Skip to main content

Trust centre

Sooner or later someone will ask how much they should believe a Merlin result — a client, a board, a colleague who has commissioned traditional research for twenty years. This section is written for that moment. It sets out what Merlin actually did, the evidence behind it, where its limits are, and how to describe it without overclaiming.

It is also for you. Knowing where a result is strong and where it is thin is what turns a number into a decision.

The trust case in three lines

  1. Grounded in research about a defined audience

    Every answer starts from real survey data about real people, not from a prompt.

  2. Many modelled respondents, not one AI opinion

    Each persona answers separately, so a result is a distribution with a spread — not a single plausible-sounding view.

  3. Tested against answers Merlin has not seen

    Real answers are withheld, Merlin predicts them, and the two distributions are scored against each other — currently 92%, against a 94% ceiling set by real people answering the same question twice.

Why Electric Twin built Merlin

Research often arrives after the decision has moved on. By the time a survey has been fielded and written up, the campaign is booked, the roadmap is set, and the findings become a post-hoc explanation rather than an input.

Merlin makes an existing audience model available while the decision is still open — for the questions and the iterations that would otherwise be settled by assumption. That is the gain. The trade is that a modelled answer is evidence about a model of an audience, not testimony from the audience itself, which is what the rest of this section is about.

Where to go next