It’s not hard to see why research teams are interested in the possibilities of artificial intelligence (AI). Its ability to process, analyse and extract feedback from huge amounts of data can dramatically reduce the time and cost of developing actionable customer insights.

One 2023 study found that while only 20% of research professionals currently use AI in at least some of their work, an additional 38% plan to incorporate it eventually.

However, there’s little guidance on what to do or how to use AI in research. While the concept of AI has been around for decades, genuine practical applications are much newer. It took the arrival of generative AI to bring it into the mainstream, but that’s created a situation where many businesses feel they should be using AI without necessarily understanding both the opportunities and limitations.

Using AI effectively in research
There’s a tendency when something new and shiny comes along to think we need to use it everywhere possible. However, deploying it with targeted use cases might have more benefits than trying to add AI to everything right away.

For instance, could a large language model (LLM) be trained to produce research questions, or at least the first draft of them? This would free up researchers without necessarily exposing customer data to the model. This could be hugely valuable for smaller teams that don’t have the resources to properly prepare customer data for the ethical and secure use of AI but still want to unlock some of the productivity benefits.

Could you be using AI to extract more value from your research? Many research teams are using AI to help them uncover themes across their customer interviews, research reports, and conversations. AI is great for identifying key topics and relevant themes, and for helping researchers to curate them faster than they can manually. When it comes to customer research, accelerating the curation of customer knowledge is not just helping researchers – it’s also unlocking value and insights for product designers, product managers, and sales and marketing people.

Are you asking enough of your AI? Many new AI tools for researchers feature search leveraging natural language processing so that researchers can ask plain English questions of their customer data. With the right prompts, these AI tools are capable of returning highly relevant and insightful results. So, ask your AI your most complex questions – you may be pleasantly pleased with the results and be able to make sense of your vast existing customer data.

How to mitigate risk when using AI for research
First, everyone needs to be aware of the potential challenges surrounding using AI in research. It’s like any successful technology deployment: there needs to be training to get the most out of the investment. When it comes to AI, this needs to extend to users knowing the potential side effects, such as hallucinations and in-built bias.

One good example of the risks of AI is when it comes to data. AI without data is nothing, yet researchers wanting to deploy AI to analyse mountains of customer information must be careful that their use doesn’t contravene privacy regulations. For instance, are you certain that the LLM you’ve just prompted to summarise the trends in a document full of customer interactions won’t store sensitive information somewhere insecure?

Then there’s the issue of bias. AI models are trained on available data sets. How do you know those data sets don’t contain racial, gender or age-related biases? Another study found that when asked to produce recommendation letters, LLMs such as ChatGPT would use language such as ‘expert’ and ‘integrity’ for men and ‘delight’ or ‘beauty’ for women. In addition, marginalised populations are already underrepresented in research and data, which also contributes to further amplifying biases when AI is used.

A proactive approach to managing AI for research
As far as research goes, we’ll always need humans in the loop. AI is a tool that will speed up the research gathering and analysis process, but humans need to be part of the actioning of any AI-driven research and put it to use.

As well as training, there needs to be clear and ongoing communication about how and when AI is used. Think of it like a label – if something has been produced using AI, say so. That way, anyone looking at the insights knows how they have been created. This transparency also makes it easier to track where data comes from and understand potential privacy risks.

Related to privacy and security is knowing the tool you’re using and the vendor you are purchasing it from. It is critical to work with trusted vendors and understand the training data they are using for AI.

There’s no denying AI could have huge potential value to researchers, but the reality is that AI isn’t smart enough to do everything. It’s great at summarising and aggregating large amounts of data, but not great at identifying and understanding outliers of the data, which are very important to understanding the big picture of customer needs when conducting research. Investing in training, promoting transparency, maintaining communication and interrogating its use will help any team take the necessary steps to avoid the pitfalls without restricting the opportunities AI can create. 

Serena Chan is research advocate at Dovetail