Presented by Insight Enterprises
Midmarket companies can’t move AI from pilots to production by simply copying the Fortune 500 playbook. They’re competing for the same AI engineers, data specialists, security experts, and forward-deployed engineers (FDEs) — at the same salaries — with a fraction of the budget.
For companies with $500 million to $5 billion in annual revenue, that makes building a large internal AI organization neither realistic nor necessary.
Three choices separate a workable midmarket AI strategy from a scaled-down enterprise one, says Jack Azagury, president and CEO of Insight:
A handful of deliberate AI bets tied to measurable business outcomes
A core in-house team that owns AI strategy and architecture
Outside partners that supply specialized engineering and security expertise as projects scale
“What you needed your engineers to know six months ago is radically different from what they need to know now,” Azagury says. “You need a team that can learn and adapt very fast, and that is very hard for a company with $1 billion in revenue to build.”
The AI productivity trap: Saving time isn’t the same as creating ROI
Most companies have experimented with AI and trained employees, but translating those efforts into measurable revenue growth or cost savings remains a challenge.
“A lot of AI work has given people an extra coffee break here and there, maybe freed up half an hour in the day,” Azagury says. “How does a CEO or CFO quantify that, when they’re spending hundreds of thousands, if not millions, of dollars on tokens, skills, and cloud subscriptions?”
For midmarket organizations looking to go from hype to how, closing that gap requires two tracks: bottom-up experimentation and a small number of top-down initiatives where leadership funds the redesign of entire processes and tracks the return.
Insight is putting millions of dollars into three such “mega processes”:
AI sales coach:
Supporting thousands of employees in sales and presales
Customer operations:
Improving order fulfillment, shipping, and customer service
Finance:
Transforming accounts payable, receivables, planning, and forecasting
At the same time, the company encourages bottom-up AI innovation across its global workforce of more than 14,000 employees.
“When I started in technology 30 years ago, we talked about people, process, and technology,” Azagury says. “Today, the tech is so amazing that people forget the other two legs of the stool.”
Two or three big bets beat a portfolio of AI pilots
Choosing those bets doesn’t require months of analysis. Companies can evaluate end-to-end processes against two questions: How much value could they deliver, and how difficult will they be to implement? Data quality is often a deciding factor, since a process whose data is in poor shape will take longer to transform regardless of its potential value.
“Don’t spend months trying to find something exact,” Azagury says. “Pick two or three areas and go after them full speed. And deploy them at scale with a real business case.”
Nor does every bet have to succeed. If one of three projects delivers only 20% of its expected value after six months, he says, cut its scope and budget and redirect resources to another area. That can be a boon for a midmarket company without necessarily needing to source new, additional budget.
The goal isn’t to identify the perfect use case. It’s to focus enough resources on a select few to determine whether they can deliver measurable business value.
Own the decisions, borrow the depth
For midmarket companies, the question isn’t simply what to outsource. It’s which capabilities they must retain to stay in control of their AI direction when much of the execution happens elsewhere.
“You can’t outsource your AI strategy,” Azagury says. “Your architectural principles, your data strategy, and your security strategy have to live in the organization too, along with a core set of engineering capabilities linked into the business, so you can guide your external help.”
That creates a clear dividing line:
Keep in house: AI strategy, architecture, data and security strategy, core engineering, governance, and change management.
Bring in as needed: Specialized engineering and security expertise and additional capacity as projects scale.
Azagury describes this as a “build, buy, and borrow” strategy: determine what to build internally, what software and cloud capabilities to buy, and what expertise to borrow from partners. Before building, any company, midmarket or otherwise, should also determine whether a software or cloud provider already offers it — or is likely to soon.
Governance and change management also belong inside the organization. AI ownership remains murky at many companies, while employees may still be learning the technology or be reluctant to adopt it.
“How you train people and how they adopt the use cases is a big part of whether you succeed,” Azagury says. “You need those capabilities in house, even if you get help.”
Production readiness, then, is as much organizational as technical: companies need clear ownership and adoption alongside the data, security, and architecture required to scale.
When experimentation creates 10,000 agents, governance gets real
Insight’s own experience illustrates how quickly widespread AI adoption can create a new operational problem. The company discovered it had almost 10,000 agents running across the organization. Some were set up by teammates across business units to do specific tasks like summarizing documents, routing requests, or automating reports. Others were created and then operating without a clear owner or purpose in limbo.
That raised a different set of questions: Do those agents have the right security access? Which are consuming tokens without adding value? How should their performance be managed?
“You need that control plane,” Azagury says.
And that control plane is only one piece of the AI stack. Leaders must also make choices about compute, cloud providers, data, semantic context, agent platforms, and security — areas where the technology and available products continue to evolve rapidly.
For a midmarket company, trying to build deep expertise across every layer is precisely the trap to avoid. Azagury’s advice is to retain control of the architecture while making a limited number of technology and partner bets, rather than spending scarce engineering resources building capabilities a vendor may commoditize six or 12 months later.
The companies that move beyond experimentation, he says, tend to have the same fundamentals: core AI talent, a clear build, buy, and borrow strategy, and governance that lets them make decisions and scale what works.
“Focus on understanding the ROI from your AI deployment, and pick a few end-to-end processes you’re going to transform to get to that ROI,” Azagury says. “Vet the people side and the process side, prioritize two or three areas with a clear business case, and execute on those.”
For midmarket companies, the advantage won’t come from matching the AI resources of the Fortune 500. It will come from being far more deliberate about where those resources go.
Sponsored articles are content produced by a company that is either paying for the post or has a business relationship with VentureBeat, and they’re always clearly marked. For more information, contact sales@venturebeat.com.
