By Phil Pelucha
McKinsey’s 2026 B2B Pulse Survey, roughly four thousand buyers and sellers across thirteen countries, found something worth paying attention to. High-growth firms are three times more likely to have raised their AI spend by double digits, seventy-one percent versus twenty-five. In financial services specifically, McKinsey found the firms that rewired prospecting and relationship management around agentic AI saw revenue per relationship manager climb by up to fifteen percent, with cost-to-serve falling by up to forty. And yet fewer than ten percent of organisations have actually scaled AI in any function at all. McKinsey and BCG published almost identical findings on the same day in July this year.
That gap, between what the winners are doing and what almost everyone else is stuck on, is the entire story.
I think about agentic AI as building a sales cyborg, not replacing the salesperson. AI does the heavy lifting, research, scheduling, qualifying, the mechanical parts of the process, so the human is freed up for the part that genuinely cannot be outsourced. Reading between the lines. Understanding the actual problem, not just the problem the prospect thinks they have. Any salesperson who’s been in the industry more than five minutes will tell you those two things are rarely the same.
Here’s a real example. A business tells you their problem is lead generation, they need more leads. Nine times out of ten, it isn’t. Say they’re getting a hundred leads and closing twenty percent. The instinct is to chase two hundred leads instead of improving conversion on the hundred they already have. But doubling lead volume usually makes conversion worse, not better, because now your sales team is stretched across twice the calls and meetings, with less time for the research and detail that actually closes a deal.
That diagnosis, is this really a volume problem or a conversion problem, is exactly the kind of judgement an agent should never make alone. Let AI qualify the hundred leads, surface the pattern in the data, even draft the first follow-up. But deciding what the client’s actual problem is, and having the conversation that uncovers it, stays human. The moment an agent is running that judgement autonomously instead of a person, you’ve automated the wrong part of the job.
So why are so few businesses actually scaling this, when the upside is this clear? I see the same two reasons every day, and neither of them is really about technology.
The first is anxiety. Companies feel they need an AI policy just to avoid falling behind, but they’re paralysed the moment they try to choose. It’s like being handed a menu written in a language you don’t understand and told to pick something you’ll enjoy. You don’t know what you don’t know, and that uncertainty is precisely what keeps most businesses stuck. The appetite to implement AI is real, it’s just matched exactly by the fear of not knowing where to begin.
So, here’s what I’d tell a South African B2B business to do instead. First, look inward before you look outward. Find your actual inefficiencies, the specific places you’re losing time or money in the sales process, before you go shopping for a solution. Second, read the case studies, but as inspiration for what’s possible, not a shopping list. The goal is one good idea, not everything on the menu. Start small, start specific. Don’t commission an expensive, business-wide “get our data AI-ready” exercise before you’ve even proven the concept on one real problem. That’s not thorough. It’s just an expensive way to stay stuck at the pilot stage, along with the ninety percent of everyone else.
Phil Pelucha is a Revenue Architect with a PhD in AI integration
