Three IPOs Later, Frank Slootman Still Thinks the Job Is Urgency

Frank Slootman is the three-time CEO of Data Domain, ServiceNow, and Snowflake, the kind of operator other operators look up to for relentless execution. Sarah Guo and Elad Gil, hosting No Priors, bring him on to talk about Snowflake's horizon and the enterprise AI opportunity, then start where his story starts: he grew up in Holland, the first person in his family to go to college, and later immigrated to the United States. As a kid he was focused and disciplined, with a chip on his shoulder he generally likes in other people. You need a reason to get up in the morning. The U.S., he says, was a much better canvas. There is no doubt he would not have done where he came from what he has done here.

The through-line is not a secret formula so much as a temperament he will spend the hour naming: choose the circumstances you cannot change, then relentlessly drive the ones you can. Life can look like a random walk, a series of events that go from one to the other. In hindsight he has to go back and find the method in the madness. The method, once he finds it, is unfashionably simple. Intensity is not a mood. It is the job.

The résumé the hosts recite is already a greatest-hits reel, and he still has to correct it. They have him taking Data Domain from less than $3 million through an IPO and a $2 billion EMC acquisition; ServiceNow from $75 million through an IPO to about $1.4 or $1.5 billion; then Snowflake, one of the seminal companies in the data world and one of the largest software IPOs of its moment. He will come back to Data Domain in a minute. The question underneath is what is common across those jobs, and what he still has to prove after already having accomplished so much by the time he got to Snowflake.

Guo notes that he talks about being on the right elevator, and that some of the companies he joined were not the hottest at the time. He uses the image because there are aspects of opportunity and circumstance you cannot change. You will be subject to them, for better or worse, so you need to choose carefully. Some people think they can will their way to anything. That is not true. Where you are going to be, what industry, what company, what people — those choices are formative. Combine good choices with great execution and you get the cocktail for future opportunities. It matters a whole lot.

A couple of times a year he speaks to graduating classes at prominent business schools. They always want one message. His is: do not go work for a consulting firm out of school. Get a real job in the real economy, building real products, selling real products. You need to feel what it is like to be in the drive stream of the economy instead of eating out of somebody else's trough, sitting on the bezel, gliding along, feeling good about yourself without having touched a real economy yet. He wants people early in their careers to feel the heat of competition and the cold wind when markets disappear. A lot of people choose jobs very removed from the real world. He does not think that is helpful for development.

Gil's usual advice is choose the right industry, then the best company in it; the role is secondary. Slootman totally agrees. The role is not that important. You will have many. Roles come and go. His first job was one he really did not want. Being an immigrant, beggars could not be choosers. He figured he would get in and make his way from there. He landed in a corporate planning group of about six people attached to the CEO of a large computer company, as far removed from the real world as he could be. These were the heydays of affirmative action; they did not have a lot of picks. He spent two years doing typical MBA work — M&A, board presentations — and after that they pretty much gave him whatever he wanted. From there he made his way.

He first corrects the record on Data Domain. They had no revenue, no customers, nothing. There were 15 people. When they first started to sort the product it had one terabyte of usable space and ran around 30 megabytes a second, useless for 99.9 percent of applications. Why take the job? He did not know that. He had been rejected numerous times for CEO opportunities; the ones interested in him were second and third string. People cautioned him not to hold out. You need really good investors. They were a startup, one of hundreds. He would walk the halls of NEA and Greylock and people would look at him: who are you, what company is that? They were a no-name, lectured about other companies that in hindsight ended up being no-names. He lives for that kind of drama. They did not have product-market fit. They just did not.

A little fit showed up in a CIO meeting. The company later got acquired by EMC. They were testing the product. The guy said their little box was a real hero on Friday. The email database had been backed up on the Data Domain device. There was a massive corruption — common then, not now — at four o'clock on a Friday afternoon. The old plan was recover from tape all weekend, sleeping on cots. Then they remembered they had a backup on disk. By seven that evening they were going home. You do not need to be a rocket scientist to figure out another use case you can sell a few times more. They stayed alive. They did do that $3 million the first year. He still remembers the first contract: a $5,000 service deal with Stanford University, and they bitched and complained the whole way. He thought, this is going to be a great business.

