We live in exciting times.

AI is improving by leaps and bounds week after week. Everyone’s trying to bolt AI onto their product. But AI only adds value when it’s solving the right problem.

One way I've explored using GenAI is offering a scalable, inexpensive conversational experience - the kind that could support engaging advisor-like interactions.

To test this I quickly prototyped a ChatGPT-based experience with an elaborate prompt. What became apparent was that while the AI could generate responses, the quality was inconsistent, and - more importantly - users were hesitant to open up in a computer-mediated interaction. AI lacks the genuine feeling of empathy you find talking with a human.

The truth is - AI isn't a magic bullet.

Time for a Rethink

It became clear that while AI had potential, I hadn’t yet aligned its capabilities with a well-defined user need.

Once I revisited the original problem I was trying to solve - offering a superior service to our users - it was clear the initial direction wasn’t fully aligned with the problem. The breakthrough came when I started asking 'what does our team actually need to offer a superior service?' instead of 'what can AI do?'

Rather than AI replacing the capabilities within the team, we pivoted toward using AI to amplify our teams' capabilities instead. Our goal became to equip them with superpowers.

Conversations with users would still be human-to-human, but supported by Natural Language Processing, integrated services, and content resources.

A huge number of different approaches become available when you start thinking about how you can give superpowers to your team. In thinking through potential applications, I focused on ideas that balanced feasibility, usefulness to advisors, and user trust. Possible approaches could include:

• Summarising conversation history to reduce the prep time before a session.
• Automated note-taking during conversations.
• Surface real-time sentiment cues to support advisor awareness.
• Suggest relevant exercises and draw from anonymized insights or content.
• Progress tracking and visualisation across multiple sessions.
• Automate scheduling and follow-up reminders based on the conversation outcomes.

To prioritise and understand how we should approach it, we considered how it was adding business value and the feasibility of building and supporting it.

A Better AI Product

It probably won't strike you as a surprise that taking a problem-first approach for AI led to far better user satisfaction and higher engagement.

As tempting as it is to go all-in on AI and be blinded by the latest and greatest release, be sure to:

• Quickly prototype ideas and test things out.
• Be willing to fail and pivot.
• Remember the reason your product exists in the first place.

With these things in mind, you'll create a better problem-first AI experience.

Have you tried to integrate AI into a human-first workflow? I'd love to hear what worked for you.