Ali Ghodsi on Scaling as a CEO

Ali Ghodsi has been CEO of Databricks since 2016, taking over during the company's 2015 near-death year, when revenue was $1.6M. By 2015, Databricks' open source success had become Databricks’ biggest commercial enemy. Amazon, Google, and on-prem vendors were offering cheap Spark-as-a-service, and customers refused to pay them a cent. A decade later, Databricks has ~10,000 employees and is one of the most valuable private companies in the world.

What follows is the framework Ali built over the next decade to turn it around.

1. The STEER Framework: The Five Ingredients of a Great CEO

Your job as CEO is to steer the company: pick a direction, and get people to follow you there. STEER decomposes that job into five ingredients: Spark energy, Targeted strategy, Exceptional team, Execution, and cultuRe. 

Spark Energy (S): Make Every Team Believe They Have the Most Important Job

• Founders are intrinsically motivated because they have the most skin in the game. It is unfair and unrealistic to expect employees to match that baseline without deliberate effort. As the company grows, people have other things to do and will not inherit your vision by osmosis.
• The core move is to give every team a purpose that is bigger than hitting a metric. The framing should make people feel they are working on the single most consequential problem in the company.
• When a team is unhappy, do not blame the culture. Look at the manager. Bad morale correlates almost perfectly with a specific manager who is weak at this ingredient - they might be smart, hardworking, even high-performing, but they are not giving their team a reason to care. Find the manager before diagnosing anything else.

Example: Reframing the Company Keynote Around Regeneron

• For years, Ali's company keynotes were execution-focused, with the message being: here are the goals, here are the metrics, go hit them. 
• At ~200 people, employee excitement cratered, with the feedback being that employees weren’t inspired and didn’t understand the bigger picture impact their work was having
• The next year, Ali rebuilt the keynote around Lucas, a data scientist at Regeneron who used Databricks to identify the genome responsible for chronic liver disease, leading to a drug in Phase 1 trials. That year, employees loved the keynote. A group of them flew to Boston unannounced to meet Lucas at his office.
  
  • Find the specific human stories of your product's impact and make them the centerpiece of how you communicate the company's purpose.

Example: Turning Around a Burned-Out Customer Support Team

• Databricks' early support team was under-water. The open source software was built for PhDs and full of bugs, customers yelled at them constantly, and leadership was just cracking the whip for more tickets closed per day. Scores were rock-bottom.
• A new leader came in and reframed the job entirely: "You have the most important job in the company. Lucas at Regeneron is impossible without you - the engineers are shipping buggy code and the sellers are selling things that don't fully work. We are the ones who make all of it real." 
  
  • Then he laid out a two-year plan: knowledge base for year one, and root-cause fixes for year two so the team could see a path out.
• Scores went up and they became one of the happiest teams at the company. While the job did not change, the meaning behind it did.

Sales Teams Need Purpose Too 

• Nothing insults a great salesperson more than being told they are "just here to hit quota and follow the comp plan." 
  
  • Money matters, but the best sellers want to be treated as frontline soldiers making the product real inside customer accounts - getting past politics, budget blockers, and competitive noise.
• The pitch Databricks used for its sellers was: "Without you, the AI/data revolution literally cannot reach our customers. You are not just selling - you are having a huge impact, and you are getting paid for it."
  
  • Show them the purpose first. Sales people will still work hard for the comp.

Targeted Strategy (T): Apply Your Strength to the Enemy's Weakness

• Vision is cheap; anyone can paint an inspiring future. Strategy is where it gets hard, because you actually have to be right.
• In the next 12 months, get ruthlessly specific: What is your single biggest bottleneck? What is your strongest asset? Who is the one competitor blocking you, and what is their most exploitable weakness? Concentrate all force there.
• The mental model is David vs. Goliath. You do not out-wrestle the giant. You pick the opposite strategy and aim the slingshot at a specific weak spot. 
  
