Startup Stories - Mixergy

Episode #2278: How to build a $10M/year AI company

Date: September 3, 2025
Host: Andrew Warner
Guest: Pavel Dolezal, CEO and co-founder of Keboola

Episode Overview

This episode dives into how Pavel Dolezal and his team built Keboola—a data automation platform using AI—into a $15 million ARR company. Going beyond the hype, Pavel shares:

• The real challenges of making data accessible in modern businesses
• How startup founders can leverage consulting roots to build software
• The role of community and hackathons as a sales engine
• The journey from bootstrapped beginnings to PLG and major enterprise wins
• Practical AI opportunities in 2025

Warner’s relaxed, energetic inquisitiveness brings out stories of real-world chaos, learning, and dogged determination behind the scenes at an AI scale-up.

Key Discussion Points & Insights

1. Making Data Accessible—and Why It Matters

• The Data Dilemma at Scale ([00:00]):
  Andrew introduces Keboola as a company solving the tough problem of making business data available instantly, not just monthly or quarterly. Pavel illustrates with a client story:

  • The Mechpen Story ([01:03]):
    Mechpen, a stationary retailer, wanted everyone (even 60+ year-old clerks) to access and use sales data to improve business, not just a data team.

    “His vision was that he’s going to teach every clerk, every seller on the floor… Every single person is gonna be a data analyst and is gonna contribute to actually, you know, making business better.”
    — Pavel [01:25]

  • Iterative Rollout ([02:15]):
    At first, giving every employee dashboard access created confusion, not empowerment, so they rolled out access in phases, layering in change management.
    • The company grew fivefold in locations, stayed profitable through COVID and regional instability, all with minimal outside funding.

• The Gold Flakes Metaphor ([05:23]):
  Pavel likens widespread data use to panning gold:

  “If you have a thousand people and they do it every day, it’s more than a big nugget a day.”
  — Pavel [05:39]

2. Founder’s Journey: From Portals to Keboola

• Atlas & Pain of Data Access ([06:09]):

  • Pavel’s experience at Atlas, a major Eastern European web portal, exposed him to the pain of getting actionable user data. Even with engineering resources, it took months to get simple usage reports.
  • “I needed to understand what do people search for... It took me like half a year to get engineers to build this view for me.” — Pavel [08:01]

• Industry Evolution ([08:59]):

  • Rise of cloud and SaaS paradoxically made data harder to access due to proliferation of systems (Gartner: average of 300 SaaS tools in an enterprise), and legacy tech was “freakishly hard.”
  • Founding vision: Data should be accessible to business people, not just engineers—“should not be a part of the voodoo clan that knows how to SQL or Python.”
    — Pavel [11:30]

• Built for the AI Wave ([11:56]):
  Keboola launched as API-first, anticipating the arrival of LLMs to turn natural language to queries—a seven-year wait.

3. Consulting DNA & Product-Led Growth

• Consulting to SaaS ([14:47], [15:17]):

  • Founders cut teeth in consulting, solving data problems hands-on before building productized solutions.
  • Early clients co-designed features—e.g., the Rohlik Group (Europe’s “Amazon Fresh”) helped shape Keboola’s API-first approach.
  • “By that we actually learned what are the hard problems.” — Pavel [15:57]

• Pivot to Product-Led Growth (PLG) after Covid ([23:56]):

  • COVID forced dramatic reevaluation; after pro-bono government work nearly sank the business, Keboola launched a generous freemium model and opened signups (previously all word-of-mouth).

    “We started just word of mouth... you couldn’t actually start Keboola online. You had to talk to someone... So after Corona we’re like, well, this PLG motion... We started like free account. We now have very, very, very generous freemium.”
    — Pavel [24:52]

  • Now, hundreds of companies pay by credit card, tickets range from $20 to $150k.

4. Hacking Customer Acquisition: Community & Events

• Cracking Enterprise via Community ([26:08], [29:04]):
  • Banks, notoriously risk-averse, were won over by running large-scale, value-driven hackathons and community events (no monetary prizes, just fun ones like “Spanish ham”).
  • “So we used our data, we anonymized the data and together we did the hackathon for over 500 people. AWS was sponsoring, IBM was there, Google was there… Three new companies were started out of that hackathon.”
    — Pavel [27:41]
  • Partnered with “women in tech” groups, teaching data to 30,000+ participants; events were an organic sales engine.