Gil calls *Tape Sucks* a hidden gem for go-to-market and sales, dense with channel strategy and partnership tactics other business books skip. Slootman self-published it, homebrew, few pages, no filler. He gets an awful lot of inbound — coffee, speeches, consulting — and if he took it all it would become a full-time job, so he writes. The last book, *Amp It Up*, he did not want to write. Denise Pearson, Snowflake's CMO, pushed him and made it easier. He wrote every word; a ghostwriter asked for examples so people outside the business would understand. Americans, he says, always think there is a formula you can extract and, if you just get your hands on it, do it too. The book is his answer to what explains the three successes across different times, markets, technologies, and competitors. He is not trying to sell it. He does not care whether you agree. He did kill customer success at every company he has been in. He thinks it is the biggest thing going on in Silicon Valley. That does not mean he needs you to agree. He is just telling you what it is.

Why do so many CEOs and leaders not push for more urgency or higher standards? He asks if they have been to a California DMV. That is what a lack of urgency looks like. It is what naturally happens to human beings. It is innate. We slow down to a glacial pace unless there are people who will drive tempo and pace and intensity and urgency. That is what leaders need to do, because people naturally slow down, minds wandering to the next vacation or the weekend. You need high focus, high intensity, high preoccupation with what we are doing. People ask for the message of the book. Read the title. There is an X-factor, an enormous amount of room in the margin, right under your nose. You can take it up in the next meeting, the next podcast, the next email, the next Slack message: push the urgency, the standards, the alignment. Easy message. Hard to have the mental energy to bring it to every single instance of today.

He has seen young CEOs who think they hire a bunch of people and sit back and wait for greatness. They have no idea they have to relentlessly drive every second of the day, every interaction, and seek the confrontation. CEO jobs are insanely confrontational, which is not human nature. We avoid it. He had a founder-CEO who, every time somebody had to get fired, had the CFO do it and stayed home that day because it was just so hard. He did not have the disposition for it. Slootman understands. There are still people in the enterprise who have to do that stuff.

The fear, Gil says, is that if you push hard enough people will leave, and you will lose in the talent marketplace. Slootman's answer: if they leave they should leave. Culture sorts and sifts. You attract the right ones and start losing the wrong ones. It is actually quite perfect. They are just not your DNA, not your blood type, and you need to create your blood type around you or you will have nothing but conflict. He remembers people after two weeks who said they could not take the pace and intensity. It was not him personally; everybody was like that, calling people out, driving expectations they were not used to. They wanted to go home at 4 p.m. and pick up the kids from school. He told them they needed to go back to HP and sleep in a cubicle. This was not the place.

Culture can be incredibly helpful, he adds, but it is not a general thing. There is no such thing as general goodness. A culture needs to enable your mission. Whatever enables the mission effectively is a good culture. There is no universal culture that is good. It depends on the type of leadership, the type of business, where you are on the journey. If you do not fill the void, somebody else is going to.

Asked for a Snowflake 101, he says the founders would argue immediately that they were never a warehousing play. They were dealing with semi-structured data from the get-go; the workloads were more than batch analytics on purely structured data. The founders were two Frenchmen, longtime Oracle technologists and architects, responsible for taking Oracle from a departmental platform to the enterprise platform it became. Parallel SQL came from them. They left and wanted to reimagine database management for cloud computing, carrying as little old technology forward as they could — not taking a Postgres kernel and hacking it up for the cloud, a move he is being unflattering about because plenty of people have done it.

They did breakthrough things. Most notably, the separation of storage and compute. Back in the day you bought them in combination; you could not buy one without the other. In the cloud you can consume them independently. The commercial model evolved into consumption, eventually billed by the compute-second, after a world of nodes and machine-hours. The other move was taking the control plane out of the cluster. Clusters are now all stateless, clueless, which is great because you can run tons of them concurrently. There is no master inside the cluster; the master lives outside. In the old data-warehouse world you had to beg for a 2:30 a.m. slot three months out because the cluster was consumed quickly. Now there is no limit. He often tells investors he is not creating demand. He is enabling it. It is so pent up it is insane. You can provision for economy — cheapest possible — or for blistering performance, and make those choices. They opened up demand in the legacy marketplace and started migrating massive Teradata plants, old Hadoop and Cloudera, tons of Oracle and SQL Server. Still early innings. Those platforms are not easy to move.

