Biggest AI Mistakes Companies Are Making with LeadGen & Lead Management Today

Philip Levinson (00:00)

Welcome, everyone. We’re excited to host you for today’s webinar: Biggest AI Mistakes Companies Are Making with LeadGen and Lead Management Today, and How to Fix Them. This covers two topics that are top of mind for marketers like me: leads and AI.

This session is part of our HubSpot User Group, also known as HUG. Our group is the AI and Automation in Sales and Marketing user group. We’re thrilled to host our members, as well as friends of the great companies here today.

I want to introduce my partner in crime and fellow co-leader of our HubSpot User Group, Jeralin. Jeralin, how are you?

Jeralin Hamann (00:58)

I’m good. Thanks, Phil. I’m excited for today’s session. Thanks to everyone who carved out an hour to chat with us about this very interesting, top-of-mind topic.

Philip Levinson (01:09)

We have well over 250 registrants, and we’re thrilled. Let me introduce our panelists.

From HubSpot, we have two superstars: Brenda Lando Fridman and Adam Wainwright. Also from Silicon Valley, we have the VP of Demand Gen at LeanData, Tara Montanez. Tara has a big event next week, which she’ll talk about as well.

We also have Kartik Hosanagar, co-founder and CEO of Bodhium Labs. From Evenbound, we have Shane Torrey out in Maryland. Finally, we have Nick Natale from Mobileforce in upstate New York. Nick is also a co-lead of our user group, so thank you for helping make this happen.

Jeralin, why don’t you walk everyone through the icebreaker and the agenda?

Jeralin Hamann (03:05)

To kick off our conversation, we like to gauge the room so our panelists can see who’s in attendance.

Here’s our icebreaker. AI will automate the reality it is given. On a scale of one to five, how much would you trust your AI with the context in your CRM today? A one means “Please don’t look at our data.” A five means you are clean, connected, and ready. I see some answers trickling in, and I expect most of us will land somewhere in the middle.

We also have a HubSpot User Group on LinkedIn. We share thought-provoking content there, recap HUG events, and discuss updates from the AI and automation world. We have more than 300 community members there, in addition to more than 600 in our hub. We’ll post the link in the chat.

Here’s our agenda. The HubSpot team will kick us off with build versus buy in the AI era and Revenue Hub. Next, Bodhium Labs will talk about rethinking AI visibility and SEO in lead generation. Evenbound will then cover avoiding the AI demand generation trap. After that, Tara from LeanData will discuss when vibe-coded lead management breaks down. Finally, Nick from Mobileforce will wrap up with turning AI-generated demand into revenue.

We’ll save time at the end for Q&A. Please leave comments and questions in the chat throughout the session.

We also like to start with a recent quote from a thought leader. This one comes from HubSpot CEO Yamini Rangan on LinkedIn. She said AI adoption is everywhere, but AI transformation is not. Roughly 90% of companies are using AI, which isn’t surprising. But only 6% say they’re seeing transformational results, which is very surprising. Those companies are four times more likely to hit their revenue goals and three times more likely to hit their efficiency targets.

So what separates the 6% from everyone else? That leads us to what we’re calling the outcomes era. Keep that top of mind as we go through today’s conversation. With that, I’ll hand it over to Brenda and Adam from HubSpot.

Brenda Lando Fridman (07:14)

Thank you, Jeralin. I’m usually based in San Carlos, but I’m in HubSpot’s San Francisco office today. It’s closing day, everybody, so thank you even more for spending time with us.

I want to tee up build versus buy in the AI era. Fun story: I carpooled this morning with a friend whose company believes we can vibe-code everything and replace all the software in the world. We had an interesting debate on that hour-long car ride.

I worked at Google for a long time. Google built everything. They built their own CRM, their own applicant tracking system, and their own version of Workday. However, most companies aren’t set up like Google, with lots and lots of resources.

The question isn’t whether you can build something. We all can, and building is fun. What happens after you build it is where things get hairy.

For HubSpot, Revenue Hub is about connecting the full revenue journey. That runs from lead to deal, quote, contract, invoice, payment, renewal, and expansion. It also includes all the business rules and guardrails that make that work in the real world.

The key message is this: we want you to build with HubSpot rather than try to rebuild HubSpot. In a minute, Adam will go deeper on what’s difficult to recreate, where hidden costs and risks show up, and why the foundation matters as you scale.

