We do a massive amount of customer research here at Near Horizon. Most of the companies we work with are at the formation stage, and our early work is to understand the problem we’ll tackle, the shape of possible solutions, and how potential customers think. That means hundreds of interviews over a few months.

Customer research is a critical step in starting a company. It is a chance to test your assumptions, and to learn about your market deeply enough to craft a winning strategy. It is painful and time-consuming, but doing it well ensures all the years of effort you will invest in your startup are worth it. 

A lot of founders skip this step, convinced they fully understand their market or problem. What they don’t understand is that customer research helps you in many ways:

• Determine whether your problem is important enough to be worth years of your life. 
• Explore how many different people (or companies) experience the problem, and find your ideal customer profile (ICP). 
• Learn the language these customers use to describe the problem. This is critical because it will form the basis of your positioning and marketing in the future. 
• Learn how these potential customers buy products and the process you’ll need to sell to them. This will ensure we can both reach and sell to these customers if we build a solution. 

Even if you are an expert, you cannot know these things from your personal experience. You need hundreds of interviews with everyone you can find in the industry, across companies, levels, and backgrounds. 

Years ago, customer research was extremely painful. You would assemble notes on each interview and spend weeks sifting to extract the themes and insights. It was monotonous, difficult, and error-prone work. 

AI tools have completely changed the game. Today, our customer research process is still time-consuming (there is no shortcut for interviews), but we can analyze the results with unprecedented speed and power. 

Here’s how:

Step 1. Interview Transcription

Customer research interviews typically last 20-30 minutes, and with hundreds of them it’s easy to get overwhelmed if you attempt to take notes. Luckily, there is a wealth of tools available to transcribe your meetings, albeit with minor errors and human-speech idiosyncrasies. That’s fine for our purposes because all of the information is hidden in that text transcript! 

With hundreds of transcripts in hand, you can move on to step 2…

Step 2. Interview Summaries

Take each of those interview transcripts and feed it into your favorite LLM with the following prompt:

This is the transcript of an interview. Provide every question and a summary of the response.

The result is a well-formed and well-written set of questions and answers that you can easily double-check to make sure they match your conversation. If you have a standard set of questions, these will become consistent, but it’s useful even if you don’t. 

You should double-check these summaries to make sure they match your understanding of the interview as LLMs do make mistakes!

Step 3. Extract Themes & Insights

Customer research is useless unless we can extract insights from the massive amount of information. Once again, AI makes this painless.

Take your hundreds of interview summaries and, once again, feed them into your LLM of choice using the following prompt:

These are a series of interviews. Extract the common themes and points and provide a list.

The result should be a series of themes that you can explore further by asking the LLM to explain them in more detail. You can have an entire conversation with your customer research data and extract valuable insights that you can use to build your startup. 

Why not just feed in the raw transcripts for this step? You could, but there is a lot of danger in customer research of LLMs hallucinating and producing misleading results. By creating summaries, you can check to ensure they are accurate and reduce the chances of incorrect conclusions. You can also verify your insights against those summaries, giving you even more confidence. 

The Future of Customer Research

Even with AI tools, the hardest part of customer research is setting up and holding the interviews themselves. Getting a hold of hundreds of people and spending 30 minutes with them is time-consuming, no matter what tools you use. 

Could that part be automated as well? We suspect not, because customer research is an art as much as science. Conducting a great customer research interview requires empathy and intuition, where you react as much to how the person responds as to what they are saying. 

Still, finding and scheduling those interviews is an area where we hope tools can help. The easier it is to do comprehensive customer research, the better all startup companies will become.