Start with a workflow that consumes too much time. Give AI access to the company knowledge it needs to improve that workflow. Then use what you learn to help customers solve their own business problems.

That approach emerged from an AI panel at BlueStar’s VARTECH 2026 event, moderated by Jim Roddy, president and CEO of the Retail Solutions Providers Association (RSPA). Scott McClannahan, president and CEO of Simple POS Solutions Inc.; Blake Gillum, president of Nashville-based DCR; and Greg Buzek, president of IHL Group, described how they are applying AI across their businesses, from email and contract review to sales preparation and customer analytics.

Although their examples centered on retail technology, MSPs face many of the same pressures: overloaded owners, scattered information, expensive sales preparation and customers who expect more value from their technology investments. The panel offered several practical places to begin.

Rethink the workflow before adding AI

McClannahan, who says he receives 700 to 1,000 emails a day, built an email assistant that organizes messages around the work they require rather than when they arrived. “Must reply” messages are separated from “heads up” notices such as a changed flight, “FYI” items such as receipts and shipping updates, and what he calls “time suckers” – requests such as a proposal or a legal review that will take hours. The assistant helps schedule that work, too, and lets him ask what he can get done in the time available. “I got 15 minutes and I need to focus on sales,” he said, describing a typical request.

For an MSP owner juggling escalations, vendor updates and sales opportunities, the useful idea is organizing work by urgency and effort. A chronological inbox rarely does that well.

Buzek warned against bolting AI onto an existing process. “If you do, you might get 20% to 30% improvement,” he said. “Take the time to rethink. If you rethink this process with AI at the core, you’ll get 20 to 30 times the improvement.”

For MSPs, that could mean redesigning escalation preparation: have AI assemble the ticket history, relevant documentation and outstanding questions before a technician reviews the case. Measure the entire process, including review and corrections, to see whether it actually saves time.

Expect a dip before the payoff. Buzek described an AI “J-curve” in which productivity falls while employees learn a new way of working, then climbs once they get past it. Budget for that learning period rather than assuming immediate savings.

Build a second brain your AI can work from

The panelists repeatedly returned to the idea of a “second brain” – an organized collection of business information that an AI assistant draws on for context. Buzek credits his with letting him go from a sales call to an almost fully edited proposal in about 15 minutes. Gillum, who uses Anthropic’s Claude and its Cowork desktop app, said Buzek helped him set his up, and “this has changed the way I almost run my business.”

Buzek said the setup takes about two hours. The first hour goes to choosing the core documents that explain the company: who it is, how it operates, its key people and its offerings. Those documents are converted to Markdown, a simple text format, and organized in a single directory. When the assistant works from that folder, it reads a handful of files on every request – an index of what lives where, a memory file of what it has learned and a running log of past work – so it starts each task with that context. Buzek stressed using the desktop apps, such as Claude Cowork or OpenAI’s Codex, rather than the browser versions, so the assistant can reach the directory.

He also tags the facts the assistant records with confidence scores: 100 for information taken directly from his own material, 80 for something inferred from it and 70 for a reputable outside source. Anything below 70 gets rechecked and removed. Buzek said the approach keeps answers tied to the company’s own information. It reduces fabricated answers rather than eliminating them, so a person should still review anything that reaches a customer.

Capture tools keep the brain current. Buzek recommends Granola to record and summarize meetings and Obsidian, a free note-taking app, with its web clipper to save useful pages. Recurring update prompts can run as scheduled tasks. And the brain doesn’t have to belong to the owner alone. “It can be a shared folder that everybody has access to on your team,” he said.

Gillum connects his second brain to DCR’s CRM and ticketing system. When an escalated issue reaches him, he asks for an executive summary before the call. Beyond saving time, he said, it gives him an unbiased view. “If there’s culpability on any end, the story gets polished a little bit,” he said.

MSPs can start with a narrow set of approved materials: service descriptions, current support procedures, proposal templates and documentation for one customer or technology. Assign someone to maintain the collection and remove outdated guidance, and ask the assistant to cite the documents behind its answers so employees can check them. A useful first test is whether a technician can find a supported answer faster than with the existing search process. Expand the collection once that works consistently.

Use guardrails to speed up contract reviews

DCR doesn’t have an attorney on staff, so each round of contract edits with its outside firm costs money. Gillum built a contract review skill into his second brain that sorts terms into three tiers: non-negotiables that DCR must have or walks away, terms it prefers and “perfect world” terms it would like to get. He runs reseller agreements, maintenance agreements and customer redlines through it, then has the assistant draft an executive summary for the attorney explaining what changed and why.

“You don’t want AI to be your attorney,” Gillum said. “But if you can take those contract turns from six, seven down to one or two, that’s great.”

MSPs negotiate master services agreements, vendor contracts and customer redlines constantly. Writing down your own non-negotiables – liability caps, security responsibilities, termination terms – is valuable on its own. Pairing that list with AI can shorten negotiations and reduce legal spend, as long as counsel still reviews the final agreement.

Give salespeople better preparation

Gillum uses AI to research prospects against DCR’s ideal customer profile and to prioritize whom his team should call first. The research also surfaces talking points, such as new regulations affecting grocers or issues a prospect has raised in a press release or interview. RSPA has published a webinar on prompting AI to build prospect lists, and its AI Resource Hub lays out internal use cases from sales enablement to support automation.

