Techstars '17: The Bet on Intelligent Commerce

In the spring of 2017, Hanzo was accepted into Techstars. We had been operating since 2008. We had paying enterprise customers. We had shipped Hanzo.js, Track.js, Dash.js, and a growing suite of AI infrastructure tools — recommendations, copy generation, behavioral analytics.

We applied to Techstars not for survival but for signal. At that stage of the company, the hardest thing wasn't building — it was proving to the enterprise sales teams we needed to reach that Hanzo was a real, durable company worth betting on. Techstars was a forcing function for that.

The 2017 Thesis

The pitch was simple in retrospect: commerce would bifurcate into platforms that used AI as a native capability and platforms that bolted it on afterward. The first group would compound — every transaction improving the models, every model improvement driving more transactions. The second group would fall behind on a curve that widened every quarter.

We had been building toward the first group since 2008. The recommendation engine. The copy generation system. The behavioral analytics stack. The A/B infrastructure. Every piece was a component of a commerce platform that learned.

Techstars's interest wasn't surprising to us. What we were describing — AI as the operating layer for commerce, not an add-on — was becoming obvious to anyone paying attention. The question was whether we could build it fast enough.

The Program

Thirteen weeks in Denver. Intensive mentorship from operators who had scaled businesses we respected. The Techstars network is genuinely different — not just investors but founders who had been through the same compounding loops we were navigating.

The most valuable conversations were with mentors who had scaled enterprise software. The question they pushed on hardest: What does the data moat actually look like? If you're building a commerce AI platform, your advantage should be the proprietary training data you accumulate at scale. The models you train on your customers' collective purchase histories should be better than anything an individual brand could build alone.

That framing sharpened how we thought about the Hanzo Datastore — not just as infrastructure but as the foundation of a competitive moat that widened with every customer we added.

What We Were Building

By 2017, the AI stack had evolved significantly from the 2015 early versions:

Recommendations had moved from batch ALS to online learning — the model updating in real-time as transaction data came in, not just in nightly batches. The result: newly launched products got accurate recommendations within hours, not days.

Copy generation had expanded from email subject lines to product descriptions, ad headlines, and landing page variants. The training corpus had grown to tens of thousands of campaigns with attached performance data. The quality gap between AI-generated and human-written copy had narrowed to the point where the A/B tests were often tighter than the standard errors.

Attribution had gone multi-touch. Track.js had always captured the behavioral sequence; we'd added the multi-touch model to correctly weight each touchpoint in the path to conversion. This solved a long-standing problem for our agency clients: understanding which marketing channels were actually driving revenue versus which were claiming credit.

What the Accelerator Changed

The investor relationships from Techstars led to the Series A that funded the next phase. But the lasting impact was the operator network — specifically, the introductions to enterprise procurement teams that compressed what would have been a two-year enterprise sales cycle into months.

When you're selling AI infrastructure to enterprises, you need someone in the room who can say "I know these people, their technology is real, their team ships." Techstars was that voucher.

---

Hanzo is a Techstars '17 company. The intelligent commerce thesis we pitched in 2017 is the same thesis underlying everything we've built since.