When I first joined HubSpot’s conversational marketing team, the majority of our website chat volume was managed by humans. We had a global team of more than a hundred live sales representatives – Inbound Success Coaches (ISCs) who qualified leads, booked meetings, and routed conversations to sales reps. It worked, but it didn’t scale.
Every day, these ISCs received thousands of chat messages from visitors who needed product information, had support questions, or just wanted to explore. While we really enjoyed these interactions, they often drew the focus of high-intent prospects who were ready to commit to the sale.
We knew AI could help us work smarter, but we didn’t want another scripted chatbot. We wanted something that could do it think like a sales representative: qualify, advise and sell in real time.
That’s how SalesBot was born – an AI-powered chat assistant that now manages the majority of HubSpot’s inbound chat volume, answers thousands of Chatter questions, qualifies leads, books meetings, and even directly sells our Starter tier products.
Here’s what we learned along the way.
How we developed SalesBot and what we learned
1. Start with distraction. Then build according to demand.
When we first launched SalesBot, our main goal was to head off easy-to-answer questions with low sales intent (example: “What is a CRM?” or “How do I add a user to my account?”). We wanted to reduce the noise and free people to focus on more complex conversations.
We trained the bot using HubSpot’s knowledge base, product catalog, Academy courses, and more. We now redirect over 80% of chats on our website using AI and self-service options.
This defensive success gave us confidence but also revealed our next challenge. Business does not grow through distraction alone. To truly scale the value, we needed a tool that does more than just provide solutions – it must sell.
2. Use assessment meetings to close the gap.
When we introduced the distraction, we saw a decline in medium-intent leads – those who weren’t ready to book a meeting but were still showing buying signals. People are great at recognizing these moments. There are not bots yet.
To address this gap, we developed a real-time propensity model that scores chats on a scale of 0-100 based on a mix of CRM data, conversation content, and AI-predicted intent. If a chat exceeds a certain threshold, it is classified as a qualified lead.
This model now helps SalesBot identify high-potential opportunities – even if a customer doesn’t specifically ask for a demo. It’s a perfect example of how AI can do this Surface nuance on scale.
3. Build to sell, not just to support.
After laying the foundations of distraction and assessment, we turned our attention to something bolder: turning SalesBot into a true sales assistant.
We trained it against our skills framework (GPCT – Goals, Plans, Challenges, Timeline) so that the bot can guide prospects to the right next step: be it getting started with free tools, booking a meeting with sales, or purchasing a starter plan directly in chat.
Now we have a tool that doesn’t just respond – it qualifies, builds intent, and promotes like a rep. This shift has fundamentally changed the way we think about conversational demand generation.
4. Choose quality over CSAT.
We quickly realized that traditional chatbot metrics like CSAT (Customer Satisfaction Score) were not enough.
CSAT measures how a customer feels about their experiences, typically by asking whether they were a critic, a passive, or a supporter after an interaction. But only a small proportion (less than 1% of chatters) take part in the survey. And even if a customer rates a chat positively, that doesn’t necessarily mean that the sales bot is providing a high-quality chat experience.
That’s why we created a custom quality rubric using our top-performing ISCs to define what “good” actually looks like. The rubric measures factors such as depth of discovery, next steps, tone and accuracy.
This year alone, a team of 13 reviewers manually reviewed more than 3,000 sales conversations. This human quality assurance loop is crucial. This keeps our AI anchored in real sales behavior and helps us continually improve performance.
5. Scale globally to increase efficiency.
Before AI, one of our biggest operational challenges was staffing live chat in seven languages. It was costly, inconsistent, and difficult to scale.
Now we can handle multilingual conversations around the world, providing a consistent experience no matter where someone is chatting from. This isn’t just an efficiency gain – it’s an improvement in the customer experience.
AI has given us true global coverage without overwhelming our team and enabled growth in regions where headcount simply couldn’t keep up.
6. Build the right team structure.
The success is not due to a single person or team, but rather to a group of smart, customer-focused developers coming together in the field of Conversational Marketing and Marketing Technology AI Engineering.