Because DCR’s second brain knows the company’s branding, templates and portfolio, Gillum’s team can produce a brochure or presentation tailored to a prospect’s priorities in about 10 minutes. If a grocer’s biggest concern is shrink, loss prevention moves to the top instead of sitting at the bottom of a generic list. He draws a clear line on automation, though. “I don’t believe in it reaching out, doing the calls,” he said. “What I want is to use it to make a more personalized experience for that customer’s first touch.”

For MSPs, that starts with defining a good prospect precisely: industry, geography, size, technology needs and fit with the services the MSP can deliver. AI can then help assemble research and draft relevant talking points. The salesperson still needs to verify the research and ask discovery questions. A prospect’s website may indicate a business priority, but it doesn’t establish the condition of its IT environment.

Apply the same discipline to follow-up. McClannahan records customer meetings so concerns raised in a conversation make their way into later follow-up emails. “Sometimes you forget those things,” he said. An MSP can turn approved meeting notes into a draft follow-up that records concerns, agreed actions and unresolved questions for the account manager to review. McClannahan also produced a product marketing video in about three hours for roughly $300, a project he estimated would have cost $3,000 to $5,000 with a traditional crew.

Start with champions, not companywide licenses

When an audience member asked how to manage AI spending across teams of dozens or hundreds of technicians, Buzek was blunt. “The biggest mistake people make is they buy Copilot licenses for everybody and they just waste the money,” he said. Instead, he recommended giving tools to the most productive employees first and letting others request access as demand develops.

That advice applies twice for MSPs: to their own rollouts and to the Copilot conversations they’re having with customers. A pilot group with a defined goal makes a stronger case for a broader rollout than a blanket purchase.

McClannahan has two champions on his team. One, a technician, began resolving problems faster with AI and later built a piece of custom software for a customer, even though he isn’t a developer. But McClannahan’s experience also showed why that enthusiasm needs oversight. Another employee began using AI on his own, and some of what he brought back included more detail than the company wanted to share with customers. Some of it was simply wrong, drawn from information about a point-of-sale system that was 12 years old.

“Before you jump too far, especially bringing this to your team, make sure you have some guardrails in place,” McClannahan said. “Simple things like don’t put customer data on the public ChatGPT. It sounds silly, but people will do it.” Roddy noted that RSPA has added AI policy guidance to the legal documents on its website.

Buzek added a warning about share links. An RSPA member searching for Toast installation pricing found a bid that someone had shared with a client through a ChatGPT link, exposing all the pricing. Keep work internal and confirm settings so company data isn’t used to train models.

For MSPs, the lesson is to define approved tools, permitted data and review responsibilities before expanding access. Customer information should enter only environments the MSP has approved for that purpose, and technicians should verify recommendations against current documentation before applying them or sending them to clients. Choose a small pilot group and track a specific outcome, such as time spent preparing proposals or researching escalations. Include licensing, setup, maintenance and review time in the evaluation. These are also the policies many MSP customers need help writing.

Put AI behind the support desk

The panel ran short on time for support, but Buzek said it’s the easier place to start. IHL Group loaded its proprietary research into Delphi.ai to create “Just Ask Greg,” a private assistant people can query. He also pointed to My AskAI, which plugs into Zendesk to pull support content in real time. He put the cost at around $99 a month.

MSPs can apply the same idea to an internal assistant built on their runbooks and resolved tickets, or to a customer-facing tool for common questions. Start with content that’s current and approved, and measure whether it reduces repeat tickets or escalations.

Turn customer data into a service

Support savings show up lower on the income statement. Buzek argued that the bigger opportunity sits on the revenue side. Roddy framed it with a line from retail strategist Carl Boutet at RSPA’s RetailNOW conference: solution providers must be repositioned “from installers of products to stewards of the merchant’s intelligence.”

Gillum put it in business terms. “The big question is, how do you monetize AI?” he said. “We want to make money off of it. We don’t just want to improve our cost center.”

DCR began by asking clients what information they were missing. It then built a data pipeline that pulls information out of point-of-sale, phone and other systems, normalizes it and presents it in dashboards with AI agents on top. A grocer can see a suggested order based on sales history, season, weather and upcoming holidays, along with the price at which a product moves best, and turn that recommendation into a purchase order. “Those who can help control and aggregate your customers’ data will be the winners,” Gillum said.

McClannahan took a partnership approach. Rather than build everything, Simple POS partnered with seven or eight companies that already had the expertise, bringing customers offerings in security and loss prevention, payment statement analysis and marketing. The loss prevention service, for example, combines point-of-sale data with security camera footage to flag possible theft.

MSPs can use either route. Start with a customer decision that is difficult because information sits in separate systems. Identify the data needed, how it will be connected and who will review the resulting recommendations. An offering might combine integration, data maintenance, reporting and a recurring advisory meeting. Price the service around that defined work and test whether it helps the customer make a better decision.

Buzek closed with a warning. If helping clients use AI isn’t a key part of your offering, “somebody else will come in and take that account from you,” he said.

Keep people at the center

For all the talk of agents and automation, Gillum described AI’s role as freeing employees to spend more time on relationships. “It’s not to replace people,” he said. “It’s to augment them.” Buzek added that owners need to communicate that intent clearly, so employees see AI as help rather than a threat.

That gives MSPs a concrete test for their next AI project: Does it free employees for more valuable work, improve the customer experience or create a service customers will pay for? Choose one outcome, establish a baseline and prove it before expanding.