An AI tool that helps an employee finish a task faster is useful. An AI-enabled process that changes how an MSP delivers service, prepares proposals or supports customers can change the business.
That distinction was my biggest takeaway from Greg Buzek’s presentation at BlueStar’s VARTECH event last week at Mohegan Sun in Uncasville, Connecticut. The IHL Group president challenged attendees to look beyond individual productivity and consider how their organizations would operate with AI at the core of the work.
For MSPs, that challenge has two parts: Improve your own operations, then turn what you learn into services customers will pay for. Both require more than buying licenses or getting better at prompting. They require deciding what should change – and proving that the change creates value.
Start with the workflow, not the tool
Buzek described a familiar pattern: Organizations bolt AI onto existing processes without reconsidering the processes themselves. Employees draft emails and summarize meetings faster, but the underlying handoffs and decisions stay intact.
The McKinsey research he cited shows the cost of stopping there. Only 11% of leaders surveyed said their organizations had reached “reinvention,” redesigning roles, workflows and operating models around AI. Yet 48% of that group report realizing enterprise value, compared with 13% of those that simply give employees AI tools.
His practical starting point was to examine any activity that consumes more than four hours a month and ask how it would work if AI were at the center, with all the relevant knowledge available. “Take the time to rethink,” he said.
The pace of change adds urgency. On GDPval, a benchmark Buzek cited that measures real work across 44 occupations, OpenAI’s GPT-5.2 Thinking matched or beat industry experts on 70.9% of tasks last December, up from 38.8% for its predecessor.
For an MSP, proposal development is a useful example. Having AI polish a statement of work addresses one step. Connecting discovery notes, service definitions, pricing rules and approval requirements could improve everything from the customer conversation to a proposal ready for review. Measure accordingly: How long does a proposal take, how much staff effort does it require and how often does someone correct scope or pricing?
Give AI the context to do useful work
Buzek called a “second brain” probably the single biggest unlock in AI. He described it as a single cloud-drive directory that agents can access, with a roadmap to where everything is.
Because his system knows his organization’s products, offerings and discount schedules, he said he can go from a client conference call to a finished proposal in about 15 minutes. He typically still adjusts it, since the draft may be more or less aggressive than he wanted.
That is his example, not a guarantee for every MSP, but the lesson travels: AI becomes more useful when it can draw on the information that makes your business different. For MSPs, that includes service documentation, customer configurations, support history, escalation procedures and approved commercial terms – kept accurate, accessible and appropriately restricted.
Knowledge management therefore belongs near the beginning of an AI initiative. If experienced technicians routinely compensate for missing documentation, automation will hit the same gaps.
Expand autonomy with accountability
Buzek then showed his own agent team: 15 specialized agents covering research, writing, marketing, reporting, security and more, directed by a “chief of staff” agent that routes work across AI models and supports IHL’s five full-time employees. Across 3,600 agent runs since April 1, he said, the agents have averaged a 1,500% improvement over human performance on the same work. That is IHL’s internal measure, not an independent benchmark, but it shows how far redesigned work can move beyond assisted work.
For MSPs, the practical lesson is to start with a bounded responsibility and expand as results justify it. A support agent might first summarize a ticket and recommend a next step. Greater authority should follow demonstrated reliability, clear permissions and a defined escalation path.
Buzek’s governance slide made the division of responsibility explicit. Humans set goals, approve tools and permissions, define the risk tier and retain final escalation authority. Agents execute within those boundaries and escalate exceptions to a human owner. An MSP should be able to explain what an agent can access, what it can change and when a person must intervene.
Governance covers everyday habits, too. Buzek warned against sharing AI conversations through public links – a risk that became real in 2025, when thousands of shared ChatGPT chats surfaced in Google search results.
Measure the business result
That discipline matters because unclear value sinks projects. In a June analysis, Buzek highlighted Gartner’s prediction that over 40% of agentic AI projects will be canceled by the end of 2027 due to escalating costs, unclear business value or inadequate risk controls.
A shorter ticket response time is promising but incomplete without resolution quality, repeat contacts and customer experience. Establish a baseline, choose a limited workflow and compare results after the change, counting the time people spend reviewing and correcting outputs along with the cost of the tools.
Hours saved become business value when the organization uses that capacity deliberately – to reduce backlogs, improve service, pursue opportunities or handle growth.
Turn internal experience into a customer offering
Buzek told the channel partners in the room that internal adoption and customer opportunity belong together, and that the opportunity comes with a threat. What happens, he asked, when a customer that uses three functions of its point-of-sale system decides to vibe-code its own? For MSPs, the equivalent is a customer that decides AI lets it build what it now buys as a service. His answer was to expand into digital services, reporting and analytics.
Successful internal projects lend credibility to that case. A demonstrated improvement in proposal turnaround or support handling is more persuasive than a general promise of AI transformation. It also reveals what the service actually requires: documentation, integration, permissions, monitoring and ongoing refinement. Those responsibilities give MSPs a role that a do-it-yourself tool can’t fill.
I left Buzek’s presentation with a practical challenge for MSP leaders: Pick one recurring workflow, define the result that matters and give someone responsibility for improving it. Measure the full outcome, then use that experience to tackle the next process.
That is how AI adoption becomes an operating advantage – and an offering worth buying.
