The Real AI Risk for MSPs: Who Verifies the Output When Clients Don’t Ask?

[00:00:02] AI as the platform layer consolidates and its own vendors push into the channel. The MSP's only defensible billable position is the layer above the tools, orchestrating which model does what and verifying that what it produces can be trusted. This is the Business of Tech. I'm Dave Sobel. Two sets of numbers moved across the channel, and on the surface they belong to completely different conversations.

[00:00:28] Start with Ninja One. The RMM platform announced a secondary funding round that more than doubled its private valuation to $12.3 billion. The company is still founder-led, still profitable, and still debt-free, and have reported roughly 70% year-over-year growth. A platform built to serve MSPs is now worth more than $12 billion, and it got there while making money, not burning it.

[00:00:54] The Pax8 moved in the same direction. The distributor expanded its marketplace with a batch of new tooling for partners. Subscription insights, a partner analytics hub, customizable finance-ready reports. All of it built to pull more of how an MSP actually runs its business onto Pax8's Rails. And alongside that, Island, the enterprise browser company, launched its small business platform directly on the Pax8 marketplace.

[00:01:21] One more vendor, landing on the same rails, reaching partners and their S&B clients through the same front door. Now the second set of numbers, and they point somewhere else entirely. Glean, the enterprise AI company, surveyed 6,000 full-time digital workers across the US, UK, and Australia, and found AI is saving them around 11 hours a week. But it's quietly taking most of that back. Workers now spend an average of 6.4 hours a week,

[00:01:50] nearly a full working day, on what Glean calls bot-sitting. Feeding the AI the context it's missing, checking its outputs, and cleaning up the confident but wrong answers it leaves behind. And when that work goes untracked, people stop doing it. More than two-thirds admitted to shipping AI work they hadn't checked, didn't fully understand, and couldn't defend if asked. And then Avanti put a number on the control problem.

[00:02:19] 85% of IT teams say every AI agent in their environment has a named owner. Only 42% actually do. A 43-point gap between who's supposedly accountable and who actually is. So that's the picture. The platforms underneath the MSP are getting bigger, richer, and more concentrated. And the AI running on top of them still can't be left alone in the room.

[00:02:45] If you're listening to this and you haven't hit follow yet, on Apple Podcasts, search Business of Tech. It takes five seconds and you'll get the next episode automatically. Quick note before we get into it. The SMB Online Conference is June 23rd through 25th and registration is $129. Three afternoons, noon to 3.30 Eastern. 12 sessions. Pricing. Service delivery. AI for the independent operator. M&A. Private equity.

[00:03:15] All practitioner speakers. Zero vendor keynotes. Jay McBain live. Brendan Ballew on the private equity playbook. An M&A panel run by operators, not buyers. Every session recorded. So even if a client call wins, you miss nothing. Small Biz Thoughts members are in free. Everyone else? 129 at smbonlineconference.com. That's smbonlineconference.com.

[00:03:43] Now, let's get into the show. Underneath both of those is one shift. The model. The actual engine every one of these tools is built on is turning into a commodity. And when the engine becomes a commodity, the value doesn't vanish. It migrates to the layer that wraps around it. The companies that make the model have figured that out before the channel has. Start with the share numbers.

[00:04:09] TechCrunch, citing SensorTower, reported that ChatGPT's share of the AI market slipped below 50% for the first time. Down to 46.4% as users spread out across Google's Gemini, Anthropics Claude, and a widening field of competitors. No single model owns the market anymore. They're becoming substitutes for one another, which is another way of saying the model itself is also becoming interchangeable. Also group named the shift directly.

[00:04:38] One of Europe's largest technology distributors, through its group lead for vendor ecosystem development, Mark Appleton, argued that AI success won't be decided by how sophisticated the model is. It'll be decided by ecosystem preparedness. The skills, the governance, the wiring built around the model. Strip that down, and it's a simple claim. The model is no longer the thing that separates a winner from a loser. Here's where it turns.

[00:05:05] If the model is becoming a commodity, the people who own it can't sell it to enterprises one deal at a time and call that a business. They need volume, and the channel is where volume lives. So OpenAI did exactly what Anthropic had already done. It launched a partner network, a $150 million program aimed at consultants, system integrators, and managed service providers, to push its models out through partners instead of around them. Read that carefully.

[00:05:35] The company that makes the model just walked into the reseller's spot and sat down. And that's the move that quietly rewrites the math for the MSP. When the model is interchangeable, and the company that makes it is now your channeled competitor or partner, reselling the model is the one position that's already gone. Whatever is still worth paying for has to live above it. So what does living above the model actually look like inside a client environment?