We also announced a lot at INBOUND. Our three main themes are that HubSpot is a CRM so smart it updates itself, with context and control built in.

The next slide shows the build versus buy argument in terms of five-year total cost of ownership. HubSpot is one of the trickier platforms to build yourself. The cost of building is probably the smallest piece. Then come the costs of running, maintaining, risk, and the quality gap over time.

Finally, here’s how HubSpot partners with you. We’re model agnostic, transparent on pricing, and backed by a large partner ecosystem. We’re also governed by default, and we offer a direct line to our executive team. Over to you, Adam.

Adam Wainwright (11:28)

Thank you, Brenda. I hope everybody is supporting their sales teams by rocking green for quarter end.

I like to think of Revenue Hub as downstream from HubSpot’s marketing capabilities. As we think about our long-term investments in Revenue Hub, especially the role AI plays, we learned something early. AI will automate the reality it’s given. If that reality isn’t organized, optimized, and oriented around best practices, we’ll get poor results, or maybe slop.

We tested this hypothesis with about 300 of our executive customers and sales leaders. At scale, we found that AI does not fix fragmentation. Good infrastructure is still at the center of what most businesses are trying to accomplish. Getting good returns from AI depends on having a strong foundation.

So where does HubSpot come in? From a build versus buy perspective, customers invest in the foundation. In my world, that means quoting and what we now call the contract object. Customers get strong, durable, reliable data architecture.

The contract object picks up commercial detail and becomes the stable rails for what we call revenue context. That’s the North Star for how customers use Revenue Hub within their broader HubSpot investment. Revenue context can power AI and Breeze to surface white space opportunities and support net dollar retention strategies. The goal is to give AI the actual reality the executive team needs to build a more predictable business.

So where does building come in? I like to reframe the question. With AI, the question is now how much you should build. If you try to vibe-code HubSpot, you’ll probably dig yourself into a hole.

HubSpot is a platform, and it becomes far more powerful when builders create process workflows on top of it. Guided selling is one example. Over the last year or two, the distance between a required business process and the technical skill needed to build it has virtually collapsed.

So it’s less about building and more about composing on top. Make the investment in the foundation. Then use AI to tweak, tune, and iterate business processes that still feed that foundational architecture.

Practically speaking, many customers are building complex guided selling workflows. These used to be complex Excel calculators or outdated homegrown solutions. Now customers rebuild them directly on HubSpot using serverless functions, HubDB, and React modules. They all live in the context of the platform.

The future isn’t really whether to build or buy. It’s how much you compose on top of the foundation you invest in. Vibe-coding from the beginning will probably set you up for failure.

Jeralin Hamann (17:43)

Thank you, Adam and Brenda. With that, we’ll hand it over to Kartik.

Kartik Hosanagar (17:51)

Thank you, Jeralin. Nice to meet everyone. I’m Kartik Hosanagar, co-founder and CEO of Bodhium Labs. I’m also a professor at the Wharton School at the University of Pennsylvania. We’re working on simulations to help build self-improving marketing AI. Over the next five minutes, I’ll explain what that means.

Five years ago, this would have felt like magic. Today, marketers can dream up any content, give it as a prompt, and get it from AI almost instantly. The new question is whether that content is any good.

Will AI engines like it? Will search engines like it? Will human audiences engage with it? Is it consistent with brand voice? Does it contain hallucinations or claims that create legal liability? Today, no model, tool, or platform answers those questions.

As a result, marketers are paying a tax. The bottleneck has shifted. It used to be generating content. Now it’s approving and verifying it.

Imagine your team has data on your AI visibility, and different people have different ideas. One person wants more blog posts for top-of-funnel education. Another wants clearer product attributes on product pages. Someone else wants posts on affiliate or social sites.

You select a few ideas and spend weeks designing the campaign and developing assets. Brand approvals take a few weeks, and legal approvals take a couple more. Then you ship the content and wait two or three weeks to see whether AI visibility moved. If it didn’t, you rinse and repeat.

Discovering what works takes months. Meanwhile, the environment changes daily or weekly. By the time you learn something, the AI models and the way they use search tools have changed. What you learned is already irrelevant.

