How did Kaizan start?

Glen and I had been discussing how large organisations waste considerable time acting as information conduits between teams — repeatedly sharing the same messages both internally and externally.

When we witnessed the AI breakthroughs, we were amazed by the possibility of using sophisticated AI to handle this communication at scale.

What was your vision when you started building the AI for Kaizan, and how has it evolved?

We wanted to build an AI friend/buddy who would be with you from a young age, helping you with your learning — the best teacher/friend/older sibling when it comes to learning. Then it graduates with you and enters the workforce, helping you learn and get better at your job.

It’s a repository of everything you have done, understands your strengths and weaknesses and helps you throughout life.

Since the problem is huge, with Kaizan, we’ve started with optimising client-facing teams – an often undervalued and neglected area of businesses.

Eventually our AI would establish itself across the wider business. I’d even love to see it evolve beyond AI for the workplace and become AI for life.

What unique challenges do client services and success teams face that Kaizan’s AI was designed to address, and how did those challenges shape the technical development?

Client success teams, particularly in professional services, have to juggle multiple roles — project management, customer support, upselling, and cross-selling. When I say project management, I mean coordinating internal teams to complete client work while maintaining constant communication. They’re responding to emails, chats, and meetings, gathering requirements — all of which requires significant effort.

There’s also considerable administrative work involved in updating systems and transferring information between platforms. As Glen puts it, CSMs are “water carriers of information” — they’re the glue connecting multiple perspectives, linking clients to internal teams.

They have the crucial task of truly understanding clients’ current needs and accurately translating them in-house. This complexity means a simple CRM isn’t sufficient; you need to operate across multiple systems — CRM, project management tools, note-taking applications — while coordinating across teams. Before AI tools emerged, this comprehensive integration was practically impossible because you needed tools that could “understand” human language – but not anymore. That’s what’s developed in recent years; although AI might just be sophisticated auto-complete, it’s exceptionally effective!

These challenges are unique because client success teams don’t operate in silos. Unlike sales teams (who acquire clients and finish when a deal closes) or development teams (who build products internally), client success teams must interface across the organisation. This means you can’t build systems for them using traditional approaches — the solution must function as a central hub that communicates with other teams’ systems. Client success teams serve as the nexus, transferring information between clients and the organisation. While sales typically ends at the point of sale, client success continues with ongoing client interaction, handling renewals and retention.

We know it’s much easier to retain a client with good service than to acquire a new one. Kaizan approached this by examining the daily challenges CSMs face, identifying three key areas to address:

1) Automating administrative tasks by facilitating information flow between systems

2) Enhancing productivity through recommendations about priorities and client focus

3) Providing “superpowers” by offering insights across the entire company – essentially creating a ChatGPT-like tool powered by the company’s internal documents and communication data

How do you tackle challenges related to data availability and quality when building AI solutions?

There are so many models out there and most of them are comparable to each other. There are slight advantages and differences between them, but the capabilities of models will largely progress at similar rates.

Where the power comes in application companies like ours is by providing domain expertise and user insight to train specific outputs and outcomes. Since most of the enterprise models have been trained on public data, when you have non-public data like client data, which is specific to the clients and industry that we are working with, we are able to train and fine tune these models to provide a superior output.

Today, when designing an AI product you need to think about where you’re going to get these data sources from. For us, we are thankful to our clients who are able to plug in their workspace and communication data which gives us domain- and client-specific data, which we can use in conjunction with our domain expertise to transform the data into the outcomes that our clients would want.

On a product level, we build pipelines and feedback mechanisms that look at user action and user inaction to train our models to become better.

How does feedback shape the engineering roadmap, and what trade-offs do you have to make when implementing new features?

Feedback is the most important thing when developing products in today’s world.

Everything is changing so fast, and to really stand out, you need user love. And user love comes when you deeply listen to users — when you really understand their problems.

During feedback sessions, you need to think about what the user is actually trying to accomplish. Users will often suggest solutions, but these don’t necessarily address the underlying problem they’re trying to solve. You need to dig deeper and understand their actual goals. If they say, “Change this feature to work like this,” it might solve their immediate concern, but the real pain point may exist elsewhere.

This is where the “5 Whys” technique is valuable. If you ask “why” enough times, you truly understand what they’re trying to do and what problem they’re attempting to solve. Then you need to understand their workflow and day-to-day challenges so that you’re able to solve that particular problem effectively.