Are we ready for AI sales reps?

RB2B replaced a sales-assist team with an AI avatar. Here’s what happened.

👋 Hi, it’s Kyle Poyar and welcome to Growth Unhinged, my weekly newsletter exploring the hidden playbooks behind the fastest-growing startups.

---

I’ve been fascinated by emerging AI use cases for go-to-market. Folks are vibe coding ROI calculators, quickly spinning up lead magnets, hiring AI agents for marketing, and running signal-based GTM plays. The use cases are endless, and a bunch of fun.

But very few are suggesting AI as a replacement for salespeople. Remember, for example, how Salesforce said they’re actually hiring 2,000 more salespeople to sell AI?

Well, there’s an outlier: Adam Robinson, the outspoken founder and CEO of RB2B, which helps de-anonymize website traffic to convert them into high-value leads. Adam has grown RB2B to $5.5M ARR with three people (down from five at the beginning of June). His goal is to bootstrap to $10M ARR as profitably as humanly possible.

Adam recently ran an experiment to replace RB2B’s two person sales-assist team with a $99 per month AI avatar trained to respond to inbound requests from prospects. The AI avatar, which is also a clone of Adam, had 1,683 (!) conversations in its first 22 days, and with surprisingly good answers. I sat down with Adam to get the full unfiltered story of what happened.

---

An on-demand rep for anyone who wants to ‘contact sales’

Meet the Adam Clone. It’s an AI avatar that lives on the public-facing side of RB2B’s website, designed to handle what a sales-assist rep would normally do: respond to inbound interest, answer common questions, and move people toward purchase. Think of it as an AI-powered “Contact Sales” button available 24x7 without the two-day wait for a reply.

This is distinctly not customer support chat. Adam draws a hard line: once someone becomes a customer, they interact with Fin (Intercom’s AI, profiled earlier), which is better suited for RB2B’s troubleshooting and post-sale help. The Adam Clone was made with Delphi and exists purely to accelerate new sales conversations, especially for the prospects who already have an affinity for Adam through his LinkedIn #influencer content.

So what does the avatar actually handle? 90% of what buyers ask falls into six topics:

• Can I get a demo?

• How does pricing work?

• Do you have SOC 2 docs?

• Will I need to update my privacy policy for GDPR?

• Does this work outside the U.S.?

• If I have 10,000 visitors, how many credits do I need?

Here’s an example of what this looks like in practice:

Prospect: I blew through 1000 credits in 4 days. I need custom pricing.

AI avatar: Alright, if you burned through 1,000 credits in just 4 days, that’s a pretty heavy usage rate. Let’s break it down:

1. Estimate Monthly Usage: If you’re using 1,000 credits every 4 days, that’s roughly 250 credits per day. Multiply that by 30 days, and you’re looking at around 7,500 credits per month.

2. Pricing Overview: For RB2B Pro, pricing scales based on the number of credits you need. You can check the detailed pricing here: [LINK]. For bulk credits (like 7,500/month), you’ll get a discounted rate when you go annual. If you’re unsure about committing to annual, you can always start monthly and adjust as needed.

In the first three weeks, the AI avatar handled 304 demo requests, 98 SOC2 docs, and 16 custom pricing quotes. (As an aside, it can’t actually deliver a demo yet. The avatar currently sends people to a pre-recorded demo video on YouTube. Adam would love to see the AI avatar ask qualifying questions, then personalize the demo experience for each prospect.)

---

Rethinking a broken sales-assist motion

RB2B’s growth motion starts with Adam’s LinkedIn content to feed top-of-funnel. They convert prospects with a freemium, product-led experience. And they offer a layer of sales-assist for higher-touch prospects.

Under the hood, things were breaking down. “We thought we had an enterprise motion,” Adam told me. “Turns out, we didn’t.”

RB2B initially tried to go upmarket, chasing bigger logos with more traditional sales cycles. But it quickly became clear that others were better suited to serve that segment. RB2B lost or churned most of those deals.

Instead of forcing an enterprise motion that wasn’t working, Adam refocused on a different goal: make it dead simple for anyone to start using RB2B. That meant building for speed and self-serve conversion.

Sales-assist was supposed to help with that. But even that didn’t hold up to scrutiny. “We kept asking: is sales-assist actually incremental? And there was no convincing data that it was.”

