Built a $50M+ AI Support Empire by Owning the Edges
with David Karandish, Capacity
Built a $50M+ AI Support Empire by Owning the Edges
Show Notes
In December 2016, the top-selling product on Amazon was not a toy, a book, or a video game. It was Amazon's Alexa. For most people, that was a holiday novelty. For David Karandish, it was a starting gun.
David had just finished one of the most successful runs in the history of vertical search. He built Announced Media, acquired answers.com, merged the two companies, scaled them into a powerhouse, and sold the whole thing for $900 million. He took five months off. He made his laundry list of what to do next - somewhere around 50 ideas. Forty were terrible. A handful were decent. One made him feel like he would regret skipping it for the rest of his life.
That one became Capacity.
He founded it in early 2017, before the world understood what AI was about to become. Blockchain was the darling. AR and VR were getting the hype cycles. AI was maybe fourth on the list of things people were excited about. David bet on fourth place, and he was right. Today, Capacity serves 20,000 customers - including T-Mobile, Verizon, Nike, and American Express - with annual revenue surpassing $50 million. The strategy behind all of it is something David calls the Compound Startup.
Frameworks from This Episode
These frameworks have been added to the AI for Founders Frameworks Library. Filter by David Karandish to find them.
The Compound Startup Strategy
Instead of building one deep point solution, map the entire customer journey end to end and own the integration layer connecting every node.
- •Identify every touchpoint a customer has with your business, from first inbound question to post-interaction loyalty loop.
- •Acquire, build, or partner to cover all steps of that journey.
- •Lower the cost of individual SKUs but raise the take rate of the bundle.
- •The integration layer becomes the product - not the individual features inside it.
- •Value lives on the edges connecting the nodes, not on the nodes themselves.
The Salad vs. Brownie Problem Framework
Before committing to any AI project, identify whether a partial solution still delivers value - or whether it must be complete to work at all.
- •Salad problem: remove one ingredient and the salad is still good. You still get 80–90% of the value.
- •Brownie problem: miss one ingredient and the entire batch is ruined. You need a complete solution or you get zero value.
- •AI-powered customer support is a salad problem - even automating 60% of tickets is a win.
- •Ask before any AI build: if this only works 70% of the time, does it still create value or become a liability?
The Data Flywheel Loop
Build compounding intelligence directly into your product architecture so every interaction makes the system smarter.
- •Virtual agents handle level-zero support, freeing human agents for higher-value work.
- •When a question exceeds the AI's confidence threshold, it escalates to the right human with the right expertise.
- •That human interaction feeds back into the virtual agent as new training signal.
- •The loop closes, and the system gets smarter with every single interaction.
- •This is not just automation - it is compounding intelligence built into the architecture itself.
The Ideal Calendar Framework
Run your life by a fixed schedule, not a reactive one - assign shape to your week before the week begins.
- •Assign specific nights and mornings to deep work so the calendar has structure before any requests land.
- •Protect one non-negotiable personal anchor per week - David's is a weekly date night with his wife.
- •Create rotating one-on-one rituals with the people who matter most.
- •The goal is not work-life balance. It is rhythm. Rhythm lets you run harder and longer without burning out.
Founder Experiment: Build Your Own Support Flywheel in 48 Hours
Build a lightweight internal support bot that logs every question it cannot answer - so you can identify your highest-value automation targets before spending a dollar on enterprise software. No developer required.
- 1Open your AI coding tool of choice (Claude, Cursor, Replit). Prompt it: "Build me a simple Python chatbot using a JSON knowledge base. When a user asks a question the bot cannot answer, log the question with a timestamp to a CSV file called unanswered.csv. Give me a basic web interface using Flask so my team can use it in a browser."
- 2Populate the JSON knowledge base with your 20 most common customer or team questions and their answers. Your support inbox, Slack history, and FAQ page are gold mines for this.
- 3Run the bot for five business days. Have your team use it instead of emailing you or each other for those common questions.
- 4Open unanswered.csv. Sort by frequency. The questions that appear most often are your highest-value automation targets and your roadmap for what to build or buy next.
Why this works: You are replicating the core insight behind the Capacity flywheel at zero cost. You are not guessing at what to automate - you are letting real usage tell you exactly where the leverage is. That is the problem behind the problem, surfaced in 48 hours.
Key Terms
These terms have been added to the AI for Founders Glossary. Search by David Karandish to filter them.
Links & Websites from This Episode
Capacity
David Karandish's AI-powered customer support platform serving 20,000 customers including T-Mobile, Nike, Verizon, and American Express. The platform automates level-zero support, routes escalations, and closes the data flywheel loop.
Create A Loop
A nonprofit co-founded by David Karandish that teaches computer science and AI to underserved kids. 90% of participants are on the school lunch program. The organization has served approximately 1,700 students and is now expanding to train teachers.
Q&A
What is the Compound Startup strategy and how does it apply to AI companies?
The Compound Startup strategy, popularized by Rippling CEO Parker Conrad and executed in the CX space by David Karandish at Capacity, involves bundling multiple point solutions into one integrated platform. Instead of competing as a single-feature product, you map an entire workflow end to end, cover it with owned or partnered solutions, and make the integration layer itself the core product. For AI companies, the connective tissue between features is often more defensible than any single feature on its own.
How did Capacity grow to 20,000 customers?
Capacity grew through a combination of direct outbound sales (roughly 60%) and partner-led growth (roughly 40%). The core engine was the Compound Startup strategy: covering the full customer experience journey, acquiring companies in key areas, partnering for non-core components like payments via Stripe, and aggressively cross-selling and upselling across the customer base. The unified platform eliminated the integration pain that competitors could not solve.
What is the difference between a platform and a product suite?
In a product suite, value lives inside each individual product. The tools may share a login or a brand, but they are loosely connected and data does not flow freely. In a platform, value lives on the edges connecting the products. Data flows across the system, actions in one product trigger responses in another, and the whole becomes significantly more valuable than the sum of its parts.
How accurate is AI customer support in 2025 and 2026?
According to David Karandish, Capacity's AI now answers between 92 and 94 percent of questions thrown at it, with less than 1 percent receiving negative feedback. In 2017 when Capacity launched, the same system answered about 55 percent of questions with a 10 percent thumbs-down rate. The improvement is a result of better underlying models, more training data, and a closed feedback loop between human agents and the virtual agent system.
What is the Salad vs. Brownie framework for evaluating AI projects?
This framework helps founders decide whether a problem is a good candidate for early or partial AI deployment. A salad problem still delivers most of its value even if one ingredient is missing - meaning a partial AI solution still creates real ROI. A brownie problem fails completely if one ingredient is wrong - you need a complete solution before any value is realized. Founders should identify which type of problem they are solving before committing to an AI build.
Who is David Karandish?
David Karandish is a serial entrepreneur and the CEO and co-founder of Capacity.com, an AI-powered customer support platform serving 20,000 customers including T-Mobile, Nike, Verizon, and American Express. He previously co-founded Announced Media, acquired answers.com, merged and scaled the combined company, and exited for $900 million. He is also a board member and co-founder of Create A Loop, a nonprofit teaching computer science and AI to underserved youth. He is based in St. Louis, Missouri.
What is the AI for Founders podcast?
AI for Founders, hosted by Ryan Estes, is a podcast and newsletter for founders navigating the intersection of artificial intelligence, venture capital, and startup growth. Each episode features operators, investors, and builders sharing frameworks, strategies, and real-world experience.