Artificial Intelligence has moved from buzzword to business reality. For middle market companies—those with annual revenues between $10 million and $250 million—AI for middle market businesses represents a genuine opportunity to do more with less, without adding headcount.
But here’s the key: AI is a tool, not a replacement for people. Think about when spreadsheets replaced ledger paper. When Lotus 1-2-3 (and later Excel) came along, everyone worried that accountants would become obsolete. They didn’t. Spreadsheets just allowed them to do their jobs more efficiently and focus on higher-value work. We view AI the same way.
The value you create isn’t in completing a task—it’s in what you do with the results once the task is done.
How to Use AI to Improve Business Operations: Where to Start
Most middle market companies don’t need cutting-edge AI infrastructure to see results. The biggest wins come from identifying where your team is spending time on repetitive, low-value tasks, and finding ways to automate or streamline them. Knowing how to use AI to improve business operations starts with looking at where your team is losing the most time.
Here are three areas where we’re seeing real impact with our clients:
1. Automating Repetitive Data Entry
One of the most immediate opportunities is invoice processing. Rather than having staff manually key invoices into your accounting system, AI can automate that entire workflow. Teach your suppliers to email invoices to a designated address, and the system automatically loads that data, whether you’re running QuickBooks, NetSuite, SAP, or another ERP platform.
Real-Life Example
A regional building products distributor running NetSuite processed about 1,400 vendor invoices a month across two AP clerks. The manual routine was the usual: pull invoices from email and mail, key header and line data into the system, scan paper copies, attach the PDF to the transaction, then file. Each invoice ran roughly 7 to 9 minutes end to end, and duplicate payments and keying errors surfaced 2-3 times a quarter, usually caught after the check cleared.
They turned on the AP automation already available in their platform and set up a dedicated inbox for suppliers to send invoices to. The system now reads each invoice, populates the bill in the AP register, and attaches the source document to the transaction automatically. Around 80 percent of invoices post clean. The remaining 20 percent, mostly PO mismatches, missing GL coding, or pricing that does not tie to the agreement, route to a clerk as exceptions.
The result over the first two quarters:
- Processing time on a clean invoice dropped from 7 to 9 minutes to under a minute of review. Total AP hours fell from about 150 a month to 40.
- Duplicate and overbilling catches went up, not down, because the clerks now reconcile and audit instead of key. They recovered a little over $9,000 in the first quarter from vendor overcharges and duplicate submissions they previously would have paid.
- No documents went missing. Every invoice lives electronically with its transaction, which also cut the time spent hunting for backup during the year-end review.
- One of the two AP clerks moved to accounts receivable, where the company had been planning to add a hire as volume grew. That hire, budgeted around $55,000 plus benefits, was shelved. Same headcount, more work absorbed, no added payroll.
2. Cleaning and Manipulating Data
Many middle market companies are running on disparate systems that don’t talk to each other cleanly. Data gets collected but never used, not because it isn’t valuable, but because nobody has the time or tools to make it usable. AI can close that gap faster and more affordably than most business owners expect.
Real-Life Example
A mid-market manufacturer was capturing compliance data on the shop floor every shift. The problem was not collection, it was format. The reporting system exported to Excel in an unstructured layout: merged cells, inconsistent headers, duplicate rows, readings scattered across columns that did not map to anything their ERP could ingest.
Cleaning it required either advanced Excel skills the team did not have, or brute-force manual work. One person spent the better part of a day each week reshaping the export by hand, deduping, and reformatting before it could go anywhere. Most weeks it did not get done. The data sat in files nobody could use. The company was gathering compliance information it had already paid to collect and getting almost no value from it except in one-off pulls when someone had a spare afternoon.
We wrote a prompt that mapped the unstructured export into the structured format the ERP required. It took under an hour to write and test.
The new routine: pull the morning download, run it through the prompt, get clean structured output in under five minutes, upload to the ERP. What had been a full day a week became a five-minute step in someone’s morning.
The result:
- The weekly cleanup, roughly 6-8 hours, dropped to about 25 minutes a week. No one was hired to own the task, and the person who had been losing a day to it got that day back for actual analysis.
- More important than the hours, the data became usable. Compliance readings that had lived in dead spreadsheets were now in the ERP, visible globally, and could be queried for risk and trend patterns the company had never been able to see before. They went from having data to having information.
- The company is now moving this to a scheduled prompt so the structured file loads to the ERP with no human in the loop at all. The manual version already paid for itself. Automating the handoff removes the last few minutes of touch.
3. Reducing Turnover in High-Repetition Roles
Some roles have high turnover simply because the work is tedious. Think hours of manual data entry with little variety. AI won’t eliminate those roles, but it can reshape them. When repetitive tasks are automated, the people in those roles can take on more meaningful work, which improves both retention and output.