So Adam made a provocative bet to remove the human sales-assist role and replace it with a bot. To be clear, this is part of Adam’s ethos. He believes a lot of founders should have a goal of hitting $10M ARR with as few people as possible. “If you can get there, you can do whatever you want with your life,” he said.

RB2B had previously placed an early bet on Fin (Intercom’s AI), and while it flopped at first, it’s now 5x better than it was 18 months ago and successfully resolves 75% of support requests. Combine that with a high-volume funnel – 80 signups a day, 10–15% converting at $149/month — and the downside of getting a few interactions wrong was fairly low.

It’s worth contrasting this with the approach taken by Calendly, which I profiled last year. Calendly went the route of investing significant time and money into sophisticated lead qualification and routing, sending prospects to a combination of on-demand demos, the sales-assist team, velocity sales, or enterprise sales. While their results were great, the level of complexity involved doesn’t necessarily make sense for a three-person business like RB2B.

---

Promising results, but not mind-blowing (yet)

Four weeks into replacing a human sales-assist rep with the Adam Clone, RB2B’s metrics look… almost identical.

“There haven’t been any major leading or lagging indicators that changed,” Adam admitted. “It hasn’t been a massive leap in anything.”

The AI clone handled about 1,700 inbound requests in 22 days. It asks for an email address after someone asks a certain number of questions (about 10% of conversations capture an email). Adam set up a Zapier automation to ask for feedback immediately after folks chat with the bot, and 70% say the experience is positive. Almost none of the AI answers were flat-out wrong.

Adam sees the AI avatar experiment as a high-leverage replacement for two people with a bonus of allowing him to capture more product insight directly from customers.

“At our stage, we want sales to do two things: usher the deal along and gather product feedback.” And now, instead of a rep on the phone chasing down those insights, it’s Adam himself reviewing every AI conversation (roughly 50 a day). He spends 30-60 minutes tweaking bot responses, refining answers, and updating how the system handles edge cases. Like what happens when someone has 10,000 website visitors? Turns out most customers exclude pages like their blog or careers page. Now the bot knows that, too.

---

Your documentation makes all the difference

If there’s one takeaway from Adam’s AI sales-assist experiment, it’s that your support documentation is the entire game. This includes both the content and how it’s structured for the bot.

“Missing info leads to hallucinations or escalations,” Adam told me. “If you want this to work, you need to obsess over the docs.”

As an aside, the AI avatar is explicitly not trained on Adam’s LinkedIn content. “When it was, the voice was perfect, but the answers it was giving were wrong.”

If you’re starting from scratch on documentation, Adam’s advice is to commit to writing one new support doc per day based on the most common question from users. And rather than answer someone’s question directly, write the doc and send it to them. Now you’ve scaled the answer.

Here are the specific categories Adam recommends building out (you can view the RB2B support docs here to see an example):

• Getting started and onboarding

• Product features and functionality

• Billing, plans, and account management

• Troubleshooting and errors

• Security, privacy, and compliance

• Integrations and API access

• User management

• Analytics and reporting

• FAQs and common objections

RB2B also uses a custom GPT to optimize support docs for both human readers and AI agents. That’s step one. From there, it becomes a cycle: let users surface the gaps → write better docs → run those docs through AI to QA the formatting → refine based on real conversations.

And critically, it’s important to set guidelines for the AI. Tools like Delphi let you specify that vague questions should be met with clarifying prompts rather than guesswork. Over time, Adam has moved settings from ‘strict’ to ‘medium creative’, now allowing the bot some freedom to provide an answer for things not explicitly in the support docs.

“It’s a constant loop of documentation and refinement,” Adam said. “But if you get that right, the AI will feel like magic.”

---

The TL;DR: Is it time to rethink the B2B sales experience?

It’s too early to call this an unqualified win, but the results are positive. And Adam’s bet is that his big investment in support docs and QA will continue to pay off over time.

“I am so pumped about what this means for bootstrappers,” Adam told me. “Smaller teams will get more done and spend time on the highest-quality human interaction only.”

My take: I’m not sure AI is ready to be a replacement for sales. But it can help augment the sales motion, filling the gaps left behind between fully self-service versus fully sales-managed. Whichever way you go, it’s probably time to take a fresh look at your documentation.