Domino
Data Lab

An infrastructure company in a costume jewelry industry

In a downtown office tower somewhere, a model risk officer is signing off on an underwriting algorithm. She does not know it, but the audit trail in front of her - every dataset version, every parameter, every approval - is rendered by a piece of software called Domino. The model will go live on Monday. It will run for years. It will be re-validated quarterly. Nobody outside the bank will write about it.

That is Domino Data Lab's actual business: keeping enterprise AI shippable, traceable, and out of the newspaper. The chatbots get the press. Domino gets the renewal.

Today the company sells an enterprise AI and MLOps platform that helps large, regulated organizations build, deploy, monitor, and govern machine learning models across whatever combination of clouds, on-prem hardware, and GPU clusters they have lying around. The customer list reads like a defensive industrial directory - Allstate, Bristol Myers Squibb, Bayer, Fannie Mae, USAA, BNP Paribas, Cigna, Dell, Lockheed Martin, Lloyds Banking Group. More than 20 percent of the Fortune 100. Roughly 1 million models running through the platform at last public count.

Data science was a craft. The Fortune 500 needed an assembly line.

In 2013 the phrase "data scientist" was being called the sexiest job of the 21st century, which was nice for the data scientists and inconvenient for everyone who had to manage them. Notebooks lived on laptops. Models lived in emails. Regulators lived in a state of mild alarm.

If you were Citibank or Pfizer and you wanted to deploy a model that decided who got a loan or which compound to advance to a trial, the path from a PhD's Jupyter notebook to a production endpoint involved a heroic amount of duct tape and a non-trivial number of arguments. There was no Git for models. No version control for training data. No way to prove, three years after the fact, why a particular score had been produced for a particular customer on a particular Tuesday.

This was the gap Nick Elprin, Chris Yang, and Matthew Granade kept staring at. All three had spent years inside Bridgewater Associates, which is roughly what you would design if you wanted to invent a worst-case for data infrastructure: enormous datasets, life-or-death model accuracy, and a regulator on speed dial. They had built internal tools to make Bridgewater's research reproducible. They suspected the rest of the Fortune 500 needed the same thing.

Three quants walk out of a hedge fund

Nick Elprin became CEO and stayed there. Chris Yang became CTO and stayed there too. Matthew Granade made a different bet - he went back into asset management and kept a board seat - which has the lovely property of giving Domino a permanent customer-side conscience in the room.

The founding bet had two parts. First, that the right unit of work in data science was not the script or the notebook but the project - a self-contained, versioned bundle of data, code, environment, and results that any colleague (or auditor) could reproduce on demand. Second, that enterprises would pay real money for a platform that treated MLOps as boring infrastructure rather than as the latest novelty.

Investors agreed, eventually. Zetta and Bloomberg Beta showed up early. Sequoia led the Series D in 2018. Coatue, Highland, and Great Hill Partners filled in the rounds. By the 2021 Series F the cheque was $100 million, and by August 2025 UBS had joined an extension - a signal that the European regulated-banking crowd had decided Domino was not optional.

Boring on purpose, in the best possible way

If you asked a Domino salesperson to describe the platform in one breath, they would probably say: an end-to-end enterprise AI platform covering the full lifecycle from data exploration to model deployment, monitoring, and governance, across hybrid and multi-cloud environments. That is a lot of hyphens. The shorter version is that Domino is the place a data scientist opens in the morning and the place a compliance team opens at audit time.

The pieces that matter:

The Domino Enterprise AI Platform. A workbench, a project store, an environment manager, and a deployment engine, glued together. Every experiment is reproducible. Every artifact has a lineage. Every model has an owner.

Domino Nexus. The multi-cloud orchestrator. One job, one workbench, computed wherever the data is allowed to live - which, in pharma and banking, is rarely where you would prefer it to live.

Model Monitoring. Drift detection at production scale, with automatic retraining triggers and dashboards that an oncall data scientist can actually read at 2am.

Governance. Audit trails, policy enforcement, model validation workflows, and the paperwork no founder wants to talk about and no regulated enterprise can ship without.

Numbers, customers, and the absence of drama

Enterprise software lives and dies by a single metric, which is whether the customers renew. Domino's do. The platform sits inside more than 100 large enterprises, including a fifth of the Fortune 100, in industries where switching costs are measured in regulatory filings rather than weekends.

The customer roster does most of the storytelling. Allstate writes more auto insurance policies than almost anyone in America and uses Domino to put models behind those decisions. Bristol Myers Squibb runs drug-discovery workflows inside it. Bayer and BNP Paribas and USAA and Lockheed Martin sit on the same logo wall - companies whose risk appetite is, to be polite, not high.

None of these names are accidents. They are the customers who pick a platform on a five-year horizon and then refuse to switch, because switching means rewriting the validation paperwork for every model already in production. Domino has built a moat out of audit trails.

Make every company model-driven. Preferably before lunch.

The mission Domino talks about externally is some version of "help enterprises accelerate research, increase collaboration, and deliver high-impact AI models at scale." The mission you can infer from the product is narrower and probably more honest: make AI inside large companies look less like a science project and more like a supply chain. Reproducible. Audited. Monitored. Boring.

There is a quiet ideology in this. If you believe AI is going to make consequential decisions about credit, medicine, employment, and national defense - and many of Domino's customers run models in all four - then the interesting question stops being "can we build the model" and becomes "can we trust it tomorrow, can we retrain it next quarter, can we defend it in front of a regulator in three years." Domino's bet is that the second question is the one with the budget.

Generative AI changed the conversation. It did not change the audit.

The interesting part of the next five years isn't whether large language models will be useful inside the enterprise - that part is mostly settled. The interesting part is who is going to govern them. Who keeps the lineage. Who proves a model didn't drift. Who handles the moment when a hospital's triage tool starts behaving oddly at 3am.

Domino has spent twelve years building the unglamorous answer to those questions, and it has spent the last two extending the same platform to cover generative AI specifically. The bet is that an enterprise that learned to govern a credit-scoring model in 2018 will need the same scaffolding for a clinical-summarization LLM in 2026 - just with more parameters and a more nervous legal team.

Back in the downtown office tower, the model risk officer hits approve. The bank ships a new underwriting model on Monday. The customer never thinks about it. Neither does the regulator, which is the whole point. Somewhere in San Francisco, an engineering team is already shipping the next release. The dominoes keep falling, very quietly, in the right direction.