![\"Download](\"https:\/\/no-cache.hubspot.com\/cta\/default\/53\/6f674af4-3116-43b0-8a54-4a64f926afb6.png\")
<\/a><\/p>\nEvery day, those ISCs fielded thousands of chat messages from visitors who needed product info, had support questions, or were just exploring. While we loved those interactions, they often pulled focus from high-intent prospects ready to engage with sales.<\/p>\n
We knew AI could help us work smarter, but we didn\u2019t want another scripted chatbot. We wanted something that could think<\/em> like a sales rep: qualify, guide, and sell in real-time.<\/p>\n That\u2019s how SalesBot was born \u2014 an AI-powered chat assistant that now handles the majority of HubSpot\u2019s inbound chat volume, answering thousands of chatter questions, qualifying leads, booking meetings, and even directly selling our Starter-tier products.<\/p>\n Here\u2019s what we\u2019ve learned along the way.<\/p>\n When we first launched SalesBot, our primary goal was to deflect easy-to-answer, low sales intent questions (example: \u201cWhat\u2019s a CRM\u201d or \u201cHow do I add a user to my account\u201d<\/em>). We wanted to reduce the noise and free up humans to focus on more complex conversations.<\/p>\n We trained the bot on HubSpot\u2019s knowledge base, product catalog, Academy courses, and more. We are now deflecting over 80% of chats across our website using AI and self-service options.<\/p>\n That success in deflection gave us confidence, but it also revealed our next challenge. Deflection alone doesn\u2019t grow the business. To truly scale value, we needed a tool that does more than resolve \u2014 it has to sell<\/em>.<\/p>\n Once we introduced deflection, we noticed a drop-off in medium-intent leads \u2014 the ones that weren\u2019t ready to book a meeting but still showed buying signals. Humans are great at spotting those moments. Bots aren\u2019t \u2026 yet.<\/p>\n To close that gap, we built a real-time propensity model that scores chats on a scale of 0\u2013100 based on a blend of CRM data, conversation content, and AI-predicted intent. When a chat crosses a certain threshold, it\u2019s raised as a qualified lead.<\/p>\n That model now helps SalesBot identify high-potential opportunities \u2014 even when a customer doesn\u2019t explicitly ask for a demo. It\u2019s a perfect example of how AI can surface nuance<\/em> at scale.<\/p>\n Once we\u2019d nailed the foundations of deflection and scoring, we turned our attention to something bolder: turning SalesBot into a true selling assistant.<\/p>\n We trained it on our qualification framework (GPCT \u2014 Goals, Plans, Challenges, Timeline), enabling the bot to guide prospects toward the right next step: whether that\u2019s getting started with free tools, booking a meeting with sales, or purchasing a Starter plan directly in chat.<\/p>\n Now, we have a tool that doesn\u2019t just respond \u2014 it qualifies, builds intent, and pitches like a rep. That shift fundamentally changed how we think about conversational demand generation.<\/p>\n We quickly realized that traditional chatbot metrics like CSAT (Customer Satisfaction Score) weren\u2019t enough.<\/p>\n CSAT measures how a customer feels<\/em> about their experience, typically by asking whether they were a detractor, passive, or promoter after an interaction. But only a small portion (less than 1% of chatters) complete the survey. And even if a customer rates a chat positively, that doesn\u2019t necessarily mean the Salesbot was providing a quality chat experience.<\/p>\n So we built a custom quality rubric with our top-performing ISCs to define what \u201cgood\u201d actually looks like. The rubric measures factors like discovery depth, next steps, tone, and accuracy.<\/p>\n This year alone, a team of 13 evaluators manually reviewed more than 3,000 sales conversations. That human QA loop is critical. It keeps our AI grounded in real-world selling behavior and helps us continuously improve performance.<\/p>\n Before AI, staffing live chat in seven languages was one of our biggest operational challenges. It was costly, inconsistent, and hard to scale.<\/p>\n Now, we can handle multilingual conversations around the world, providing a consistent experience no matter where someone\u2019s chatting from. That\u2019s not just an efficiency win \u2014 it\u2019s a customer experience upgrade.<\/p>\n AI has given us true global coverage without overextending our team, unlocking growth in regions where headcount simply couldn\u2019t keep up.<\/p>\n Success didn\u2019t happen because of one person or team \u2014 it happened because a group of smart, customer-driven builders came together across Conversational Marketing and Marketing Technology AI Engineering.<\/p>\n Conversational Marketing owned the strategy, user experience, and quality assurance, always grounding decisions in what would deliver the best experience for our customers. Our AI Engineering partners in Marketing Technology built the models, prompts, and infrastructure that made those ideas real \u2014 fast.<\/p>\n Together, we formed a unified working group with shared goals, a common backlog, and a rhythm of weekly experimentation. That mix of deep customer empathy and technical excellence let us move like a product team \u2014 testing, learning, and improving SalesBot with every release.<\/p>\n The biggest unlock in our journey was embracing a product mindset. SalesBot wasn\u2019t a one-off automation project. It\u2019s a living product that evolves with every iteration.<\/p>\n Over the past two years, we\u2019ve moved from rule-based bots to a retrieval-augmented generation (RAG) system, upgraded our models to GPT-4.1, and added smarter qualification and product-pitching capabilities.<\/p>\n Those upgrades doubled response speed, improved accuracy, and lifted our qualified lead conversion rate from 3% to 5%.