3 / S C
Over the last few years, I have been observing a set of familiar user behaviors: people struggle to articulate what they want before encountering it. Trust has always been earned incrementally. Understanding has always required participation rather than passive consumption.
These aren’t interface problems. HCI and Cooper’s About Face have done the foundational work of understanding how people interact with systems and how to design those systems around user goals. Hooked, by Nir Eyal, maps how products build habits. BJ Fogg’s Behavioural Models identifies the triggers behind people’s actions. Jobs To Be Done organizes intent and outcomes. Indi Young’s Mental Models maps the reasoning. This library tackles the next logical question: how systems could be designed to meet people at the level of their intent, behavior, and journey; rather than asking them to adjust to the system.
While designing Felvin early in 2022 and later that year building an early conversational agent prototype for Flipkart Healthplus, it became increasingly clear that the conversational interface was merely a stepping stone. The field is wide open to design how intelligent systems should behave in response to a person’s needs, intent, context, and mental model.
The assumptions that current systems operate from do not match with how people actually decide things. Or how they learn. Or move from a vague position to a clearly formed one. When we want to go out, we are presented with grids of options arranged in endless scrolls. When we want to learn, systems give us answers directly instead of probing our critical thinking. These behaviors are not new, but we now have systems that can observe, calibrate, and respond to them in ways that were previously difficult to achieve.
Over time, I started extracting these patterns from real product work and documenting them through essays and prototypes. The result is a library of Human-System Relationship Models.
Each model starts with a recurring human behavior, proposes a system behavior in response, explains the mechanism behind it, identifies where it works and where it breaks, and rounds off with applications across domains. But naming a behavior isn’t enough to call it a model. A model has to earn that status: it needs real provenance, a mechanism specific enough that it couldn’t describe an adjacent model, failure conditions named without hedging, and applications stress-tested across enough domains to confirm the principle is genuinely transferable.
That’s why the library is organised into three tiers: stub, partial, and developed. A stub is an formed idea. A partial has found provenance and a stated problem but has not been fully stress-tested. A model only reaches developed when the mechanism holds up under scrutiny and the applications confirm it travels beyond its origin context.
For example, Decision Compression, that originated as Orion framework, is one of six models in the developed state.
Decision Compression
The human behaviour: most people don’t know exactly what they want until they see it. Asking them to describe it first produces a best guess, not a real preference, and systems built around that guess stall the decision instead of driving it.
The mechanism flips the usual order and instead of leading with a question, the system leads with a proposal. A small set of options functions as a diagnostic. Each reaction narrows what comes next without ever asking the person to explain why. Across a few rounds, the right answer becomes obvious.
The same model applies to event discovery, restaurant booking, travel planning, fashion, gift selection, content recommendation, experiential retail, and personal styling. That transferability is the actual and final test for whether something belongs in the library or gets pruned. So when a healthcare leader reads a model that originated in e-Commerce commerce and recognise the same human problem in their own domain, that is the true success of the model.
Six models are fully developed. The Socratic Emergence leads itself to Constructed Insight, while its off-shoot, the Essay Companion leads to Cognitive Scaffolding. Each model identifies the models where it shares axis. Most models have a working prototype that you can request.
Several others are in progress. New ones keep emerging as new situations reveal the same or entirely new recurring patterns.