When he started, the tagline was the data warehouse built for the cloud. He did not want to stick with it. You taint yourself with a brush and pretty soon you cannot get it off, which is pretty much what happened. He has an allergic reaction every time he hears data warehousing. To him it is just a type of workload now, no longer a market or an industry. Cloud data-management platforms, and Snowflake, are seeking to become full-spectrum workload capable: from the most batch-analytical to streaming, online transactional, massive scale, extremely low latency. The premise behind the data cloud is that the work comes to the data; the data does not go to the work. Historically data has always been pumped around to the work, and you get massive siloing whether you try or not. A new app comes with its own database. The siloing prevents you from exploiting what lies in the data. He has started telling large banks: do not go re-siloing your world in the cloud. Your data-science, ML, and AI teams will be very frustrated trying to overlay, blend, fine-tune, and train. The data universe is an ecosystem bigger than the enterprise — FactSet, Bloomberg, S&P, hundreds of hedge-fund data flows. That is also why they are multi-cloud. You cannot have a data cloud on a single public-cloud platform, by definition.

The other change is that this is no longer a self-contained database with ODBC and JDBC for apps sitting outside. That breaches the governance perimeter. The application needs to execute inside the platform, not outside. They have a programmability layer called Snowpark, a native application framework, a different stack than the on-prem world he grew up with. Streamlit, the Python visualization project they acquired, is the same logic: Python programmers reflexively reach for it to make a machine-learning model consumable by a business user who can manipulate variables and watch the visualization redraw. Bring it inside Snowflake as Streamlit in Snowflake because you need a hardcore, trusted, sanctioned governance perimeter. You can bring Python to your data in two weeks; people download libraries every couple of weeks and have no idea what exfiltration risk they are exposed to. Snowflake spent two years making Python non-porous. Large financial institutions will not let Python anywhere near core data. Snowflake's answer is: do it inside the boundary, without creating exposures. In a world of AI, if you do not have highly organized, optimized, sanctioned, trusted data, what do you want your models to do — train on a data lake? He calls it a landfill. You have no idea what is in there. Everybody dumps their stuff. A data product is data taken out of that, turned into a trusted, optimized, understood object you can give to the business and stand behind. That is the chief data officer's job.

Gil asks how the ChatGPT and GPT-4 wave, then a few months old, is showing up in the enterprise, given that a large business can take six months just to plan. Slootman's starting point is almost too plain: large language models are about language. It is a huge deal. He was taught the basics of COBOL in school — Common Business Oriented Language — and there was nothing common or business-oriented about it, extremely cryptic, but compared with assembler and machine code the syntax was amazingly comprehensible. In the 1980s SQL was also positioned as something mere mortals could use to query data. People's relationship with data has evolved and still been immensely frustrating: ad hoc, standardized reporting, dashboards. Going to natural language is the last mile. The effect on demand will be enormous, because every mortal, if you are semi-literate, maybe not even literate, can just talk and get value from data.

He draws a hard line around generation. The generative aspect is very cool when you are planning a trip to Yellowstone. In the enterprise you are dealing with structured proprietary data, and you are not planning trips. An insurer might ask: we had disproportionate bodily-injury claims in Florida; surrounding states did not; what explains that, will we have it next quarter, do we stop underwriting, do we change pricing? You are not going to get the answer to that from a large language model. Separate text-to-SQL, which he thinks is incredibly valuable, from going after structural proprietary data. That is a different realm. You will need business models, not just language models, because that question requires you to understand underwriting and pricing. In auto insurance, Geico, Progressive, Liberty Mutual and the rest, telematics data is number one through ten: the device in the car knows when you are speeding. That is how they now price risk, lowering prices and increasing profits. Extremely productive data. The difference between winners and losers. And that is not even AI. That is machine learning. Analysts might take weeks, or you bring in McKinsey or Bain. Systems should start giving insight into those questions. That is where Snowflake lives: proprietary structured enterprise data, coupled with language. Pretty powerful.

