Our next couple episodes of Deployed will look specifically at AI in customer service – one of the biggest areas generative AI is making a significant impact for businesses.
For the first conversation, we sat down with Nick Francis (CEO & Co-founder) and Luis Morales (VP of Engineering) of Help Scout to discuss their journey transforming a 13-year-old customer service SaaS company into an AI-first business. With 12,000 customers worldwide, Help Scout had built a significant "traditional SaaS" business over the years that was already operating at scale. Now they've come so far so fast on their AI journey that they've made a huge bet on AI – to the point of getting rid of their traditional seat-based pricing model.
What makes their story particularly interesting is how they've approached this transition: rather than simply adding AI features, they've fundamentally reimagined their business model and product strategy around AI capabilities. And always with an "AI for delight, or not at all" principle guiding the way.
We dig into several big themes that will be interesting to other teams considering this shift:
How they built conviction to change their pricing model
Their journey to build generative AI expertise within their existing team
The success they’ve had from their new "AI Quality Lead" role that bridges customer service domain expertise and AI development
Shifting their product and engineering culture to embrace the experimental nature of building with AI
Below we highlight some key parts of the conversation – check out the full episode here on Spotify, Apple Podcasts, and YouTube.
Making a Big Bet on AI After 13 Years
When many established companies consider adding AI capabilities, they often default to charging extra for a new AI feature add-on or other bonus item to test the waters. Help Scout has taken the dramatic next step. After their early success delivering valuable AI features to customers, they completely restructured their pricing model to align their incentives better with their customers — and generative AI features are included for free. Help Scout’s pricing is not based on the number of “contacts helped,” not the number of support agent seats required by a business.
"This has been the only sort of incredibly disruptive cycle that we've been through as a business," Nick explains. "It doesn't happen that often, but it did feel like a founding moment. And whenever it feels like a founding moment in your market, it's important that you adopt day one thinking."
Nick talks about that Day One thinking in this clip, including:
Moving away from per-seat pricing to charging based on customer contacts
Making AI features free on all plans
Offering unlimited seats to democratize access to customer data
Building AI Expertise From Scratch
Like so many other software companies that have recently entered the AI space, Help Scout didn't start with a machine learning team or significant AI expertise. They had to build this new expertise from the ground up.
Looking back two years ago, Nick shares: "We were literally starting from scratch. We didn't have an AI team. We didn't have a machine learning team. We didn't have really any core competency within the walls of our business around AI."
Their approach:
Started by creating a working group that prototyped seven different generative AI features in 6 weeks
Initially shipped simple features like conversation summarization and translation
Then started building more complex capabilities like AI-powered email drafts with a custom RAG system
Acquired a small AI-focused company to accelerate learning
Luis talks more about how they initially built technical expertise in this clip.
Creating the "AI Quality Lead" Role
Help Scout has taken an approach that we’ve seen other successful teams take too. One of their most impactful innovations has been creating a new role to bridge the gap between AI development and customer needs. They took someone from their customer success team – someone with deep product knowledge but no engineering background – and made them responsible for ensuring AI features truly serve customer needs.
This role that they’ve called “AI Quality Lead” has become central to their AI development process. The person is responsible for:
Improving prompts based on deep customer understanding
Defining and running evaluations
Helping prioritize what to build next / what issues to prioritize fixing
Ensuring AI features maintain the right tone and quality
"He definitely has been improving the prompts to make them especially tone and quality. A month in, he did a couple of tweaks to one of the prompts and our own customer team was like, 'it's amazing, it feels like it's one of our own writing those drafts,'" Luis shares.
Embracing an Experimentation Culture
Perhaps the biggest change for Help Scout has been shifting from traditional software development practices to the more experimental approach required for AI products. We’ve found this shift to be surprising to lots of teams, and especially to leaders who are looking for dependable delivery schedules.
"Initially, it was hard because the engineers were not used to that," Luis shares. "They wanted to do their research and their discovery. And once they were done with that, they just wanted to go and implement it and everything was going to go following the plan. The non-deterministic nature of LLMs means that's not going to happen at any time."
Nick emphasizes how this requires a different approach to shipping: "We now have this phrase that we use a lot in the experience org, which is 'ship when it's dangerous.' If you're not shipping when it's dangerous, it's too late."
This cultural shift has meant:
Accepting longer timelines from POC to production
Higher tolerance for risk in shipped products
Recognition that improvement requires real usage data as a precursor, so they know where to focus
More emphasis on monitoring and evaluation in production
Take Aways
Help Scout has taken bold steps to make the transition to an AI-first product company and business. Hopefully their lessons learned are helpful For other leaders considering similar transitions
To recap some of their biggest learnings and advice:
Start with clear principles and a vision for how AI should serve customers
Be willing to question fundamental business model assumptions
Invest in building the right team structure and culture
Focus on shipping and learning from real customer usage – even when it feels risky
Thanks to Nick and Luis for sharing their experience! If you're interested in learning more about Help Scout's approach to AI, check out their website here.
