Our Approach to AI

Overview

Artificial intelligence offers genuine opportunities to improve efficiency and even to uncover insight, but only when applied thoughtfully and with appropriate safeguards. Here we outline our approach to using AI for customer insight, but do get in touch if you have any questions.

Here we are talking mostly about LLM-based GenAI tools, such as Claude, Gemini and Copilot, but we have also used traditional machine learning approaches as outlined below.

"We see AI as a useful set of tools rather than a substitute for expertise."

AI principles

  • We will never upload confidential client information or customer personal data to AI services unless appropriate contractual, security and data-processing safeguards are in place

  • We will only incorporate or recommend AI when it demonstrably improves quality, efficiency, or value for our clients

  • We will be transparent about where and how AI has been used in our work

  • AI can support our work, but professional judgement remains essential

Our use of AI

Currently we are using or actively developing our approach to AI in...

Comment coding

We use a combination of human researchers, machine learning, and LLM approaches to comment coding, depending on the needs of the client, and will always discuss this with you openly.

Transcription

We are experimenting with LLMs for transcription of interview recordings, and we're happy to share these with clients as a cost-effective means of providing transcripts that are suitable for many purposes, but not 100% accurate. We are working on ways to improve accuracy through contextual prompting.

Thematic analysis

We are exploring the use of LLMs to support thematic analysis of qualitative transcripts. While these tools can assist with pattern identification and summarisation, we do not currently believe they can replace the interpretive judgement required for robust qualitative analysis.

AI Assist

Our AI Assist tool can be deployed on your portal to make it easier to understand data and find the insights you're looking for. We always add the caveat that AI can still make mistakes.

Our perspective on potential uses of AI

There are other potential applications which we do not yet believe to be beneficial.

Synthetic responses

We actively monitor the debate around the usefulness of synthetic responses. While the evidence is mixed, we do not believe that synthetic responses are effective or cost-effective for most of our work.

In particular, we believe it is a mistake to turn to synthetic responses instead of engaging with fundamental challenges such as low response rates or hard to reach audiences.

Fieldwork

AI, for example chatbot-driven interviews, may have some applications in fieldwork, but we do not yet believe that it is the best option for most situations. While it offers flexibility and the opportunity to ask open questions and probe at scale, it tends to elicit a negative response from customers and can be perceived as rude.

While AI makes it possible to gather and analyse larger volumes of qualitative data, we believe organisations should be careful not to lose sight of the distinct purposes of qualitative and quantitative research. In many cases, deeper insight comes from combining focused qualitative inquiry with robust quantitative measurement, rather than attempting to scale qualitative methods indefinitely.

Research Design & Reporting

We use automation in our reporting workflows for efficiency, but AI does not determine our conclusions. Research design, interpretation, recommendations, and final reporting remain the responsibility of experienced researchers.

In summary

We see AI as a useful set of tools rather than a substitute for expertise. We will continue to evaluate emerging technologies carefully, adopt them where they create genuine value, and remain transparent about how they are used. Our goal is not to use AI for its own sake, but to deliver better insight, better decisions and better outcomes for our clients.