Customer Research Analysis tasks I let AI do to save me hours weekly ⬇

What everyone says you should do these days?
"Continuous discovery."

What no one can actually do in one short team meeting every Friday?
ANALYSIS. 😵‍💫

There are a lot of things that can take a ton of time in the research process, but Analysis is certainly a big one.

(It’s why I created a course specifically for Customer Research Analysis with AI).

Unfortunately, not even the most beautiful affinity mapping template in Miro makes analysis happen faster.

But LLMs actually can.

It’s worth noting that I'm someone who does not believe (yet) that my experience and analysis skills will be replaced by AI anytime soon.

I’m also highly aware of the bias that can be involved in a lot of AI processes.

But in the process of analyzing my own customer research, LLMs have had a huge impact for me.

As someone who works with multiple clients and a ton of tasks at once, LLMs feel like extremely trainable interns - and even partners.

Plus, I'm a parent of a tiny kiddo and just don't have any time. 🫤 I’ll take time hacks anywhere I can (just not at the expense of quality).

If you’re looking for a few ideas for where specifically to use LLMs in your analysis process, here’s a list -

How I use LLMs to support me with Analysis and Insights:

▶ Data cleaning: Use LLMs to identify potential problems in data, like poor quality transcripts with missing phrases, or strangely formatted CSVs.

▶ Data set formatting: Turn notes and transcripts into other formats like a table so you have a clearer overview of your data set

▶ Theme checking: Group customer quotes by theme / patterns, and checking themes AI identifies against your own notes (you ran the research, right? 🙃)

▶ Frequency: Identify how “proven” an insight is by having ChatGPT find the number of observation instances in your quotes

▶ Top Quotes: Find the best quotes on [X] from a transcript that best fit the story you’re telling

▶ Contexts & Use cases: Identify the contexts where customers experience problems, log all use cases and frequencies

▶ JTBD Statements: Write statements for you from the quotes that explicitly define Jobs, desired transformations and more.

▶ Profiling: Summarize what some or all customers have in common based on transcripts, build data-backed segments faster

▶ Insight statements: Write or re-write insight statements that are clear and concise, tailored to specific audiences and actionable

▶ Need more evidence: Highlight when I don’t have enough information about something specific. If it’s important, I know to include this in the next discovery round

▶ Highlight biases + gaps: Point out how reliable the research and analysis might be, based on biases and missing pieces identified in the raw data. This can be critical for helping me and a client team know just how much we can trust the research

Can I really trust AI?

This is a big question with a lot of dependencies, but sometimes yes - IF you have the skills to guide it.

LLM builders and prompting experts all consistently share one message:

The quality of outputs from any LLM depends on the quality of the input AND your skill at writing great prompts.

AI alone can’t solve everything for you.

Plus, I’m still really careful about one particular thing…

🚨 Warning 🚨
ChatGPT still often inserts statements and observations that weren’t in the transcripts! Yikes. In the best of cases, it might synthesize your customers’ input in a way that kills the value.

We have to be sure that we know how to tell it what we need, and keep it on track.

__

LLMs can be a major time-saver for me.

But no chat window is ever a casual conversation I’m just winging.

I go in with a clear plan, and I don’t trust AI blindly.