Real-Life Example
A Tier 1 automotive supplier, family-owned and middle-market, shipped to its OEM customer just-in-time and sequenced, generating roughly 5,000 unique transactions a day. The reconciliation problem came from a mismatch in how payment was tracked. The OEM paid off the EDI transaction number. The supplier invoiced off the number its ERP generated. The two never tied cleanly.
So every month when payment came in, the AR staff spent the better part of a week matching what the OEM paid against what the supplier expected. Short pays and debits made it worse, each one a small investigation with no clear starting point. The work bled into the close. Books ran late, and the cash forecast was unreliable one week out of every four. It was also the part of the job everyone dreaded.
We built it in two prompts. The first stitched together three sources that had never been joined: the AR register, the customer remittance advice, and the EDI transaction data. That produced a single dataset linking each supplier invoice number to its OEM EDI transaction number, the connection that had been missing all along. The second prompt reconciled that dataset, flagged every invoice paid in full so it could be closed in the ERP, and kicked out an exception report for everything that did not tie: short pays, debits, the occasional overpayment. Exceptions could be routed to the right department, so a quality debit went to quality with the backup attached, not to an AR clerk guessing at cause.
The result:
- The monthly reconciliation went from roughly a week of manual matching to an exception queue the team could clear in a fraction of the time. The close stopped slipping. The one bad forecast week a month went away, because AR now knew what had actually been paid instead of estimating.
- The staff’s job changed more than their workload did. They stopped keying and guessing and started resolving. Payment issues that used to drag on for months, sometimes past a year, got worked while they were still fresh, because the exception report surfaced the the same week the remittance landed. Working with quality and other departments to run down a debit became the job, instead of the thing there was never time for.
- Morale moved. The month-end reconciliation had been the worst stretch of the AR calendar. Taking it off their plate did as much for retention in that group as any comp change would have.
4. Enhancing HR and Employee Experience With Sentiment Analysis
Sentiment analysis is transforming how companies listen to and retain their people. By leveraging AI to analyze employee feedback and surveys, organizations can extract actionable insights that improve operations and foster engagement.
Real-Life Example
A mid-sized company ran its own employee engagement surveys and ran them well. Participation was strong and the surveys were anonymous, so people answered honestly. The trouble was on the back end. Management had numerical scores and a pile of written verbatims and no reliable way to turn either into action. Scores told them the temperature, not the cause. The comments were where the real signal lived, but reading them was subjective, and HR tended to set aside whatever looked like an outlier. So the loudest problems got smoothed over, and leadership was often guessing at what employees were actually trying to tell them.
They brought AI in to do sentiment analysis across the full set of verbatims, outliers included. The point was not a prettier dashboard. It was a richer read of what people were saying, including the uncomfortable parts HR had been discarding, and the ability to see themes across hundreds of comments instead of reacting to the few that stuck out.
What they did with the read is what made it work. Rather than quietly adjusting policy, they held town halls and led with “this is what we hear you saying,” reflecting the sentiment back to the people who had reported it. Then they split the response two ways. Where the friction came from a misunderstanding, they communicated the why behind decisions employees had been reacting to without context. Where the friction was real, they stood up employee-led cross-functional teams to work the actual problem.
The result:
- Employees felt heard, which is the outcome the whole section is about. The town halls closed the loop that anonymous surveys usually leave open. People could see their own words reflected back and see something happen because of them.
- Communication improved in both directions. Some friction dissolved once employees understood why the company did what it did. Other friction got solved by the people closest to it, on teams that surfaced issues management had never fully seen.
- Over time the culture moved, and the company’s reputation as a place to work moved with it. Employees and management were solving problems together that had previously gone undetected or unaddressed. The people who took the surveys got what they were really after: a voice, and a seat at the table.
5. Streamlining Quality Control With Image Recognition
For manufacturing companies, AI-powered image recognition is changing the game on quality control. High-resolution cameras combined with machine learning algorithms can detect defects that might elude the human eye, catching problems earlier in the production process and reducing costly errors downstream.
Real-Life Example
A hardwood flooring manufacturer graded and inspected its product the way the industry always had: by eye. As boards came off the line, finished and ready to pack, graders visually sorted them for defects: knots and knot holes, checks and splits, wane, mineral streak, color that fell outside the run, and finish problems like skips, bubbles, or uneven sheen. Wood is natural and inconsistent, so this took a trained eye and years to develop. The method worked until volume and fatigue worked against it. At line speed, and late in a shift, marginal boards got graded generously or missed entirely. A board that should have been pulled or downgraded went into a premium bundle, and the customer found it. That came back as a claim, a return, or a quiet downgrade of the company as a supplier.
