Early in HubSpot's AI transformation, CEO Yamini Rangan shared a video showing employees how she used AI in her own workday.
It wasn't a grand vision for an AI-powered future. It was a practical example of how AI already fit into the work.
The video gave employees a simple signal: AI was important enough for the CEO to learn in public. Jon Dick, HubSpot's chief customer officer, remembers people responding, in effect, that if Rangan was using it, they should try it too.
That visible example helped prepare HubSpot for a much larger change. The customer-platform company had spent years helping businesses generate demand, win deals, and retain customers. Those goals had not changed. But the work behind them remained full of stubborn constraints: sales reps lost time to research and administration, marketing depended heavily on search traffic, and customer teams struggled to give hundreds of thousands of customers personal help at scale.
HubSpot saw a chance to break those constraints. Over roughly three years, the company rebuilt how it attracted prospects, engaged buyers, and delighted customers. Today, shared AI systems carry parts of the research, coordination, and customer work that once depended more heavily on individual employees.
Keep reading or get the full story in Episode 234 of The Artificial Intelligence Show, as part of our special AI Transformations series presented by Google Cloud.
HubSpot Gave Employees Room to Learn
Rangan's video worked because it modeled a behavior instead of merely announcing a priority. HubSpot's leaders talked about AI, used it themselves, and gave employees dedicated time to learn.
They also made the learning visible. Employees shared wins and examples in Slack, joined hackathons, and experimented with a wide range of tools. Dick says that openness mattered because AI creates an unusual kind of anxiety: The tools change so quickly that even experienced users feel behind.
HubSpot tried to lower that fear before asking people to redesign important customer work. The early goal was fluency. As employees grew more comfortable with AI, teams could stop treating it as a separate technology project and start applying it to the customer journey.
Support Forced HubSpot to Change the Score
Customer support became HubSpot's first major customer-facing test. Dick says that was where generative AI had the clearest fit.
The team watched resolution rate, a support metric meant to show how often the system resolved a customer's issue. It looked strong.
Then HubSpot's customer satisfaction score, or CSAT, moved in the wrong direction.
"Everybody got really excited about resolution rates," Dick says. "I care way more about CSAT, honestly."
HubSpot changed what it optimized. For Dick, resolution was not enough if CSAT fell. The team made CSAT the priority and continued tuning the support experience around it. Dick says resolution also rose as customer satisfaction improved.
The lesson was bigger than support: Automating more work meant little if the customer got a worse result.
Support was one early test, not the whole transformation. As teams rebuilt content and coding workflows, HubSpot moved toward a broader operating question.
"We pretty much just looked at the entire customer journey and said, 'What are the constraints, and how can AI break them?'" Dick says.
That question gave HubSpot a way to prioritize. Leaders could map the steps customers take, identify where old limits held back the work, and decide where an AI system might make a meaningful difference. They could not rebuild everything at once. (In fact, Dick's advice for leaders undergoing AI transformation is to start with the business problem causing the most pain.)
Today, HubSpot's agents carry out bounded jobs, such as researching accounts. Assistants help employees apply shared company knowledge and can call on agents to complete parts of the work.
The resulting AI architecture now stretches across HubSpot's familiar attract, engage, and delight model. And each part of the journey required a different change, a different test, and a different measure of success.
From Search Traffic to AI-Led Discovery
HubSpot helped pioneer inbound marketing, a model built in large part around earning attention through useful content and search. Then search traffic declined, while more buyers began asking questions in AI answer engines.
The constraint had changed. Publishing more content for the old discovery path would not guarantee that HubSpot appeared in the new one.
AI answer engines created a new visibility contest: whether a company's material appears and gets cited in a generated response. HubSpot built an answer engine optimization strategy and used AI to help execute it, shifting part of the marketing team's work from ranking in search results to earning visibility inside generated answers.
Dick says conversions HubSpot attributes to this work grew nearly 2,000% over a recent two-month stretch.
Today, HubSpot attracts prospects differently from the inbound model it helped popularize. Instead of relying only on traditional search, the company is working to appear wherever buyers ask questions, including inside AI-generated answers.
From Individual Research to Shared Context
Sales presented a different constraint. For years, leaders tried to increase the share of time representatives spent with customers. The number barely moved because complex deals required so much account research, administration, and coordination.
Today, HubSpot splits that burden across two shared systems. A prospecting agent researches accounts and flags companies worth contacting, work that would otherwise fall to a sales development representative. Once a conversation is underway, Guided Success, a custom sales assistant, puts HubSpot's own guidance on winning deals in front of the representative working the deal.
That changes the division of labor. The agent takes on repeated prospecting research. The representative works the deal, while the assistant supplies HubSpot's context for deciding what to do next.
Dick says the prospecting agent booked about 10,000 meetings in the previous quarter.
From Broad Segments to More Account Coverage
When it came to delighting customers, HubSpot faced another scale problem. Customer teams wanted to give each account relevant advice, but the old rules-based automation could do little more than insert a name or sort customers into broad segments. Customer success managers also had many accounts to cover and the administrative work that came with them.
Generative AI opened a different path. Dick saw the potential to create more individualized guidance. Today, customer success managers use one assistant that coordinates other agents to handle parts of the account work. The goal is to clear more work off their plates so managers can spend more time with customers.
Dick says save rates now rise by seven points when managers use the assistant.
The Playbook Is a Sequence, Not a Tool List
HubSpot's way of building changed as the work became more consequential. The first phase was deliberately loose: Give people tools, invite experiments, share the wins, and learn quickly. But HubSpot wanted gains that showed up in company results, not only faster work for one employee.
So, the next phase brought subject matter experts, engineers, and data specialists together in cross-functional pods. More recently, HubSpot consolidated much of its go-to-market AI work under one leader and a shared set of priorities.
Today, that evolution remains unfinished. Central focus makes it easier to commit to larger work, but it also creates trade-offs between company-wide priorities and local experiments. HubSpot is still deciding what to standardize, what to keep testing, and which measures best reflect a better customer outcome.
But the company's experience points to a practical sequence of operations when it comes to this type of AI transformation:
- Make AI use visible from the top. Leaders should show how their own work is changing, then give employees time and safe places to learn.
- Start with a real constraint. Map the customer journey and choose a business problem, not a fashionable tool.
- Test the customer outcome. A higher automation rate is not enough if satisfaction, conversion quality, or retention suffers.
- Put company context into the system. Shared knowledge and go-to-market judgment matter more than a generic agent completing a task quickly.
- Turn useful experiments into shared systems. Individual exploration creates ideas. Cross-functional owners, common priorities, and agreed measures make the strongest ideas usable at scale.
- Keep people responsible for quality. Automation volume is not the goal. People remain responsible for whether the work helps the customer.
Rangan's video made one new behavior visible. And HubSpot's larger transformation then came from giving that behavior somewhere to go: into different work, different measures, and a rebuilt model for attracting, engaging, and delighting customers.
Did you enjoy this transformation story? Go deeper on how AI is reshaping work and business with The Artificial Intelligence Show. Each week, we break down what matters in AI and what leaders should do about it.
Mike Kaput
Mike Kaput is the Chief Content Officer at SmarterX and a leading voice on the application of AI in business. He is the co-author of Marketing Artificial Intelligence and co-host of The Artificial Intelligence Show podcast.