The Challenge
In 2015, Best Buy was losing the personalization war to Amazon. Customers expected tailored experiences — product recommendations that actually made sense, emails that felt relevant, a website that adapted to their interests. Best Buy's existing systems weren't cutting it.
The company went to market looking for a solution. Vendors came back with proposals ranging from $20M to $30M, with timelines of 18 to 24 months. For a retailer fighting for digital relevance, that timeline was a death sentence.
That's when I proposed a different approach: build it internally, from scratch, in 90 days.
The Approach
I knew we could move faster by avoiding the enterprise software trap. Instead of buying a monolithic platform and spending months on integration, I assembled a small team and focused on three things:
1. Start with the Data, Not the Platform
Most personalization projects fail because they start with the tool and then try to force the data into it. We did the opposite — we mapped the customer data landscape first, identified the signals that actually predicted purchase intent, and built the platform around those signals.
2. Open-Source Stack, Cloud Infrastructure
Rather than licensing expensive enterprise software, we built on open-source technologies running on cloud infrastructure. This gave us the flexibility to iterate daily and the cost savings to deliver the entire platform for $3.2M — less than one-sixth of the lowest vendor quote.
3. Ship and Iterate, Don't Perfect and Launch
We put a working version in front of customers within the first month. Every week we measured, learned, and improved. By day 90, we had a production system that was outperforming what the vendors had promised to deliver in 18 months.
Key Insight: The vendors weren't wrong about the complexity of the problem — they were wrong about the approach. Enterprise personalization doesn't require enterprise software. It requires understanding your data and moving fast.
The Results
The numbers speak for themselves, but the real story is what happened after the initial launch:
Year 1: $120M in attributable revenue. Conversion rates on personalized experiences jumped from 1% to 17%. The platform was processing more data than any other system at Best Buy — combined.
Years 2-3: Revenue exceeded $1B as the platform expanded across channels — email, web, mobile app, and in-store. The team grew from a small skunkworks project to a core part of Best Buy's competitive strategy.
Cost savings: 85% less than the cheapest vendor proposal. And we delivered in 90 days instead of 18-24 months.
Frameworks Applied
This project is where the Velocity Gap Framework was born. The organizational friction we had to overcome — the permission loops, the planning inversion, the polish paralysis — became the foundation for understanding why AI projects stall everywhere.
The resource allocation decisions we made also informed the B1/B2/B3 Innovation Framework, specifically the distinction between table-stakes features (B1) and competitive-edge capabilities (B2) that let us prioritize ruthlessly.
What This Means for You
If you're looking at vendor proposals for AI capabilities and the numbers feel wrong — they probably are. The build-vs-buy calculation has shifted dramatically in favor of building, especially with today's tooling. The question isn't whether you can afford to build it yourself. It's whether you can afford to wait 18 months for someone else to do it.
Want to apply these frameworks to your company?
Book a discovery call to discuss your specific challenges, or subscribe for weekly AI strategy insights.
