// recruiter? 60-second version analytics architect · Amsterdam

// Current Meditations / working notes

What I have been thinking about lately.

Working notes on open trade-offs in methodology, AI-assisted workflows, schema design, and the gaps the official measurement layer does not settle. I think of them as koan-like: compact contradictions worth carrying around until the problem looks different.

Methodology / Mixed methods

What does a complete qualitative and quantitative data feedback loop look like, from instrumentation to optimisation?

Instrumentation can flag a change in behaviour; qualitative method can explain what it means. Experimentation and optimisation can test whether the response actually improved the system. The missing piece is a practical methodology that treats those as one loop rather than separate research, analytics, and CRO rituals.

Working theory: the loop is: instrument the behaviour, detect the anomaly, investigate the meaning, redesign the interaction, test the intervention, and feed the result back into the schema. Quantitative data narrows the search, and the qualitative work supplies the next question. Optimisation closes the loop by proving whether the answer held.

AI-assisted workflows / Human agency

What do humans cede when AI becomes the gatekeeper to action?

Before, the human prompted the AI: analyse this, summarise this, check this. Now the AI increasingly prompts the human: react to this, approve this, escalate this, ignore this. That reverses the agency problem. The question is no longer only whether AI can assist judgement, but whether it begins to own the queue of human attention.

Working theory: the risk is not only that AI makes decisions for us. It is that AI starts deciding what deserves our attention. Volition survives when humans maintain the gates: suppression, escalation, evidence, interruption, and feedback. But if every gate requires approval, the system collapses back into the bottleneck it was built to solve. Human-in-the-loop everywhere recreates that bottleneck; the workable design keeps humans at a small number of gates with strong control surfaces.

Measurement design / Session continuity

What replaces sessions when users stop and start their visits across hours and days instead of minutes?

The session was useful when a visit looked like a bounded sitting: arrive, browse, convert or leave. Increasingly, the journey looks more like a thread picked up across tabs, devices, messages, notifications, saved carts, and delayed decisions. What unit of measurement survives when the user's actual visit no longer fits inside the analytics platform's clock?

Working theory: the session measured attention inside a single sitting and says nothing about intent left unfinished across days. Instead of stretching the session timeout, the better move is to model a task sequence across a longer conversion window: start, pause, resume, complete, or abandon. Sessions remain useful locally, but the user journey is measured by the open loop, not the sitting.

Schema design / Event taxonomy

What deserves to be an event, and what should only be a parameter?

Older analytics habits encouraged tracking across explicit steps: the user saw this, clicked this, skipped this, continued here. Newer event models make a different question more important: did a distinct action occur, or is this just context for the action that did occur? A skipped step may not need to become its own event if the next event can carry the prompt, offer, path, or choice set that shaped it.

Working theory: event minimalism only works when parameter richness is strong enough to preserve meaning. The event should name the user action; the parameter should explain the conditions that made the action meaningful. A skipped step becomes an event only when the skip itself is an intentional action. Otherwise, it belongs as context on the event that followed.

Practitioner observation / Insight visibility

When does tool adoption become a new data silo?

Personal dashboards can look like analytics adoption: more users are building, exploring, and answering their own questions. But if the insight stays on one person's dashboard, device, or team workspace, the company has not become more data-driven. It has just moved the silo closer to the user.

Working theory: adoption is not the same thing as visibility. A shared company data layer should let insights travel across teams instead of getting stuck in one analyst's personal dashboard. The success metric is not how many people made charts; it is whether useful findings become reusable, discoverable, and governed enough to change decisions outside the team that found them.