top of page
AI philosophy & FAQs
At Aldavin, we approach AI with curiosity, but also with the same critical scrutiny we would apply to any research method. We recognise its potential, while remaining realistic about where the evidence is strong, where it is still developing, and where human expertise remains essential.
We believe AI is most powerful when combined with human expertise, judgement and understanding.
AI is evolving rapidly. We keep a close eye on new developments and emerging evidence, continually reassessing what the technology can and cannot do well in research. The perspectives set out here reflect our view today, not a fixed position. As the technology advances, some of the opportunities and limitations we see today may change with it.
We use AI selectively across the business, including to automate routine administrative tasks and streamline day-to-day operations. Agentic workflows help us reduce the time spent on low-value manual work and free up more time for what we do best: thinking deeply about our clients’ questions, solving sticky problems and applying our expertise where it matters most.Within research projects, we may use AI for bounded tasks such as supporting desk research, developing code frames for open-ended data, and as an additional point of triangulation to challenge our thinking and identify potential blind spots.There are also clear boundaries. Methodological decisions remain with the researcher and AI does not write questionnaires or replace human analysis. Our research remains senior-led throughout.
Yes, we are always transparent about how AI is used in client work and discuss any relevant applications upfront. Client data is never shared with or processed by an AI system without specific prior approval.We also include a clear AI disclosure in our deliverables, for example where AI-generated or AI-modified imagery has been used.
Our default in all research is to use human respondents. This is particularly important in studies focused on animal health, as specialist target populations are still difficult to simulate reliably. Veterinary professionals are less likely to be well represented in the data used to train general-purpose AI models. Their responses also draw on clinical knowledge, professional experience and a behavioural context that those models may struggle to reproduce. We therefore believe synthetic respondents in these type of populations require particular caution.That does not mean we dismiss synthetic research. Its value depends on the research objective, technical implementation and, where human data or other empirical evidence are used to ground or calibrate synthetic respondents, the quality and relevance of those inputs. Where clients want to explore synthetic data, we can partner with specialist third-party providers to develop a solution suited to your needs.We can also provide consultancy to organisations developing synthetic respondents or personas within their own AI environments. Working alongside their internal AI and technology teams, our role is to bring the research perspective: helping ensure that what is technically possible is also methodologically sound.
Every synthetic study needs to be validated on its own merits. Evidence that a synthetic panel or platform has performed well against human data does not automatically validate every study conducted using it. Results may differ when the audience or subject matter changes.For client projects, we agree the appropriate validation approach upfront. The most rigorous validation is to run a parallel study with real participants and compare the findings directly. Where that is not practical, relevant existing research can provide a useful benchmark. Without an empirical benchmark, we can still assess the quality and consistency of the synthetic results, but we would not describe the study as validated.We are also running our own controlled experiments, comparing synthetic responses with human research data. This allows us to build our own evidence about how well the method performs and better understand its limitations.
Yes, we can offer AI-moderated research through selected partners.AI moderation can offer practical advantages in cost and speed, for example in the exploratory phases of mixed-methods projects. We do, however, take a more measured view of the concept of ‘qualitative research at scale’, as more interviews do not necessarily mean more insight. There is strong evidence of diminishing returns setting in fairly quickly in well-executed qualitative research, with additional interviews often reinforcing rather than adding to findings.As things stand today, we also do not believe AI can fully replicate the skills of an experienced human moderator. Qualitative interviewing is about more than people's words. A skilled moderator will also look for subtle signals such as slight hesitations, changes in tone and body language, using these to assess when to probe further and in what direction.For us, the question is not whether AI or human moderation is better, but which approach is better suited to the research objective.
We never process client data using AI without the client’s specific prior approval. All data is anonymised beforehand to protect personally identifiable information (PII).In cases where the client has approved data processing, we use business-grade services with appropriate data protection and security safeguards in place. These include encryption of data in transit and at rest, controlled access, and protection against client data being used to train the provider’s models.
No. We do not use client data to train AI models.
bottom of page