An AI Brewery Spokesperson: Efficiency, Authenticity and Trust

The psychology and its limits

Dietvorst, Simmons and Massey’s 2015 paper studied algorithmic and human forecasting. Across five studies, seeing an algorithm make errors reduced willingness to choose it, even when it outperformed a human forecaster. The research concerns forecasting judgments, not synthetic brewery presenters or sales. [2]

Logg, Minson and Moore’s 2019 research found algorithm appreciation in several estimation and advice tasks: participants sometimes gave more weight to advice they believed came from an algorithm. The tasks and comparisons differ from the earlier work. [3]

Together, these papers caution against a simple “people hate AI” rule. A reaction can depend on the task, expectations and comparison. Criticism of a fictional spokesperson cannot, by itself, diagnose algorithm aversion or establish why a customer would buy.

A practical six-step workflow

These are Maqtrix’s proposed applications to test, not outcomes established by the brand story or a promise of results.

1. Define the promise before the format

   Write what customers value about the business: convenience, expertise, personal attention, local craft or something else. Then ask what a presenter needs to communicate. A production tool should support that promise rather than become the entire strategy.

2. Separate useful automation from representation

   List where automation could help without changing who appears to speak for the business. Editing, drafting and organizing assets are different from inventing a person who seems to work there. Evaluate each use on its own role and consequences.

3. Keep factual claims accountable

   Assign a real person to review product statements, availability and any implied experience. A synthetic character should not be used to invent a testimonial, staff biography or first-person customer experience. Make the business responsible for what its content says.

4. Test with relevant people

   Show the proposed message to people similar to intended customers. Ask what they think the business offers, who they believe is speaking and what feels unclear. Avoid leading questions such as asking whether they hate AI; those can frame the response before you learn anything.

5. Compare a meaningful alternative

   Try a staff explanation, product demonstration or simple voiceover alongside the synthetic format. Keep the core message similar so the comparison is useful. A poorly made alternative would tell you little about whether the format itself matters.

6. Plan how to respond

   Decide who will answer questions and how misleading impressions will be corrected. Listen for concerns about authenticity, accuracy or service separately from general dislike of the style. Revise the content when it communicates something the business cannot support.

A small-business example

A hypothetical local bakery wants to explain a new ordering service. It can use automation to organize a script while a staff member demonstrates how an order is prepared. It could also test a stylized animated explanation, provided the format does not pretend to be a real employee’s personal experience.

The useful comparison is whether customers understand ordering, trust the information and know what to do next. The business does not need to prove that one technology is universally good or bad. It needs a format that communicates its actual service accurately.

What to measure

Track comprehension, relevant enquiries, recurring trust questions and completed actions over a defined period. Separate views from customers and negative comments from verified cancellations. Include production and review time in any efficiency calculation. Without a suitable comparison, describe observed changes rather than claiming the creative caused them.