5 Honest Lessons on How AI is Actually Rewiring B2B Sales (From Salesforce, Momentum, and Mangomint)

We put three very different revenue leaders on stage at SaaStr AI London: Greg Beltzer CCO of Salesforce Agentforce (former CIO), Ashley Wilson COO and co-founder Momentum (AI-native B2B company), and Marchelle Mooney from Mangomint (VP of Sales running a very SMB vertical B2B org selling to salons and spas).

They had wildly different contexts. Salesforce has tens of thousands of sellers. Momentum is a 50-person AI startup. Mangomint’s AEs close 20 logos a month at SMB ACVs. But they kept arriving at the same conclusions. Here’s what actually matters right now.

1. Your AI Rollout Will Be Bumpy. That’s Normal.

Greg was blunt about this: the Agentforce rollout at Salesforce itself has not been smooth. Data wasn’t clean. Processes weren’t ready. Ways of working had to be rewritten on the fly.

This is Salesforce we’re talking about. The CRM company. If their data wasn’t in a perfect place for AI, yours probably isn't either.

The key insight from Greg: AI is much easier to deploy on the service and ops side than on the sales side. Why? Because service and support workflows already have structured data and defined processes. You’re already doing those activities with humans in a documented way. Sales is different. You’re relying on individual sellers, and most of what they know lives in their heads, their text messages, or nowhere at all.

If you’re starting your AI journey, start where your data is cleanest. For most companies, that’s support or operations, not sales.

2. Every Company Has Leads They Never Followed Up With. AI Fixes This Immediately.

This was the single biggest “everyone in the room nodded” moment.

Greg admitted that Salesforce gets a massive volume of inbound web leads and a shockingly low percentage were being followed up on. Reps cherry-picked the ones they wanted. The rest got a fake loss reason with no notes in the record. Sound familiar?

Amelia SaaStr’s Chief AI Officer shared the same experience at SaaStr AI: post-Annual, hundreds of leads from people who literally raised their hand to sponsor, and the sales team just… didn’t call them. Not because they’re bad people. Because humans prioritize, and the bottom 40% of leads always gets ignored.

This is the single easiest AI use case in sales right now. Put an agent on the leads nobody is working.

Not because the agent is going to close a $50K deal. But because it will follow up consistently, enrich the data, qualify or disqualify, and surface the ones worth a human’s time.

Greg’s number: the leads worked by agents at Salesforce led to direct revenue they would not have had otherwise, because nobody was working those leads at all.

If you do nothing else with AI in sales this year, do this.

3. Your CRM Data Is Worse Than You Think, and AI Will Show You

Marchelle had the most honest take on this. When she started rolling out Momentum and other AI tools, she realized her Salesforce instance was a mess. Deals were closed with no notes. No call logs. No contact records. Her marketing automation platform had 10,000 more data points than Salesforce, even though Salesforce should have had more (since it gets both marketing and sales data pushed into it).

Her top rep was closing 35 logos a month. Absolute machine. But everything lived in text messages. No notes in the CRM. Another rep literally had to explain to his wife why he had so many women in his text messages because he’d never sold to salon and spa owners before. These are real humans doing real things that don’t show up in your system of record.

Here’s the problem: if data never existed as a record in your CRM, your AI agent can’t go back and find it. It’s not going to magically reconstruct what happened. So the first step for many teams isn’t deploying an agent. It’s deploying tools (like Momentum, Gong, or whatever fits your stack) that ensure data actually makes it into Salesforce going forward.

Marchelle’s advice: pick your AI tools based on what will ensure everything flows back to one source of truth. Then build agents on top of clean data.

4. Don’t Raise Quota Yet. Use AI to Lift Your Middle Performers.

Marchelle announced to her team that she is not raising quota in 2026. She could have. Her top performers are closing 35 logos a month against a 20-logo quota. That’s a 7x ARR-to-OTE ratio in very SMB SaaS.

But she chose a different strategy: keep quota flat and use AI to push the middle of the bell curve up toward top-performer levels. Her logic is sound. We’re not at the point where AI is truly native in most sales workflows. If you raise quota now, you’re asking reps to do more before you’ve actually given them the systems to do more. You’ll just create frustration.

Ashley reinforced this from her research for a report Momentum published. Multiple revenue leaders, including Kyle Norton at Owner.com, said the same thing: the companies winning with AI in sales are investing heavily in ops and rev ops headcount to run agents, manage data, and architect systems from top of funnel through post-sales. The efficiency gains only happen if the business supports the reps with all the underlying infrastructure.

Ashley’s direct quote on the “two reps to $100M” narrative: sure, that works if you’re Lovable. But most of us don’t have that business. Holding yourself to that standard is unrealistic. Better to invest in the systems that will compound over time.

One exception: if you’re an AI-native startup and everyone is already crushing a low quota, you probably have the wrong quota. One founder Ashley spoke with had a head of sales running the “industry standard playbook” at a company where every rep was hitting quota easily. That’s not a sign of a great team. That’s a sign the bar is too low. First principles matter more than borrowed playbooks.

5. AI for Customer Retention Is the Underrated Play

Most companies are focused on AI for top-of-funnel or support chatbots. Almost nobody is using AI seriously for retention. Ashley thinks that’s a mistake.

Customer success teams have been underserved in tooling for years. The expectation is still high on them to get data into Salesforce or Gainsight, but the tools haven’t matched the ask. Meanwhile, churn is usually a slow drip, not a sudden event. A customer doesn’t wake up one day and leave. They have three unanswered support tickets, a feature request that went nowhere, and a quiet conversation where they mentioned they’re “evaluating options.”

With AI listening to those calls and flagging dissatisfaction signals in real time, CS leaders can actually intervene before the renewal conversation turns into a save attempt. Ashley shared an example: a customer doing 3-hour support calls in another language, then having to manually write notes into Salesforce. Now that’s automated, translated, and searchable. The time savings alone change what’s possible for a CSM’s book of business.

Greg added context from financial services: 40% of customers want more outreach from their vendors, but they want it to be relevant and personalized, not a drip campaign. That’s exactly what an agent can do. It builds the relationship digitally, one small interaction at a time, so the human can close it when it matters.

A Few More Quick Takeaways From the Panel

AI is easier to sell into organizations than it was two years ago, but internal resistance from legal, CIOs, and works councils is still the #1 blocker. You need a champion inside the company raising the flag. It doesn’t happen by just giving people a Claude license.

Your top seller will never put notes in the CRM. Stop fighting it. Deploy tools that capture the data automatically. The personality type that closes 35 deals a month is not the personality type that writes call summaries.

AI-generated outbound sequences work, but video messages on LinkedIn still break through the noise better than anything automated. Ashley’s team in Argentina uses AI to write sequences, but the human-recorded videos are what get the meetings.