Cost of contact

Research methods sort by how much contact they need with real people. The 2011 deck called these tiers no-touch, low-touch and high-touch. The point was to make research cheap enough to actually happen. With agents doing part of the work, the same tiers decide who does what.

The tiers

Tier	Methods (2011 list)	With agents
No-touch	Web and video analytics, social sharing, heat-mapping, mouse-tracking, heuristic reviews	Agents run it, unattended, on a schedule
Low-touch	Remote usability, online card sorting, A/B and multivariate tests, surveys, engagement analytics, social listening	Agents draft and run; a person reviews in one pass
High-touch	Participatory design, contextual inquiry, focus groups, moderated usability, stakeholder interviews, hallway testing	A person does it. This is where judgment and discernment live

A field note: what the sensors missed

In September 2026, a pet parent trying out PayPaw needed more than 40 minutes, with the founder helping, to almost pay a $100 invoice. The invite link broke inside Yahoo Mail's in-app browser, the wrong email went out, and there was no way to copy the link. None of the automated audits caught it. A real person on a real mail client did.

The fix became a nightly check that walks a cold start through a real inbox. High-touch contact found the problem; a no-touch sensor now guards it. That is the whole idea in one incident.

In the harness (working proposal): a routing rule. Cost of contact decides what an agent runs on its own, what it drafts for review, and what it hands to a person.

Open questions

• Checks can prove a finding is grounded in evidence. They can't prove it matters. Where does that judgment get recorded?