  • (Recommended reading: Good Strategy Bad Strategy by Richard Rumelt.)

Example: Databricks vs. Amazon

• In 2015, AWS was the invincible 800-lb gorilla. At every AWS Re:Invent, a dozen startups died. AWS was offering cheap, commoditized Spark, and Databricks' open source success had become its biggest commercial threat.
• Ali studied Amazon obsessively and concluded their strategy had a real flaw. They had lifted the retail flywheel (everything store, economy of scale, lowest prices) and applied it to software. But in software, customers do not want 200 services to stitch together with IAM roles, DevOps scripts, and configuration hell. They want one thing that works.
• Databricks took the total opposite strategy: one unified analytics platform. More expensive up front, but no DevOps army needed. Their pitch was: "You can technically build this yourself on AWS with 15 services - or you can use us." This became Databricks' defining product position for years.

The Biggest Founder Mistake Is Having No Strategy At All

• The most common mistake is not a bad strategy, it is no strategy. Founders (and frontier AI labs) look at what competitors launched last week and chase their tail. By the time the roadmap matches, the competitor has moved on.
• This is the thesis of Clay Christensen's The Innovator's Solution, Chapter 3: Markets become commoditized when every PM benchmarks against every other PM's feature list.  Eventually, every product looks identical (for example, smartphones).
• Real strategy requires a thesis about a secular weakness your competitor cannot fix. OpenAI spent six months on Sora, blocking meaningful work they could’ve done with partners, then canceled it. 
  
  • Every great company makes strategic mistakes like this

Exceptional Team (E): Raise the Bar

• Optimize for false negatives, never false positives. Hire as if every mis-hire is catastrophic. 
  
  • Databricks has passed on many great people who went on to do phenomenal things elsewhere, and that is fine. The inverse mistake (bringing in someone mediocre) is far more expensive.
• Codify what this means at the company level, and separately at each function level.
  
  • At the company level, Databricks' filters include:
    
    • No job-hoppers. A candidate with a pattern of two-year stints gets auto-rejected by recruiting. Brilliant people sometimes hop; Databricks does not want them - they will spend a year ramping and then leave just as they start producing.
    • Everyone has to be exceptional at something. Harvard and number-one-in-class is fine. So is "didn't go to college because they worked three jobs to support their family and self-taught to build an important open source project." Mediocre is the disqualifier.
    • First-principles thinking and cognitive aptitude are tested for, not assumed.

Make Candidates Do the Actual Work

• The best predictor of job performance is watching someone do the job.
  
  • Interviews - especially brainteasers and "what is your biggest weakness" - have almost no signal.
• Instead, put real work in front of the candidate
  
  • Example: if someone is interviewing for the Databricks Ventures team, walk them through three live deal decisions and ask how they would evaluate them. Good candidates dig in, ask to meet the founders, and form a sharp view. By the time Databricks makes an offer, they have effectively pre-validated the fit.
• In geographies where it is legal (e.g., parts of Europe), hire people on a six-month trial before converting. In the US, approximate this with substantive work samples over the course of the process.

 Trust Backchannels Over Interviews

• Some people are incredible interviewers and mediocre operators. Some are the reverse. Do not pretend you can size someone up in an hour; you can't. 
• A backchannel with someone who worked with the candidate for five years is worth more than twenty interviews. They know the candidate's actual operating style, blind spots, and behaviors under stress.
  
  • One of Databricks' highest-ranked engineers over ten years into the company was a hard no in his interview. Ali was ready to pass. Then Databricks backchanneled with his current employer and the feedback from them was phenomenal. Databricks gave him a large take-home, and the code came back pristine at 3am
  • He has been there ever since - he works extreme hours and is one of the most productive engineers at the company. He is just bad at interviews. The only reason he got hired is that Databricks did not trust the interview signal alone.

