ITWeb contributor Phillip de Wet.
Ever since we learnt that GenAI is failing to deliver any measurable return on investment (ROI) in the enterprise, there’s been a subset of vendors doing what bad vendors do: selling a technology patch to a business problem.
The specific patch that comes up most often is MCP, the Model Context Protocol, which absolutely holds the promise of a new golden age while, simultaneously, being peddled as snake oil.
The promise is that you find data, any data, put an API on top of it, slap an MCP server in front of that and… magical AI stuff will magically happen.
It is bad enough that I’ve come across developers who have had screaming matches with customers demanding MCP integration where it can’t possibly make sense, and executives now sitting with a bunch of MCP servers that contribute nothing to their organisations except a funky new attack surface.
The good AI and AI-adjacent vendors – of which there are many – are helping organisations become nimble. They are helping track down the forgotten unstructured data repositories, building the systems and processes to expose that to agents, and setting up the infrastructure to quickly deploy whatever agents they may find useful.
However, these vendors do not know what companies should be doing with those tools. That is the job of consultants.
The terrible consultants – of which there are many – are telling executives that they are behind the curve and need to get AI into every process and every system right this very second. Bolt it on, throw it in, you can pivot and iterate later, just go go go! (Maybe spin up some MCP servers, that’s a quick win.)
I have a working theory that there is a conspiracy of silence around the one place where there is massive ROI on AI, because that is through shadow AI. Individual employees want to keep the credit, or the time they save, operational units don’t want hiring freezes because they report productivity gains, and the executives who themselves use shadow AI don’t want to admit it exists.
If so, the hard path towards AI gains is to study the shadow AI operating everywhere, and bring it into the formal fold while enhancing capabilities, rather than crippling the official version through security restrictions.
The good AI and AI-adjacent vendors – of which there are many – are helping organisations become nimble.
Which is a whole mess of culture and policy and strategy that can bog you down for years and, let’s face it, will probably not have all that much impact. Also, you’ll end up with some nonsense measurement of how many tokens are used, or how much time is saved, rather than impact on the bottom line.
Or you could just zoom in on the place where the money is actually made.
Sales teams are whispering about some jaw-dropping results from early efforts in agentic AI. Not from polished, highly-tuned models to handle dynamic pricing, and not from high-risk, customer-facing stuff such as chatbots.
Instead, these teams are combining unstructured data such as e-mail with customer records to find long-ago lapsed subscribers that could be worth approaching – and handing off the initial approach to AI. They are digging into the very top of the funnel to find leads that never made it into pre-qualification but that are worth addressing. They are, God help us, trying to scrape LinkedIn to quickly identify new key hires at potential customer organisations, so they can hit them up during their initial period of vulnerability.
So yes, the tactics are not always pretty. This is sales, after all.
But the results are hard to argue.
Some of this is via rogue, cobbled-together systems, but there is rising buzz about the likes of Salesforce’s awfully named Agentforce Lead Nurturing and the HubSpot prospecting agent, especially when combined with an AI browser such as Comet or Atlas.
These are one-year-old agents combined with zero-year-old browsers, and some smaller organisations are claiming big sales volumes on deals only touched by humans right at the end of the process. When the big enterprises finally start publishing case studies, the numbers are going to be big.
Sure, maybe you should reengineer your entire business to take advantage of AI, and maybe you should replace all your humans with agents, and maybe you should entirely rethink your IT infrastructure.
In the meantime, though, you can throw some AI at sales and possibly make some decent money.
