle the problem. He compares this direction with large language models being used for infrastructure deployment, although he describes the change as arriving gradually, "bit by bit, Brick by Brick." The opportunity exists because distributed, serverless, and Kubernetes-based systems are powerful, but difficult to operate reliably.

### MLOps is a tool chain that changes with the use case
[12:03](https://www.youtube.com/watch?v=u0Dyb1lpfg0&t=723s)
George describes his firm's MLOps framework as three parts: data preparation, model development, and model governance and security. He then adds another distinction. Some products operate at the application layer, such as Weights & Biases and Fiddler, while others work at the algorithmic layer, such as Deci, Neural Magic, OctoML, and Run:ai. This structure helped the investment team make sense of a crowded category. George does not describe a single fixed MLOps stack. He expects tools to move in and out of a tool chain depending on the problem a company is solving.

### Structured and unstructured data need different tooling
[15:25](https://www.youtube.com/watch?v=u0Dyb1lpfg0&t=925s)
George says the shape of the data changes the tools a machine learning team needs. Structured data may require feature management, either through a feature store or inside a database. For unstructured data, he says a separate feature store may be unnecessary because much of the relevant work is captured in labeling and annotation infrastructure. Demetrios adds that robotics, autonomous vehicles, medical imaging, time-series forecasting, and recommender systems have different service-level requirements. George agrees that MLOps will "speciate" and differentiate across use cases rather than settle into one universal stack.

### Practitioner-focused tools can become essential software
[19:17](https://www.youtube.com/watch?v=u0Dyb1lpfg0&t=1157s)
George is interested in tools built around the daily work of machine learning practitioners. He names experiment tracking, version control, and hyperparameter tuning as examples of that work. Weights & Biases stood out to him because it offered a strong product experience for this user and had already shown depth before he became an investor. He says the product became a tool that practitioners considered essential to their regular work. His broader test for future investments is whether a product creates the kind of experience that users feel they cannot work without.

### Open-source traction has to lead toward a commercial path
[32:22](https://www.youtube.com/watch?v=u0Dyb1lpfg0&t=1942s)
For an open-source company, George looks beyond GitHub stars. He examines pull requests, contributors inside and outside the company, contribution activity, iteration speed, and the community around the project. He then asks how that community could turn into a natural path to commercial adoption. He uses Metabase as an example. Its downloads, pull requests, GitHub stars, and open-source activity gave the company a strong base for attracting business users who wanted an easily deployable dashboarding and visualization tool. After commercialization, he looks at new customer logos, annual contract value, gross retention, net dollar retention, new annual recurring revenue, and customer acquisition cost.

### Enduring growth matters more than inflated valuations
[39:36](https://www.youtube.com/watch?v=u0Dyb1lpfg0&t=2376s)
George describes the market as returning toward its historical valuation range after valuations became overheated. He says this does not mean the companies are bad. It changes the cost of capital and the prices investors are willing to pay. His advice to founders is to build companies that last through good and difficult periods. He is looking for more efficient growth, stronger repeatability, and less dependence on spending heavily to produce top-line growth. He also asks early founders to understand that a Series A or Series B investment can begin a decade of work, rather than a short route to an exit.

### Early-stage deals should stay simple and carefully diligenced
[44:25](https://www.youtube.com/watch?v=u0Dyb1lpfg0&t=2665s)
George says Insight Partners is multi-stage, although his current focus is generally Series A and later. The firm can support companies beyond an initial investment instead of requiring them to find long-term capital elsewhere. For early-stage deals, he prefers simple structures and says he has not personally used more than a 1x liquidation preference. He avoids piling structure onto companies that are still figuring out how to scale. He also says his team does not send term sheets after a first meeting. The firm takes the time needed for diligence and an investment decision, even though the market's cooling has made the process longer.

## Notable quotes
- "The complexity of managing it exactly is profound." (05:50)
- "We broke it down into three big areas of the tool chain: data prep, model development, and model governance and security." (12:46)
- "Ultimately what we'll see over the next half a decade is that it is truly a tool chain." (17:46)
- "You find delightful software and those companies become the seminal companies of their prospective generations." (38:59)
- "You have to think about it in terms of a decade, not in terms of, hey, how could I flip this in a year or two and see what happens." (48:26)

## Tools & references mentioned
- Insight Partners
- Shoreline Data
- Kubernetes
- Heroku
- Ruby on Rails
- HashiCorp
- Terraform
- Weights & Biases
- Fiddler
- Deci
- Neural Magic
- OctoML
- Run:ai
- Alteryx
- Tableau
- Metabase
- Salesforce
- SAP
- Kespry
- Databricks
- Snowflake

## Who should watch
- You are building an MLOps or machine learning infrastructure company and want to understand how an investor divides the category.
- You run an open-source project and need to connect community activity with a credible commercial path.
- You are preparing for a Series A or later round and want George Mathew's views on metrics, diligence, liquidation preferences, and founder longevity.

## Related talks

- [Investing in MLOps](https://mlopstalks.com/talks/investing-in-mlops) (Leigh Marie Braswell, Founders Fund & Davis Treybig, Innovation Endeavors, 48:52)
- [MLOps Investments](https://mlopstalks.com/talks/mlops-investments) (Sarah Catanzaro, Amplify Partners, 46:18)
- [Founding, Funding, and the Future of MLOps](https://mlopstalks.com/talks/founding-funding-and-the-future-of-mlops) (Mihail Eric, Storia AI, 57:31)
- [The Current MLOps Landscape](https://mlopstalks.com/talks/the-current-mlops-landscape) (Nathan Benaich, Air Street Capital & Timothy Chen, Essence VC, 58:31)
- [The Godfather Of MLOps](https://mlopstalks.com/talks/the-godfather-of-mlops) (D. Sculley, Google, 51:25)