Execution (E): Discipline Beats Genius

• Execution is the single most important ingredient but it is also the one most founders systematically underinvest in, because it does not feel intellectual
  
  • Execution is like getting fit. On paper, it’s simple: you eat well, work out, stop drinking, show up every day. The problem is doing it consistently for years.

Inspect Multiple Levels Down

• Setting metrics, establishing a cadence, and holding people accountable is table stakes. The more overlooked lever is inspection: going two, three, and four levels below your direct reports to verify that the work actually happened and that the quality is real.
  
  •  “The team made 15 sales calls this week” is not enough on its own. You need to inspect the work itself. 
    
    • For example, review a sample of calls to see whether they actually happened and whether they were good.
• You cannot run a company on reported metrics alone. Inspection is what turns activity reporting into real accountability.
  
  • Trust but verify. It may feel heavy-handed, but without inspection, standards slip.

Pick a Functional Playbook and Run It

• Each function already has battle-tested execution frameworks. Pick one per function and stick with it.
  
  • It matters less which framework you choose than that you pick one and run it with discipline.
    
    • When Ali took over engineering at Databricks, there was no process. Engineers shipped whatever was trending on Twitter, features that did not work hit customers, and frameworks changed weekly. 
      
      • Installing deadlines, two-week roadmaps, and explicit "what we are committing to ship" transformed velocity and quality.
• Engineering: Scrum. Two-week sprints, retrospectives, breakdowns of work into smaller deliverables. Databricks uses this.
• Sales: MEDDIC or MEDDPICC. 
  
• Product: Jobs to Be Done is Ali's preferred framework. 
  
  • JTBD is weaker pre-PMF but excellent once you have initial traction and want to tighten the fit.

Modulate Tone to the Team's State

• If the team is struggling or demoralized, be extra positive. If the team is crushing it and getting complacent, call it out and make people feel the standard.
  
  • The mistake is running one tone in all weather. Great execution cultures flex the volume based on the team's current state, not the leader's default personality.

Culture (R): Your Culture Should Match Who You Actually Are

• Most founders write down an aspirational culture that does not match their actual personality or operating style. This always fails. Soul-search what you actually are and let the culture reflect that.
  
  • Many cultures produce world-class outcomes, and repel certain personality types. Neither is "correct." What matters is that you choose one consciously and consistently. For example: 
    
    • Amazon: process-heavy. OP1, WBRs, 6-pagers, frugality rules. Every important move is scripted and inspected. You cannot freelance, and only if a good move is approved can you execute it. The upside is predictability and scale; the downside is slowness.
    • Netflix: the opposite — full autonomy, keys-to-the-kingdom, fear-based accountability. Do whatever you want, but any visible mistake and you are fired (and the firing is broadcast to the company). The upside is speed and empowered brilliance; the downside is a chronic fear culture

Priority Order: Promote > Hire > Fire > Write It Down

• The single most effective cultural lever is promotions. Promote people who embody the culture, and the rest of the company learns what the culture is by watching who wins.
• If you want to do more:
  
  • Promote according to the culture (highest ROI).
  • Hire according to the culture (next-highest).
  • Fire against the culture (third).
  • Write it down - handbooks, wall posters, values pages (lowest ROI, and the thing most companies start with).

2. Which Ingredient Matters Most

The five ingredients are not equally important. Ali's ranking, in order:

• 1. Execution. The single most important ingredient. Even if your strategy is wrong and your team is average, ruthless execution lets you redirect faster than anyone else and still win.
• 2. Targeted Strategy. Executing fast in the right direction. Most founders have no real strategy, so doing the work here is a major edge.
• 3. Exceptional Team. This matters - but less than founders usually believe. A great team with poor execution and no strategy loses. A decent team with both wins.
• 4. Sparking Energy / Vision. Important for follower-ship but easier to fake than strategy or execution.
• 5. Culture. Operates over long time horizons and is the least load-bearing in the short term. Many successful companies have explicitly "bad" cultures.

