AI Marketers Guild: our first public Neuralift demo
Jonathan Mendez joins the AI Marketers Guild for the first public demo of Neuralift — 2.4M-row retail dataset, 29 segments, under two hours.
A while back, Jonathan Mendez joined David Berkowitz and the AI Marketers Guild community for what was, at the time, the first public walkthrough of the Neuralift platform — covering why we built it, what it does to first-party customer data, and a live demo on a real 2.4M-row retail dataset.
▶ AI Marketers Guild · ~55 min · Jonathan Mendez on segmentation, scale, and time-to-value
The session opens with the origin story — Jonathan was Chief Digital Officer at a large travel brand coming out of Covid, sitting on real-time GA4 in BigQuery, a Snowflake instance of transactional data, and a CDP doing lookalike modelling. And still he couldn’t easily answer the basic question: who actually are our customers, and what’s relevant to them? That’s the problem Neuralift was built to solve.
From there he moves into the substance: why traditional SQL-based segmentation is structurally limited, where AI changes the calculus, and what a deep-learning approach actually surfaces on real data.
Conversion rate has been at 3% for about the last 15 years — because this is the way we’ve been doing segmentation.
The demo itself runs from around the 17-minute mark. Jonathan uploads a women’s-fashion retailer’s customer file (2.4M rows, no PII, behavioural + transactional + loyalty signals) and the platform returns 29 named segments in under two hours — each with summary, insights, persona, and channel-specific recommendations. The Q&A afterwards is worth the time on its own: skeptical questions on hallucination, stability across runs, reliability vs. unsupervised ML methods, and how AI-derived segments feed downstream next-best-action models.
A few moments worth jumping to:
- The thesis on bias in SQL segmentation (~10:40) — why hand-picking the top 20 columns from a 300-column dataset hard-codes analyst bias into the output.
- Live demo of the segment landscape (~17:00) — the circle-packed view of all 29 segments, sortable by KPI lens (loyalty, at-risk, new customers).
- Drilling into a single segment (~22:00) — summary, insights, persona, and recommendations for “Weekend Fashionistas / Evening Elegance.”
- The skeptical questions (~26:00) — Adam, Chris, Eric and others on reliability, hallucination control, and what’s third-party-data vs. first-party-derived.
Two years on, much of what’s in this video has matured significantly — the platform now runs on NVIDIA Blackwell GPUs (see our Q1 scaling note), the ML evaluation layer ships with full Weights & Biases instrumentation, and we’ve added structured outputs for activation across email, SMS, social, and search. But the core thesis is unchanged: stop minimising your customer data to fit a SQL rule, and let a neural network find the cohorts your team can’t see.
Thanks to David Berkowitz and Mitch for hosting, and to the AI Marketers Guild community for the questions that made this hour what it was.