So we started with a question. What if AI could grade itself and help verify content? How will AI answers change if I publish this? How will search engines respond? Will people find it useful? Is it on brand? Are there legal claims to worry about?

Our answer comes from our research. I’m a professor at Wharton, and my co-founder was at Google DeepMind, helping bring AI into search engines. We asked whether we could build simulators that act as verifiers. They simulate the real world and let you test before you deploy. That’s the core idea behind Bodhium Labs.

Say you want to publish a new blog post or edit existing product pages. You want to know whether it will improve your visibility in AI search. To answer that, we build a simulator for a brand or industry.

If you ask the target LLM a question and ask our simulator the same question, you should get similar answers. That means similar brands, in a similar order, with similar recommendations and claims.

Once the simulator matches the target LLM, two new possibilities open up for marketers. First, we can probe the simulator. How would the answer change without this Reddit post, this competitor post, or this affiliate post? That shows which content most influences AI answers.

Second, we can run counterfactual simulations. What happens if we publish a new blog post, edit a product page, or add content to an affiliate site? You don’t have to create assets, get approvals, and push them live. Our AI agents generate many ideas and test them in a simulation sandbox.

Strategies that were destined to fail will fail in the simulation. You never spend effort perfecting that content. The ideas that move the needle are the ones you publish. That dramatically reduces the time, money, and effort spent on content that would have failed, and it gets you to winning strategies faster.

We announced today that we’ve come out of stealth and raised $3.5 million. We work with large public companies, like Visa and Delta Electronics, as well as many startups. We’ve seen how the simulator takes guesswork out of content strategy and lets marketers test before they deploy.

For anyone attending today, visit bodhiumlabs.com/checkup or scan the QR code. Our team can build a simulator for your data and walk you through several content strategies it predicts will work.

Jeralin Hamann (24:58)

Very cool. Thank you, Kartik. A quick reminder: everyone will receive the slide deck and recording by email after today’s event. Now I’ll pass it to Shane from Evenbound to talk about avoiding the AI demand gen trap.

Shane Torrey (25:29)

Thanks, Jeralin, and congrats, Kartik. That’s a big deal.

A big part of building demand is building trust. You want buyers to believe your product, service, and brand will solve their challenges. Unfiltered AI can easily erode that trust if you aren’t careful with content creation, targeting, and outbound.

If you give AI leeway, it will take it. Context is the difference between output you can use and output you have to clean up.

HubSpot’s AI is getting good at filling in context for you. But here’s an example. I worked with a business setting up context for a prospecting agent and AEO trackers. They share a common name with a few businesses in very different industries. Based on the brand domain alone, AI decided they focused on AI for engineers, AI simulations, and estate planning. Estate planning is not at all what they do.

AI will fill in gaps with confidence. That puts the burden on us to be precise in the ask and clear about the audience. Always double-check what AI is doing.

Here’s another story. A specialty food supplier wanted to fill contact gaps for outbound. Many of their buyers don’t use big-brand domains, so tools like Apollo and ZoomInfo miss them. We built a custom solution.

Our initial target was restaurants and bakeries, but that wasn’t enough. With that loose ideal customer profile, about 41% of the contacts we pulled were the wrong fit. Many were corporate offices and chain organizations that would never carry their product. We were wasting usage, and we didn’t feel comfortable sending to those lists.

So we fixed the ICP. We did more research and added parameters. For example, we targeted restaurants and bakeries with higher price points on Google and Yelp, in top ZIP codes, with more four- and five-star reviews. AI wasn’t wrong by its own logic. It just needed the right context. Once we made those changes, the list looked like our actual buyer, and our outbound campaigns performed better.

Content is another big area. AI can match your voice, but it can still sound too polished. I recommend using AI as a sounding board rather than the author. Let it give you outlines and angles to react to, then fill in the gaps yourself.

Your voice is your stories and your opinions, sharpened. People come to your business for your unique take. If AI writes everything, you get words and rhythms no one says out loud. If you publish with little afterthought, it won’t sound human, and people will tune out.

Outbound needs care too. AI can scale outbound, but it can also scale your mistakes. Think about volume, personalization, and speed.