They acquired Neeva, a search company founded by former Google people, for that expertise. He is a search junkie. Search is addicting because it lets you explore everything known, written, published, opinionated. The problem is it has no context. It matches on strings. Search on Snowflake and you might get the company, the weather, or the social phenomenon. Enrichment and context are the name of the game in data. One attribute can take a data field from mundane to high-octane because of the context it creates: suddenly insightful, impactful, predictive. Search that becomes stateful, plus chat, becomes one natural language conversation. Combine that with new levels of intelligence specific to industries and there is a world of opportunity still waiting. The Snowflake interface, he thinks, will change over five or ten years toward more natural language and a broader user set. BI companies — Tableau, Looker, the dashboard world — still have a future, because dashboards also tell people how you want them to look at the data. Not all analysis is ad hoc. For ad hoc, nothing will be better than natural language. Internally they already push Salesforce data into Snowhouse, their Snowflake database. He asks who the two top reps are in this country or that market; it spits the answer in a fraction of a second with a graph. Addicting, like search. You just keep going. BI will be severely affected.

Nine out of ten customer conversations are not technology, architecture, and migrations. They are industry use cases. Healthcare is economically not viable at the scale we need to deliver it; data can make it predictive and prescriptive — who is at risk for what disease, when, and what they need to do. Data does not have opinions. It gives you accuracy. Institutions do not want to treat disease; they want to prevent and anticipate it. In pharma it takes on average 12 years to bring a drug to market, with five years of patent left; compress that by one, two, or three years and you change the economics of the industry. Supply chain is a third: they just announced that Blue Yonder will fully re-platform on Snowflake. Supply-chain management has never been platformed because the data problem was unsolvable; it is still an email-and-spreadsheet business, living in a Microsoft world from 30 years ago. Visibility across the entities of a chain needs a single data universe. The R&D implication is brutal. The platform has to be incredibly broad and deep. If it does not do what people need, they pump the data somewhere else and you are back to fragmenting. Cybersecurity partners run on the data cloud because customers would not buy yet another database container. Bring the data here, combine it with vulnerability sources, and analysts search one universe instead of 15.

Does the macro shift — from buying software like crazy to cutting SaaS budgets — change his view of consumption or credit-based pricing? Not really. Investors love it when you have customers over a barrel and can keep a gun to their head. He does not particularly like that. At ServiceNow he always felt the relationship was not equitable: customers signed up for many millions of dollars, took nine months to get into production, and paid for all the users the whole time. What he liked about Snowflake, cloud computing, and consumption is that you pay for what you use. A utility model. Is it painful sometimes? Yes. He talked to the CIO of a bank in Las Vegas: the bank is growing 3 percent, Snowflake is growing 22, and that cannot go on forever. The CFO gets in and says this is the size of your bread box, live with it. People go back to the drawing board. The granularity is the point. You can downgrade a provision, run a workload less frequently, change a retention period, and lower consumption of storage and compute in near real time. In a SaaS subscription they wait for the next renewal before they start cutting off a limb. Investors do not like it. They love it on the way up. They hate it on the way down.

How should operators manage through the downturn? He sees the Amazon, Meta, Google layoffs. Snowflake does not do layoffs, because they do not wait until there is a huge headwind. They are always pruning the tree, so they do not have to do it as some massive, super-unsettling event. Management of resources should happen daily: not just performance, but bringing supply and demand in sync, alignment, constantly. Culture evolved to a place where it is unfathomable to be so confrontational as to take somebody out of a job, so people look the other way until a crisis, then rip out tens of thousands. He does not think that is fair or effective. His world does not change all that much, because he was already doing it. People are not used to living in downturns. When you have been around longer, they come around. It is part of life. Double down, put your game face on, put your boots on. This is actually going to be a lot of fun. This is where it really happens. People grew up thinking trees grow into the heavens. Trees do not grow into the heavens. Everyone needs to grow up a little and get a leash on reality.

For startups, raising money is a whole different ball game. They need to think harder. At Data Domain they ran the company from one fundraising milestone to another. That is how it was. In recent years people have not had to raise money or run businesses that way, to prepare themselves for a fundraising milestone. They should, because that is how you stay alive. Fundraising is oxygen for a company. Gravity turned back on, Gil says, and everybody is realizing it. Slootman closes on the book title. In the next meeting, the next message, the next encounter, amp it up. It is a choice. Do not be afraid people will react poorly. They will not. The good people will actually love it. Inject energy and focus and intensity and quality so the whole place starts to feel exciting, not like you are watching the clock for 4 or 5. It is much easier to live in an energized environment than one that is devoid of energy.