<\/p>\n We didn\u2019t get there overnight. It took hundreds of iterations and a culture that treats AI experimentation as a core part of the go-to-market motion.<\/p>\n Even with all this progress, some things still require a human touch. Today, SalesBot can\u2019t build custom quotes, handle complex objections, or replicate empathy in nuanced conversations \u2014 and that\u2019s okay. We\u2019ll always be working toward expanding its capabilities, but human oversight will always be essential to maintaining quality.<\/p>\n Our agents and subject matter experts play a core role in our success. They evaluate outputs, provide feedback, and ensure the system continues to learn and improve. Their judgment defines what \u201cgood\u201d looks like and keeps our standard of quality high as the technology evolves.<\/p>\n AI\u2019s role is to scale reach and speed \u2014 not to replace human connection. Our ISCs now focus on higher-value programs and edge cases where their expertise truly shines. The goal isn\u2019t fewer humans \u2014 it\u2019s smarter, more impactful use of their time.<\/p>\n When we first built SalesBot, it ran on a simple rules-based system \u2014 X action triggers Y response. It worked for basic logic, but it didn\u2019t sound like a salesperson. We wanted something that felt closer to an ISC: conversational, confident, and helpful.<\/p>\n To get there, we experimented with fine-tuning. We exported thousands of chat transcripts and had ISCs annotate them for tone, accuracy, and phrasing. Training the model on these examples made it sound more natural, but accuracy dropped. We learned the hard way that too much unstructured human data can actually degrade model performance. The model starts remembering the \u201cedges\u201d of what it sees and blurring everything in between.<\/p>\n So, we pivoted. Instead of giving the model more<\/em> data, we gave it a better<\/em> structure. We moved to a retrieval-augmented generation (RAG) setup, grounding the tool in real-time context and teaching it when to pull from knowledge sources, tools, and CRM data.<\/p>\n The result is a bot that\u2019s significantly more reliable in complex sales conversations and far better at identifying intent.<\/p>\n If you’re just getting started, the biggest misconception is that you can jump straight into AI. In reality, AI only succeeds when the foundation beneath it is strong. Looking back at our journey, these three principles mattered the most.<\/p>\n AI is only as good as the human program it learns from. Before we automated anything, we had years of real conversations handled by skilled chat agents. That live chat foundation gave us:<\/p>\n If you skip this step, your AI won\u2019t know what \u201cgood\u201d is \u2014 and it won\u2019t know when it\u2019s wrong.<\/p>\n AI can\u2019t replicate the nuances that come with human interaction.<\/p>\n Study your top-performing reps deeply, and ask yourself the following questions:<\/p>\n Your human team is your blueprint. Everything great humans do \u2014 from tone to timing to discovery \u2014 becomes the foundation for an AI that can actually sell, not just answer questions.<\/p>\n AI is not a set-it-and-forget-it project. Tt\u2019s a product, and the only way to scale an AI chat program is to build a team that:<\/p>\n An experiment-driven team turns AI from a one-time launch into a continuously improving engine for growth.<\/p>\n The biggest takeaway for me is this: AI doesn\u2019t replace great go-to-market strategy \u2014 it accelerates it. <\/strong>Your tools should be a reflection of how you operate. For us, that\u2019s a blend of technology, creativity, and customer empathy to keep evolving how we sell.<\/p>\n When I first joined HubSpot\u2019s Conversational Marketing team, most of our website chat volume was handled by humans. We had a global team of more than a hundred live sales agents \u2014 Inbound<\/p>\n","protected":false},"author":0,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[13],"tags":[],"_links":{"self":[{"href":"http:\/\/buywyo.com\/index.php\/wp-json\/wp\/v2\/posts\/4017"}],"collection":[{"href":"http:\/\/buywyo.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/buywyo.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"replies":[{"embeddable":true,"href":"http:\/\/buywyo.com\/index.php\/wp-json\/wp\/v2\/comments?post=4017"}],"version-history":[{"count":0,"href":"http:\/\/buywyo.com\/index.php\/wp-json\/wp\/v2\/posts\/4017\/revisions"}],"wp:attachment":[{"href":"http:\/\/buywyo.com\/index.php\/wp-json\/wp\/v2\/media?parent=4017"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/buywyo.com\/index.php\/wp-json\/wp\/v2\/categories?post=4017"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/buywyo.com\/index.php\/wp-json\/wp\/v2\/tags?post=4017"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}How We Built SalesBot and What We Learned<\/h2>\n
1. Start with deflection. Then, build for demand.<\/h3>\n
2. Use scoring conversations to close the gap.<\/h3>\n
3. Build to sell, not just support.<\/h3>\n
4. Choose quality over CSAT.<\/h3>\n
5. Scale globally to boost efficiencies.<\/h3>\n
6. Build the right team structure.<\/h3>\n
7. Approach automation with a product mindset.<\/h3>\n
8. Humans still matter.<\/h3>\n
9. Give your model structure, not just more data.<\/h3>\n
How to Get Started Building an AI Chat Program<\/h2>\n
1. Build the foundation before you automate.<\/strong><\/h3>\n
\n
2. Understand what your humans do great. Then, teach the AI.<\/strong><\/h3>\n
\n
3. Create an experiment-driven, data-driven team.<\/strong><\/h3>\n
\n
The Bottom Line<\/h2>\n
![\"\"](\"https:\/\/track.hubspot.com\/__ptq.gif?a=53&k=14&r=https%3A%2F%2Fblog.hubspot.com%2Fmarketing%2Fwhat-we-learned-building-salesbot&bu=https%253A%252F%252Fblog.hubspot.com%252Fmarketing&bvt=rss\")
<\/p>\n","protected":false},"excerpt":{"rendered":"\n