3. Identify the One Bottleneck And Weaponize the Whole Company

Ali manages his time by constantly trying to clear his calendar so that at any point in time he’s free to focus on the single biggest bottleneck in the company.

One Bottleneck at a Time

• At any given moment, there is one thing blocking your company more than anything else. Identify it and put the full weight of the company behind it. Weaponize every employee toward removing it. 
  
  • Timeframes vary - this can last three to six months or sometimes a year or two.
• Examples from Databricks' history: fighting Snowflake, cracking engineering hiring against Google comp (2017–2018), landing the Microsoft partnership (2016).
  
  • Different year, different bottleneck, but in each case, the whole company pointed at one thing.
• The recursive version of STEER: (1) Targeted Strategy tells you what the #1 bottleneck is, (2) Spark gets the company excited about attacking it, (3) Exceptional Team gives you the people to do it, (4) Execution grinds it out.

Revenue Is a Lagging Indicator — Don't Manage To It

• Investors will ask about revenue growth because they are pattern-matching. Your head of sales will ask about revenue because it is their job. None of that means you, as CEO, should manage the company on revenue.
• Databricks' data warehousing product did $5–6M in its first year against a Snowflake doing roughly $1B. It took four years to cross $1B on that product. If Ali had managed to revenue, Databricks would never have entered data warehousing. 
  
  • Revenue comes two to three years after the right bottleneck-clearing work.
• Focus on input metrics and bottleneck removal. Revenue will show up in the lag.

4. Using One Goliath to Fight Another: The Microsoft Partnership

The clearest illustration of the "one bottleneck, weaponize the company" playbook was the 2016–2017 Microsoft partnership, which became the single thing that put Databricks on the map. 

The Strategic Rationale Behind the Partnership

• Databricks was one of a thousand undifferentiated data startups - ClearStory, Trifacta, H2O, DataStax. AWS was the 800-lb gorilla squeezing the category. The bottleneck was not product but being unable to rise above the noise.
• Microsoft hated AWS. Databricks did not have the weight to fight Amazon alone. Using one Goliath to fight the other (Microsoft's distribution, brand, and enterprise footprint) was the only path to visibility.
• Ali told his EA that any inbound from Microsoft was P0. The entire company was weaponized around the deal.
  
  • Approximately 50% of Ali's time for the year went into it.

Expect the Gauntlet

• Microsoft PMs were initially dismissive even though Databricks beat them in technical deep-dives until the senior PM was converted.
• Microsoft engineering’s 60 tech leads plus the world's top database professor assembled in a single room to tear Databricks apart. Ali personally fielded their questions, leaning on his co-founders' PhDs in databases. It was a tough meeting but Databricks walked out with technical credibility.
• Microsoft marketing’s CVP emerged late and tried to kill the deal outright. Ali took multiple trips to Redmond to win her over.
• The brain behind Microsoft's bundling strategy, pushed back hard. Ali closed the pricing terms with him at 4am after a back-and-forth following his flight from Japan.
• Once signed, Microsoft assigned their most prolific engineer to lead the project. He refused to ship under the Microsoft brand and manufactured security concerns to stall. Databricks worked relentlessly through Satya's office to get him moved aside. A replacement was installed, and a four-month delivery sprint began.
• The project was ultimately delivered one week early.

The Outcome From the Partnership

• Microsoft launched Databricks in 100+ languages simultaneously, including web pages in Japanese and across African markets. A 100-person startup went from invisible to globally branded overnight.
• Amazon saw the Microsoft partnership and came back offering their own deal. Databricks now had two giants competing for the same partnership slot. The dynamic flipped entirely from "one Goliath might stomp us" to "we can play these two off each other for better deals, better discounts, better co-marketing, and better sales motion."
  
  • The lesson is not "do a Microsoft deal." The lesson is: identify your one bottleneck, weaponize the entire company against it for as long as it takes, and look for leverage that turns a 1v1 fight into a three-way dynamic you can exploit.

‍