With the right context, you get more touches from the same team. Without it, you reach the wrong people faster, and your domain reputation pays for it. With personalization, good context gives you relevant openers at scale. Without it, personal details are generic or flat-out wrong. I get a lot of outbound offering to clean up our HubSpot portal, which wouldn’t be great for an agency like ours.

With speed, the right context helps you follow up on the right signals. Without it, you blast into the void before someone is ready, and you turn them off even faster.

Here’s how to keep AI on the right path. First, define the audience yourself. AI can help with research, but you should identify your ICP, not AI. Gather tribal knowledge from the people who talk to your customers. Learn what buyers care about, what objections they raise, and who isn’t a fit. Then give that to AI as a strict guardrail.

Second, be precise in the ask. Spell out the criteria, the exclusions, and the output you expect.

Third, train with examples. Show your AI what success looks like and what a fail looks like, and explain why. It will improve over time.

Fourth, keep a human in the loop. Before any list goes out or any content is sent, have someone review it. People want to hear from you, not from an AI system that almost sounds like you.

Finally, capture what lives in people’s heads. Sit down with your team and document how they sell and serve. Capture what they see that isn’t logged in your CRM. Transcribe conversations and organize them in a structured way you can use going forward. With that, I’ll pass it over to Tara.

Tara Montanez (32:44)

Thank you, Shane. We’ve talked about many ways to bring people into your GTM motion. Kartik covered AEO strategies, and Shane shared examples of AI in outbound. All of these signals are becoming very complex, especially as leads enter through many different systems.

AI agents have increased both the speed and complexity of GTM operations. Automation is more powerful, but it has made us all more dependent on clean data, clear ownership, and documented processes. We see that in our own UserEvidence survey of customers and prospects.

We’re also seeing failure points in vibe-coded lead routing and in routing built within people’s own tools. Here are the common places it breaks down.

First, duplicate and mismatched records. One lead becomes three contacts under two accounts. It’s the same buyer, but that’s invisible to the reps.

Second, wrong-rep routing. The same lead goes to the wrong rep, with no SLA and no explanation.

Third, broken account hierarchy. Information about which company owns another, and who should handle those accounts, quickly goes out of date. Signals then land with the wrong owner.

Fourth, handoffs with no owner. A great lead comes in, but it sits in a queue because no rule or agent says who owns it.

Fifth, enrichment that contradicts itself. Your CRM and marketing automation platform may disagree on the same account, and nothing reconciles it.

So what’s required for AI in go-to-market? First, a data foundation. GTM data must be clean, resolved, and continuously governed. Whether an agent or a person follows up, everyone needs confidence the systems are working from the right data.

Second, business process truth. Business rules must be documented and encoded so agents and people know what to do and when.

Third, shared customer context. Marketing, sales, and customer success all operate from one shared source of truth.

This leads to two approaches: probabilistic and deterministic. Probabilistic is AI reasoning. The same input can produce a different output each time. It finds patterns and interprets signals from context. That’s your creative side, and that variation can be great for personalizing each interaction.

On the other side, you need deterministic rules. That means repeatable execution, with the same input producing the same output every time. It’s consistent and reliable at scale.

How does that apply to lead management? On the probabilistic side, AI is great for judgment calls on messy, ambiguous signals. In routing, it can flag anomalies a fixed rule might miss. A standard list may not catch every variation of a job title, for instance. For handoffs, AI can apply contextual judgment to one-off exceptions. For example, it might recognize that a contact fits your SMB segment even if they’re not in the database.

On the deterministic side, matching requires exact precision you can audit. When someone asks why something happened, you have the answer. Routing requires the same lead, with the same criteria, to reach the right rep and the right action every time. Handoffs require SLAs, like how quickly your SDR team must follow up. They also require territory rules, like assigning leads by company headquarters, and governance to ensure company rules are followed.

LeanData is the backbone for AI GTM and lead management for many enterprise companies. Customers use us for lead management, and also at every stage of the funnel, from acquisition to adoption, retention, and expansion.

Finally, our customer and prospect event, OpsStars, is happening on Monday, October 5, in San Francisco. Speakers from Adobe, Atlassian, Bill.com, Visa, and PagerDuty will discuss their go-to-market motions and how they’re using AI. If you’re in the Bay Area and want to attend, I have a few free tickets for folks on today’s call.