    “Our North Star for a good hackathon was not number of new clients.” — Pavel [31:28]

5. Surviving and Scaling Through Crisis

• The COVID Government Episode ([19:11]–[23:14]):

  • In early pandemic days, Pavel rallied 5,000 tech volunteers to help the Czech government with data, almost losing the business due to months of pro bono focus.

    “We almost lost the business because... in times where every tech company was selling, we were focusing on pro bono.”
    — Pavel [21:17]

  • The experience forced a reckoning and led directly to adopting PLG.

• Fundraising Decision:

  • The GPT-3/transformers wave forced a choice: either scale up or get left behind, so Keboola raised a seed round in June 2022 with $3.5–4 million ARR at that point.
  • “It took almost a year internally... Are we gonna lose our freedom? ...Do we want to... but our mission is to automate every single business process with data and AI.”
    — Pavel [33:41]

6. Building AI Automation that Actually Scales

• AI in the Real World—Doing, Not Just Knowing ([12:26], [34:49]):

  • Two-step value: 1) Make data accessible, 2) Enable action/automation.
  • “First you need to understand: what are you solving, what is the process, and only then you can build the agent.”
  • Agents (full workflow automation) only add value after good data and process understanding. Automation follows real questions from business users—"People are very good at asking questions" [38:06]

• Success Story: AI Reducing Human Workload ([42:27]):

  • Client “Jim” (a gym network in 16 countries) used Keboola + ChatGPT to automate review analysis/responses, reducing support staff from 50 to 3, while retaining a “human in the loop.”

7. How to Start an AI Company Today (Even Without Tech Chops)

• Opportunities for Non-Technical Founders ([43:36]):
  • “Vibe coding” (no-code/low-code) is a massive opportunity:

    “Coding at least to MVP... is like now easy. And anybody can go and just like prototype what they want to build and they can sell it.” — Pavel [44:13]

  • The real “secret” is vertical knowledge.
  • Shared examples of friends building $3M–$10M ARR no-code SaaS products in a year.

Notable Quotes & Memorable Moments

• On Data Democratization Gone Wrong:

  “He calls me like 20 minutes later. ‘This is not how you do it! Immediately turn it off!’”
  — Pavel recalling Mechpen’s CEO [02:28]

• On AI and Business Process Automation:

  “Our mission is actually to automate every single business process with data and AI.”
  — Pavel [34:37]

• On Coding and Startups in 2025:

  “Coding—at least to MVP—is like now easy... What is your actual secret is your vertical knowledge.”
  — Pavel [44:03]

• On Crisis and Adaptability:

  “There’s so many crisis building the company that you need to make sure that they don’t kill you. And you need to have at least plan B and C... Go in, you know, for all. I think that’s a great advice—if it works out.”
  — Pavel [23:03]

Timestamps of Important Segments

• [01:03] – Mechpen case study: democratizing data in old-school business
• [05:23] – Gold flakes metaphor for mass data empowerment
• [06:09] – Origins: the pain of data from Portal Atlas
• [15:17] – Consulting roots and product development philosophy
• [19:11]–[23:14] – COVID crisis: government intervention, near-death moment, embracing PLG
• [26:08] – How Keboola won its first major bank via community and hackathons
• [31:28] – Hackathons as community development (and sales)
• [33:26] – Revenue milestone: approaching $15M ARR
• [34:49] – Fundraising: loss of freedom vs. scaling mission
• [42:27] – AI practical case study: reducing customer support staff with LLMs
• [43:36] – Opportunities for non-technical founders in the AI era

Tone & Language

Andrew’s questions are direct, curious, and occasionally playful. Pavel responds with candid, energetic storytelling—full of real-business “war stories,” Eastern European pragmatism, and optimism grounded in hands-on learning.

Final Takeaways

• Solve for real-world complexity, not just shiny tech.
• Deep industry/vertical understanding is a bigger differentiator than coding skill.
• Community and hands-on consulting beats flashy sales in early stages.
• AI is a force multiplier only after good data plumbing is built.
• In 2025, building a $10M startup is less technical and more about knowing your users’ true problems.

For more details, visit Keboola.com and find Andrew’s full archive at Mixergy.