i
News
News · 2026-11-16

Midmarket AI needs a different strategy from enterprise AI

Dataist Dataist

AI for midsize companies is not a scaled-down version of enterprise AI

Cover: Midmarket AI needs a different strategy from enterprise AI

Companies with midmarket revenue cannot simply shrink an AI strategy designed for the Fortune 500. With revenue of $500 million to $5 billion, they compete for the same AI, data and security specialists—and pay the same salaries, despite having far smaller budgets. According to Jack Azagury, president and CEO of Insight, these companies do not need to build large internal AI departments. Instead, they should choose a few projects with measurable returns, keep the teams responsible for strategy and architecture in-house, and bring in specialist engineering and security expertise as projects grow.

Midmarket AI requires a different strategy from enterprise AI

Time saved is not the same as returns

Many companies have already tried AI and trained employees, but turning those efforts into measurable revenue growth or lower costs remains difficult. Saving half an hour in a workday may help an employee, but executives need to know whether that benefit justifies spending hundreds of thousands or millions of dollars on tokens, training and cloud subscriptions.

Azagury says moving from experimentation to practical use requires two things: employees need the opportunity to try AI in their day-to-day work, while management needs to fund the redesign of several end-to-end processes and track their returns.

Insight is investing millions of dollars in three such areas:

AI sales coach — support for thousands of sales and presales employees.
Customer operations — improving order processing, delivery and service.
Finance — changing processes for accounts payable, accounts receivable, planning and forecasting.

At the same time, the company supports employee AI initiatives across its global team of more than 14,000 people. Azagury notes that enthusiasm for technology should not overshadow people and processes—the other two elements needed for change.

Choose two or three areas

Finding suitable projects does not require months of analysis. Companies can assess entire processes against two criteria: the value a change would create and how difficult it would be to make. Data quality often affects the timeline: even a promising process is harder to transform when its data is in poor shape.

Azagury recommends choosing two or three areas, starting quickly and deploying solutions at scale with a business case to guide the work. Not every project has to meet expectations. If, after six months, one of three projects delivers only 20% of the expected value, the company can cut its scope and budget and redirect resources elsewhere. That makes it possible to reallocate existing funds without necessarily asking for more budget.

The goal is not to find the perfect AI use case, but to commit enough resources to several projects to test whether they deliver measurable business value.

Keep strategy in-house; bring in expertise as needed

A company should retain control over its AI direction, even when outside partners do much of the work. According to Azagury, strategy and architectural principles, data and security strategy, and a core engineering team connected to the business and able to coordinate outside support should stay in-house.

Responsibilities can be divided as follows:

Keep in-house — AI strategy and architecture, data and security strategy, the core engineering team, governance and change management.
Bring in as needed — specialist engineering and security expertise, as well as additional resources as projects grow.

Azagury describes this approach as a combination of “build, buy and bring in”: the company decides which capabilities to develop itself, which software and cloud products to purchase, and which expertise to get from partners. Before starting to build something, it is worth finding out whether a software or cloud provider already offers the solution—or plans to offer it soon.

AI governance and change management should also remain the company’s responsibility. In many organizations, it is still unclear who is accountable for AI; employees may be only beginning to learn the new technologies or may be reluctant to adopt new ways of working. Training and employee involvement in specific decisions therefore affect outcomes as much as technical support does.

Nearly 10,000 AI agents create a new management challenge

Insight’s experience shows that widespread AI use can quickly create a separate operational problem: the company discovered that it had nearly 10,000 AI agents running. Employees in different departments had created some for specific tasks—for example, summarizing documents, routing requests or automating reports. Other agents had no clear owner or purpose.

This raised questions for the company: did the agents have the right access permissions, which were using tokens without delivering value, and how should their performance be assessed? Azagury says managing them requires a control layer.

That is just one part of the AI stack. Leaders also have to choose compute resources, cloud providers, data approaches, semantic context, AI agent platforms and security tools. Technologies and available products in these areas are changing quickly, so midmarket companies should not try to build deep expertise in every one of them.

Azagury’s advice is to retain control over the architecture and limit the number of technology choices and partnerships. Otherwise, a company risks spending scarce engineering resources on capabilities a provider will turn into a standard product in six or 12 months.

According to Azagury, companies that move beyond experimentation tend to rely on three foundations: their own core AI specialists, a clear choice between building, buying and bringing in outside help, and governance that enables decisions and scales solutions that work. To get there, he recommends assessing the returns from AI adoption, choosing two or three processes with a clear business case, accounting for employee and process readiness, and focusing on executing those plans.

In Dataist’s view, midmarket companies do not need to copy the staffing and technology model of large corporations. The approach described here puts more weight on concentrating resources on a few processes and keeping key decisions out of the hands of external partners. Insight’s experience also shows that even distributed employee experimentation can lead to thousands of AI agents that need to be monitored.

Daily AI news

Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.

Only what matters — every day

Follow on X