Jeralin Hamann (39:00)

Thank you, Tara. We’ll send that out after today’s event as well. To wrap this all up, I’ll hand it over to Nick from Mobileforce to talk about turning AI-generated demand into revenue.

Nick Natale (39:21)

Hey, everyone. I’m Nick Natale from Mobileforce. I work in demand gen marketing, so it’s only natural for me to bring us home.

As sales and marketing people, we have a huge abundance of tools. We have AI tools and intent tools, like ZoomInfo, Apollo, and Clay. We also have website visit tools and tools to enrich the data in your CRM.

Here’s an example that will probably resonate. I recently saw someone on our SDR team copying and pasting each of their Fathom call notes into ChatGPT. They asked whether it was a good call and what the next steps should be.

That’s not a bad thing to do. However, some of that conversation data is leaking into those LLMs, and it’s fragmented. They then copied the output into another tab with their email, and data got lost along the way.

The fix is to keep your CRM as your main source of truth. Take all your data, whether it comes from website visits, ZoomInfo, Apollo, Clay, enrichment, or Reddit scrapers. Then create fields for it within the CRM.

That may seem like an extra step. But it saves you from everyone running their own individual demand gen and outreach. Otherwise, you end up with a fragmented team using fragmented tools, with some people on ChatGPT and others on Claude. The way they connect those LLMs to your CRM can also put data out there that you don’t want leaked.

HubSpot’s prospecting agent and Breeze are great tools, as long as the data is there for them to read and write. The number one thing you can do from an operations standpoint is make sure every MarTech tool you onboard talks to your CRM. Then use that CRM as your main source of truth.

Jeralin Hamann (41:59)

Thank you. Context comes up again here, so keeping everything up to date will be extremely beneficial. Thank you so much to all our speakers. Phil, I’ll kick it over to you for Q&A.

Philip Levinson (42:26)

That was amazing. Thanks to each of our presenters. Please drop any questions in the chat.

Let’s start with Brenda and Adam. Virtually everyone here has built, tried, or tested something with AI in the past year, involving their marketing, CRM, or selling process. Can you walk us through a couple of examples of what has worked best for HubSpot customers?

Brenda Lando Fridman (43:14)

Adam, I’d love for you to add an example. I’ll start with the bigger picture.

For those of us who remember the advent of the internet, we had no idea how our lives would change. This feels similar. We’re trying to see around the corner, and we just don’t know what it will bring.

Yamini, our CEO, talked about this in her INBOUND 2026 keynote, which you can find on YouTube. All of us are out there trying to build things, and most of it is a gigantic waste of time. Whether your company is big or small, stick to what you’re good at. Participate, but don’t spend all day building.

For example, HubSpot has 3,000 sellers. Those 3,000 sellers shouldn’t each build their own prospecting sequences. Instead, let certain hot spots build it, then scale it out so everyone benefits. Prospecting better through sequences is something we’re having success with.

We’re using Breeze Intelligence for much of what we build. It wasn’t great a year ago, and I heard that a lot. But I’ve also heard many people say they’re glad they gave it another try. Adam, anything to add?

Adam Wainwright (44:58)

Brenda, you nailed it. That’s exactly how we’re seeing Revenue Hub customers approach deployment. We’re learning best practices almost as quickly as we release new features.

Here’s a grounded answer. First, interview your sellers. Understand how they price and commercialize deals with customers, and capture that at scale through a series of interviews. That becomes the foundation for your knowledge base.

Second, gather your commercial history. That includes any contractual data in your ecosystem, from signed PDFs to any dataset that represents past deals.

Third, capture pricing and policy rules that your CFO or legal team enforces, and put them into a table.

Drop all of this into a knowledge vault, and start testing the questions AI might ask. From there, you can use HubSpot’s MCP to iterate on a guided selling flow. That flow then feeds a deal record, which creates line items in Revenue Hub.

Foundationally, understand what your sellers do and understand your commercial context and history. Building with prompts and MCP is actually very easy, but you need that good context first. Without that AI reality, you may build something that leads your sellers astray.

Philip Levinson (46:54)

That’s well said, Adam and Brenda. Kartik, let’s shift to the dark side. With AI, it’s much easier to generate content and implement tactics. What risks do companies face with a spray-and-pray model? Do you have an example of guiding a client toward a better practice?