[00:06:04] And what does it cost the MSP who skips it? Let's start with the bill. Asana surveyed UK IT leaders and found that more than 8 in 10 had been hit with unexpected AI costs. Spend nobody forecast surfacing after the fact. And the same survey found that more than half had already seen an AI tool or agent take an action that caused financial, legal, reputational, or compliance harm.

[00:06:30] The AI making a call or touching data in a way nobody authorized. That's AI running inside a business with no one deciding what it's allowed to do and no one checking what it produces. And the bill, financial and legal, landing on whoever's responsible for the environment. For most MSBs, that's the MSP. Govern none of it, and you're not avoiding the work. You're absorbing it for free after it's gone wrong. Now the size of the opening.

[00:07:00] Deloitte, in its State of AI in the Enterprise report, found that only one in five companies has a mature way to govern autonomous AI agents. Which means roughly four in five are running software that acts on its own with little real control over what it does. Sit with that. The agents are showing up in client environments faster than anyone has built a way to watch them.

[00:07:23] That gap between the AI that's deployed and the AI that's actually governed isn't a problem to point at. It's the exact width of the work the MSP can step into. Deciding what each agent is allowed to do and standing behind whether its output can be trusted. So the choice is clean. You can be the MSP that owns that layer on purpose, orchestrating which AI does what inside each client environment,

[00:07:52] verifying the output before it reaches the client, and pricing that judgment as the service it is. Or you can keep reselling tools whose maker is now your competitor and absorb the cost, the cleanup, and the compliance questions the day the ungoverned stack produces something nobody checked without ever having been paid to carry it. This episode is brought to you by ControlMap.

[00:08:19] Growing MSPs are using ControlMap to build recurring revenue by expanding their GRC services. Starting now, ControlMap is offering a free plan for MSPs looking to get started with providing compliance as a service. Create a free account and run an assessment. Track key items like policies, risks, and evidence in one place. It's a practical way to prove value to a client before deciding to expand your compliance offering. Try ControlMap for free today.

[00:08:46] Visit scalepad.com slash Dave to get started. That's scalepad.com slash Dave. Why do we care? Because your client is never going to walk in and ask, who's orchestrating their AI or checking its work? They don't yet know that's the question. You bring it to them. In the next review, put one thing on the table. Here's every place AI is now making a call inside your business.

[00:09:14] And here's exactly who verifies it before it reaches your customers. The first time a client hears that from you, instead of finding the gap the hard way, you've stopped being the shop that sells them tools and become the one they trust to govern them. So what to consider? Build the map before you book the meeting. The conversation only works if you can actually produce it. So walk each client environment first and list every place AI is now making a decision

[00:09:44] or generating output that reaches a person. Copilot drafting their client emails, a support tool, auto-resolving tickets, an agent touching their data, a note beside each one who or what checks it before it goes out. Most of those who verify sells will come back blank. That blank is the conversation. Lead with the output that reaches their customer, not the tool. Don't open the review with an inventory of platforms.

[00:10:13] Open with the one or two AI outputs that touch their customers, because that's where a wrong answer costs them a relationship, not just a workflow. Pick the highest stakes path, the AI that answers a customer, prices a quote, or flags a transaction. And make the verification question concrete there first. That's how a governance pitch becomes a business risk conversation the client actually leads into. And turn the gap use service into a named price service.

[00:10:43] The point of bringing the map isn't to hand over a free audit. It's to make the service the client now wants and put a number on it. When the blank, who verifies sells become visible, set a defined offering against them. You decide which model does what in their environment. You stand behind the outputs that reach their customers. And the judgment is a line item, not a favor folded into the base rate. A client who sees the gap alongside you

[00:11:11] is far more willing to pay you to close it than the one you pitch cold. If this trend continues, within the next year, the MSPs who bring clients that map first will have turned which AI runs here and who verifies it into a standing review question and the billable governance line. While the shops that wait will be the ones answering it reactively after a client's AI has already put a wrong answer in front of their customer. This is the Business of Tech.

[00:11:42] Want more from the Business of Tech? Join Business of Tech Plus for ad-free episodes, early interviews, extended cuts, subscriber-only shows, and exclusive member perks and analysis. Sign up at businessof.tech slash plus. And follow this show in your podcast app and if you're on YouTube, hit subscribe and the bell so you never miss a story. Reviews and comments help spread the word too. Interested in advertising?

[00:12:09] Head to mspradio.com slash engage. The Business of Tech is written and produced by me, Dave Sobel, under ethics guidelines posted at businessof.tech. Thanks for listening. I'll see you on the next episode. Part of the MSP Radio Network.