Conversational Marketing was responsible for strategy, user experience and quality assurance and always based its decisions on what would provide the best experience for our customers. Our AI engineering partners in Marketing Technology developed the models, prompts, and infrastructure that quickly turned these ideas into reality.
Together we formed a unified working group with common goals, a common backlog, and a rhythm of weekly experiments. This mix of deep customer empathy and technical excellence allows us to operate like a product team – testing, learning, and improving SalesBot with each release.
7. Approach automation with a product mentality.
The biggest benefit of our journey was adopting a product mentality. SalesBot was not a one-off automation project. It is a living product that evolves with each iteration.
Over the last two years, we’ve transitioned from rule-based bots to a Retrieval-Augmented Generation (RAG) system, updated our models to GPT-4.1, and added smarter qualification and product pitching features.
These upgrades doubled responsiveness, improved accuracy, and increased our qualified lead conversion rate from 3% to 5%.
We didn’t get there overnight. It required hundreds of iterations and a culture that sees AI experimentation as a core part of the go-to-market movement.
8. People still matter.
Despite all these advancements, some things still require a human touch. Today, SalesBot can’t create custom offers, handle complex objections, or replicate empathy in nuanced conversations—and that’s okay. We will always work to expand its capabilities, but human oversight will always be critical to maintaining quality.
Our agents and subject matter experts play a central role in our success. They evaluate the results, provide feedback and ensure that the system continues to learn and improve. Your judgment defines what “good” looks like and ensures our quality standards remain high as technology advances.
The role of AI is to scale reach and speed – not to replace human connection. Our ISCs are now focused on higher value programs and edge cases where their expertise really shines. The goal is not fewer people, but more intelligent and effective use of their time.
9. Give your model structure, not just more data.
When we first developed SalesBot, it ran on a simple rules-based system – X action triggers Y reaction. For basic logic it worked, but it didn’t sound like a salesman. We wanted something closer to an ISC: talkative, confident and helpful.
To get there, we experimented with fine-tuning. We exported thousands of chat transcripts and annotated them for tone, accuracy, and wording from ISCs. Training the model on these examples made it sound more natural, but the accuracy decreased. We learned the hard way that too much unstructured human data can actually hurt model performance. The model begins to remember the “edges” of what it sees and blurs everything in between.
So we switched. Instead of giving the model more Data, we gave it better Structure. We moved to a RAG (Retrieval-Augmented Generation) setup, anchored the tool in real-time context, and taught it when to use knowledge sources, tools, and CRM data.
The result is a bot that is significantly more reliable in complex sales conversations and is far better at recognizing intentions.
Here’s how to start building an AI chat program
When you’re just starting out, the biggest misconception is that you can jump straight into AI. In reality, AI is only successful when the foundation underneath is strong. Looking back on our journey, these three principles were most important.
1. Build the foundation before you automate.
AI is only as good as the human program it learns from. Before we automated anything, we had real conversations led by experienced chat agents for years. This live chat foundation has given us:
- High quality training data
- A clear definition of what “good” looks like
- Patterns to identify what could be automated first
If you skip this step, your AI won’t know what’s “good” – and it won’t know when it’s wrong.
2. Understand what makes your people great. Then teach the AI.
AI cannot replicate the nuances that come with human interaction.
Study your top performing employees closely and ask yourself the following questions:
- How do they qualify?
- What signals do they perceive?
- Which language creates trust?
- How do they recover when something goes off script?
Your human team is your blueprint. Everything great humans do—from tone to timing to discovery—becomes the foundation of an AI that can actually sell, not just answer questions.
3. Build an experimentation and data-driven team.
AI is not a set-it-and-forget-it project. Tt is a product and the only way to scale an AI chat program is to build a team that:
- Constantly experiment
- Moves quickly through iterations
- Measures what works (and what doesn’t)
- Treat errors as inputs, not setbacks
An experimental team transforms AI from a one-time introduction into a continuously improving growth engine.
The conclusion
The biggest insight for me is this: AI doesn’t replace a great go-to-market strategy – it accelerates it. Your tools should be a reflection of how you work. For us, it’s a mix of technology, creativity and customer empathy to constantly evolve the way we sell.