Lesson 10 of 12 · 2 min read

Who sets the rules when research uses digital twins?

The researchers, once, before the pressure arrives. A twin built from past interviews and data can answer what people already said. Only people can answer something new, and someone has to draw that line before a deadline draws it for them.

Diagram: past interviews with known answers on one side of an evidence boundary, and sessions observing people today, with one new behavior, on the other.
The past answers what people said. Anything new takes a person, observed today.

The pressure to use a twin comes from good people

Vivek Srivastava, writing on LinkedIn, argues that twins are going mainstream, so the question is who decides when to use one. He names three pushes, and none of them is wrong.

Three reasons to reach for a twin, and the check none of them makes
  1. The deadlinethe product managerNeeds an answer by Friday
  2. The demothe vendorShows a headline accuracy number
  3. The budgetfinanceSees a cheaper line item
  4. The checkthe researcherWhat was it built on, and can the past answer this?

A stand-in gives the answer that sounds right

Nielsen Norman Group, a UX research firm, tested synthetic users against real people. Asked about an online course, a synthetic participant said it had completed every part. The real participant had completed three of seven.

Those were made-up participants, not twins of real people, so for twins it’s a warning, not proof.

A stand-in can’t report behavior. It has none.

What a twin can fairly answer, and what still needs a person

A twin, at best

  • What people told us before
  • A first pass at a screen’s wording
  • Which questions to bring to real people

Only a person

  • What they actually do now
  • Where they hesitate, and why
  • Anything the past never covered

Draw the line once, so it isn’t argued every week

Srivastava’s point is that a research policy, written once, ends the debate on every project. I’d write it like my testing rule: agents clear broken paths; people judge results.

The agents catch what’s broken on every build. They have never once told me whether a design was right.

A twin can tell you what people said. Only people can tell you what’s true now.

Check yourself

Which research question on your roadmap could the past actually answer?

Those are fair game for a twin. The rest belong with real people: anything new, and anything about behavior.

Is your team’s line written down, or decided under each deadline?

A line drawn under pressure moves toward the cheaper answer. Write it while nobody needs it.

The idea: Vivek Srivastava, on why researchers must draw the line on digital twins. The example: how I use agents in testing, from lesson 2.