Kartik Hosanagar (47:31)

You had me nervous when you said “the dark side” and then said my name.

On spray and pray, this has been the problem for decades, and certainly in SEO and GEO. Marketers are dealing with black boxes they don’t understand. Without visibility, there’s been no alternative but to spray and pray and figure out what’s working.

That has a few costs. First, teams launch several campaigns at once and can’t tell which one moved the needle. I spoke with a marketer this morning whose team made four or five changes. One vendor said its schema changes made the difference, but internally, the team couldn’t tell. That creates uncertainty, a lack of clarity, and wasted time and effort.

Second, spray and pray no longer makes sense when the environment changes this quickly. We hear about model updates every three to four months. However, changes happen every week, even when the training data and underlying model stay the same. AI engines keep changing how they use search tools.

So whatever you and your competitors are doing is unlikely to produce results with a spray-and-pray strategy. You need a more disciplined approach. Open up the black box and understand why the AI answers the way it does.

When my co-founder and I started, he was at DeepMind and I was at Wharton. We discussed the enormous scientific rigor and computational power behind building LLMs. Marketers don’t have any of that. No one has applied that same rigor to opening up the black box. That’s the solution to spray and pray.

Philip Levinson (50:33)

That makes a lot of sense. Thanks, Kartik. Now let’s go to the light side with Tara. You’ve worked with clients who tried things with AI and struggled, then guided them toward more efficient practices. Beyond probabilistic and deterministic, are there rules-based best practices you implement with clients?

Tara Montanez (51:13)

I loved the HubSpot team’s comments on build versus buy. We’re seeing the same thing. Teams need a standard structure and trusted software. No one wants to maintain something that has already been built.

To answer your question, it comes back to a strong foundation, clear workflows, and clear context for everyone on your team. What that means varies by company, and we work with customers on those structures. I hate repeating myself, but we keep coming back to those fundamentals. We look at where to use AI versus standard business rules and find the right mix for each company.

Philip Levinson (52:11)

That makes sense. Before we wrap up with Tara, OpsStars is next week. Should people message you on LinkedIn if they have questions?

Tara Montanez (52:22)

Yes, absolutely. Feel free to scan the code to visit the website, ops-stars.com.

Jeralin Hamann (52:40)

Now we’ll head over to Evenbound. Shane, we talk about AI with clients every day. What are you hearing from mid-market companies about AI adoption? Is it still excitement, or has it turned into frustration with the outputs?

Shane Torrey (53:18)

It’s a good question. Many companies have mandates to implement AI, especially if they have investment backing. Many are still excited. There’s a lot of untapped potential, and it’s still the wild west. People are hopeful about what AI can do.

However, it hasn’t always delivered the outcomes they hoped for. Part of the reason is that many companies didn’t define an outcome before they started building.

When we build an agent within HubSpot, or something custom, we start by asking what outcome we want to achieve. Then we assess feasibility. Do we have the right data? Is it clean enough? Many manufacturing companies we work with have decades of data, and much of it is no longer good. I wouldn’t want to start building on top of outdated data.

That’s where frustration comes from. People don’t always understand what it takes to build something repeatable, usable, and consistent. Hallucinations can throw AI off. We also want a strong case for adoption, because teams will probably stop using AI if it isn’t good right away.

There will still be a lot of focus on AI. But it may help to slow down a bit and build the right things, with more structure and a human element to remove friction.

Jeralin Hamann (55:11)

Nick, let’s wrap up with you. Teams are generating leads faster than ever, but they aren’t closing as fast as they generate. Where does that gap show up first?

Nick Natale (55:36)

The biggest place is speed to lead, or the SLA for SDR follow-up. You’ll see the task list grow and not get completed fast enough. The other signal is that leads get disqualified faster than before.

The biggest fix is realigning with sales. We have an abundance of tools, new contacts, and leads, but that doesn’t mean every one is qualified. Now we can deliver more qualified leads to sales and use AI and automation tools to nurture the rest along the way.

Jeralin Hamann (56:15)

Thank you so much for joining us during your busy afternoon, especially on close day. You’ll receive the slide deck and recording after today’s session. The final slide includes links to contact each panelist, and their LinkedIn profiles are hyperlinked. You can also join our LinkedIn group, where these panelists are members too. Thank you again, and we’ll see you at the next one.