AI's Real Sales Payoff Shows Up In Faster Follow-Up And Cleaner Data

The loudest promise in sales AI right now is replacement: AI SDRs, autonomous outreach, and fully automated selling. That's also where the technology is falling shortest of the hype. The gains that are actually landing in revenue organizations come from compressing follow-up time, automating administrative overhead, and scaling work that human reps were already doing well. AI is proving its value by removing the bottlenecks that used to cap what good salespeople could accomplish, and the leaders getting the most from it are the ones who thoroughly understood their sales process before they automated any of it.

David Baker is the Chief Revenue Officer at vendor compliance management platform VIVE, where he leads revenue strategy for a heavily event-driven, technology-forward sales organization. His approach is shaped by hands-on experience building and testing AI workflows himself rather than delegating the evaluation, which gives him an ultra-clear read on where the tools deliver and where they collapse under their own marketing.

"AI is a great tool for sales leaders who know how to lead sales. It's not a great tool for tech leaders who decide they want to sell," he says. That assertion captures Baker's core thesis that AI amplifies competence and exposes the lack of it, and nowhere is that more true than on revenue teams. 

The hype is outreach, but the gains are everywhere else

Baker is direct about where the noise is loudest. He estimates that he fields around 40 messages a week from AI companies pitching automated SDR outreach, and he places that squarely in the hype column. "As someone who has a lot of experience with some really high-end voice models, I believe there is a place for voice models and AI. But for cold outreach, it's just not there yet," he says.

Where the value shows up is in the operational layer that used to be a bottleneck. Baker's team works roughly 40 events a year, and the follow-up crunch after a large show was historically the constraint on how much of that pipeline converted. "How do you go to one event with 14,000 people and then three days later turn on another event? That's a really condensed follow-up time frame."

His solution is AI video at scale. Ahead of a show, he utilizes a third-party partner for outreach via email and LinkedIn. Then, while riding in an Uber after the show, a rep writes up a script, loads a list of contacts, and sends a personalized video that inserts each recipient's name and company on the fly. "One video takes one minute for us to send to 100 people. It's a custom LinkedIn video with a much higher reply rate than we ever got by trying to just throw it into an email campaign," he notes. 

The important distinction is that AI isn't doing the selling. It's removing the administrative ceiling that used to limit how fast and how personally a human team could follow up.

AI is only as good as the data underneath it

Baker keeps AI running on the data side, using it for forecasting, customer signals, and intent, but mostly so he can benchmark it against his own judgment. His skepticism centers on a point that gets lost in the enthusiasm for intent features and signal scoring. "AI is only as good as the data you have in your system. Just like it wasn't hard to game a CRM in years past, if my rep knows their weighting is based off the confidence of AI, well, AI has no clue if your champion went cold on you four weeks ago. It just knows if you marked it high or low."

The point is that AI signals are only trustworthy when every other input is clean enough to support them. Absent that discipline, Baker says, the model produces confident output built on stale or gamed data, which is worse than no signal at all. 

Automation without qualification just scales the garbage

Baker experienced the volume trap firsthand when he first partnered with an outreach vendor. The early results looked impressive at the surface level and fell apart on inspection. "It turned out that 80% weren't our ICP, and we would have had better time spent just not doing this exercise," he shares.

The fix was tightening the inputs before scaling the outputs, using tooling to identify each prospect's software environment, and scrubbing the lead list down to the exact titles worth contacting. His framing of the tradeoff is a direct rebuttal to the wider-net theory of automation. "I'd much rather reach out to 600 people with a 30% success rate than 5,000 with a 3% success rate. Even if the total number is more, it's not worth it if it takes three times longer to get there."

More volume is only a win if it doesn't bury the team in unqualified leads that erase the productivity the automation was supposed to deliver.

The danger of confidently wrong answers

Baker's sharpest caution is about what happens when someone deploys AI to solve a problem they don't already understand. He's comfortable rebuilding his own CRM workflows live, mid-conversation, precisely because he knows how to verify the result. "I wouldn't have the confidence to do that if I wasn't positive how to build it on my own to go back and verify it did it correctly. If you don't know how to do it and you're seeking the absolute truth, it's a very scary thing, because AI will confidently tell you 'This is the way to do it.'"

He watched the failure mode play out with a company he consults for, where demos looked spectacular but close rates were cratering. "They had competing automations set up, so anytime they sent through email, it was logging that as a demo inside the CRM. But also when the client did the recording, it was logging as a demo, duplicating the numbers. Their close rate didn't go anywhere. They were looking amazing on demos for three months and coaching to the wrong behavior."

Automating a process nobody fully understood hid the truth and trained the team to optimize the wrong thing.

The tool rewards people who already know the job

Baker's closing point is that AI belongs in the same category as the productivity tools that came before it. It's powerful for someone who understands their craft, but useless as a substitute for that understanding. "Just like PowerPoint and Excel, it's a tool to help somebody who understands their job do it better. It is not a tool to help you fake your way into a job, and that's what it's being used for."

That principle shapes his read on which revenue roles change and which endure. He sees RevOps compressing significantly, with fewer multi-person committees managing CRMs and integrations, because a knowledgeable leader with AI can cover that bandwidth. But the human core of selling stays exactly where it is. "Where AI is never going to take over is somewhere where I'm asking someone to trust me. If I'm delivering pricing, possible bad news, possible product upgrades, I'm not going to rely on AI for that. I need to feel out the actual room."

His closing caution is for anyone expecting AI to rewrite the fundamentals of a deal. It changes the bandwidth required, but not the nature of the sale. "It's not going to necessarily shorten the sales cycle. It can shorten the amount of bandwidth needed so you can get more sales in that shorter cycle. But if your sales cycle is nine months, that's your buyer committee, that's your ICP. People looking for AI to take a nine-month sales cycle to three, that's just not realistic today."