Product-to-product insight derived from your own transaction data. Explainable in plain language, on your own infrastructure.
Know what your customers really think.
What this is
Your transaction data contains a detailed map of how your customers actually relate products to each other: what they buy together, what they treat as substitutes, what they discover in sequence. bemorehuman is a recommendation engine I built and have refined since 1997 that extracts that map directly from purchase behaviour and presents it in a form your own people can verify at a glance.
No black box. No customer data leaving your environment. No ML team required to run it.
Why it’s different
Explainable by construction Every relationship the engine surfaces traces back to real customer behaviour. When it says two products belong together, your category managers can see why and judge whether it's true.
Works at your scale and below it Credible results from thousands of transactions, not just millions. That means it works per-store, per-category, or per-banner, not only on a whole business at once.
Runs behind your firewall Deploys on ordinary hardware, on-premise, or in your own cloud. Customer data never leaves your control. This is a clean answer to the data governance questions that stall most AI initiatives.
Open source core The engine is public on GitHub under the MIT licence. Your team can inspect exactly what will be running on your data. View the engine on GitHub →]
Where it applies
The same behavioural signal serves several teams:
Customer experience Genuinely personal recommendations online and in-app, driven by what customers do rather than what a rules engine assumes.
Buying, planning and supply chain Behaviour-based product relationships show which products earn their place in a range, what customers treat as substitutes, and what sells together. A supply chain planned around products that actually sell means less overstock, fewer markdowns, and less waste.
Marketing Cross-sell and campaign targeting grounded in observed behaviour, explainable to any stakeholder who asks "why this product?"
The offer
A fixed-price proof of concept
A contained engagement, typically 2–3 weeks:
1. Data audit and preparation I take a sample transaction export and structure it for the engine. This step alone usually surfaces data quality findings worth having.
2. Similarity run The engine derives product relationships from your customers' actual behaviour.
3. Validation session I walk your team through what it found. Your experts judge whether the relationships are real. That's the success test: not my claims, their judgement.
Fixed price, defined scope, your data stays with you. If the results are meaningful, we talk about where to apply them; if not, you've spent little and learned something about your data either way.
About
I've worked in recommender systems since 1997, when I built the first version of this engine. I've since been CEO of Silverstripe (the NZ open source CMS company) and held CTO roles. A previous paid proof of concept with a major NZ grocery retailer produced product relationships their merchandising team validated on sight. I'm Wellington-based and work with NZ retailers directly, in person or remote.
Contact
Curious what your data knows?
The conversation costs nothing and takes half an hour.
Brian Calhoun brian@bemorehuman.nz