They put high-resolution cameras and controlled lighting over the finishing line and trained a model on their own boards, the actual defects their species and process produced, not a generic library. Wood is hard to inspect precisely because it is variable. A knot is a defect in a clear grade and acceptable character in a rustic grade, and mineral streak reads differently across species. The model had to learn their grades, not a textbook’s. It now scans every board, checks it against the grade it is being sorted into, and flags anything questionable for a human grader to confirm. Color and sheen consistency, which is where finish complaints usually start, get checked the same pass.
The graders did not go away. Instead, they stopped eyeballing every board and started adjudicating the ones the system pulled, and feeding confirmed calls back to sharpen the model on the edge cases wood always throws.
Two things worth noting about this project: This was a build measured in months, not an hour. In addition, cameras and lighting needed to be stable enough that grain didn’t fool it, and enough labeled boards across grades and species to make the model reliable before it touched a real bundle. And it only worked because the graders’ knowledge trained it. The model learned their standards. It did not replace them.
The result:
- Escapes to the customer dropped, because every board gets inspected against its grade instead of a sample, and the camera doesn’t tire at hour seven or start passing marginal boards to keep pace. Grading also got more consistent from shift to shift and grader to grader, which for a natural product is half the battle. Two graders can honestly disagree on a borderline board. The model applies the same standard every time.
- Catching the defect at the line instead of at the customer changed what each miss cost. A misgraded board caught in-house is a downgrade or a rework . The same board caught by the customer is a claim, freight both ways, and a mark against the relationship.
- The grade accuracy is the outcome that pays. When a customer trusts that your premium grade is actually premium, you hold the premium price and you keep the account. Consistent grading protects both the margin on the high grades and the reputation that lets you keep selling them.
6. Boosting Productivity With Generative AI
Generative AI is also making a real difference in content creation and marketing. Tools that can produce high-quality materials, like articles, social media posts, personalized outreach, in a fraction of the time free up your team to focus on strategy and relationships rather than production.
The key, as with everything in this article, is keeping the human touch. AI can handle the output, but your people provide the judgment, creativity, and authenticity that make it worth reading.
Real-Life Example
A firm that published regularly, blogs, technical papers, case studies, and white papers, had the ideas and the expertise but not the time. The people best equipped to write were the same people carrying a full day of client work. Writing was the thing that got pushed to nights and weekends, and then pushed off entirely. Good thinking sat in people’s heads and in half-started drafts because there was never a clear runway to turn it into finished, publishable work.
They brought in generative AI, but pointedly not as a ghostwriter. The rule was that the ideas stayed human. What AI did was compress the distance between a rough idea and a structured draft. Someone could dump their unordered thoughts, the points they knew they wanted to make but hadn’t sequenced, and use AI to organize them, pressure-test them, and sharpen the wording until the piece said what they actually meant. The thinking was theirs. AI handled the part that used to stall them: getting from a page of notes to a coherent structure they could then refine in their own voice.
The result:
- Work that had been stalling for months started getting finished. The bottleneck was never the ideas, it was the hours to shape them, and that was the part AI took off the plate. People who had good material and no runway now had a way to bring it to life on the time they actually had.
- The quality held because the judgment stayed human. The firm was deliberate about this. AI was structured and refined, but a person decided what was worth saying, whether it was true, and whether it sounded like them. Content that is generated end to end by a machine reads like it, and nobody wants to read regurgitated output. The human side is what makes a piece worth someone’s time, and that is exactly the part they refused to hand off.
The Right Mindset for Adopting AI
We practice what we preach. Since mid-2025, we’ve been leveraging AI and tools like Microsoft Power Automate to handle tasks that previously required a dedicated assistant, freeing our own team up for higher-value client work. How to use AI for business effectively means starting with the right principles:
- AI handles the task; your people handle the thinking. The goal is to free up your team’s time for interpretation, decision-making, and relationship-building—the things AI can’t replicate.
- Start where the pain is. Look for processes with high manual effort, frequent errors, or high turnover. Those are your best candidates for automation.
- You still need to understand the work. Just like you can’t skip from elementary school to college, AI doesn’t eliminate the need to understand your business processes. It just makes executing them faster and more efficient.
- Don’t skip the human touch. We never want our client interactions—or our own work—to feel AI-driven. AI helps us work smarter, but it doesn’t replace the judgment and relationships that drive real results.
Practical Considerations Before You Start
The advantages of AI are real, but so are the challenges middle market companies face with leveraging this technology. A few things to keep in mind before getting started:
- Data privacy matters. Make sure any AI tools you use have appropriate data protection protocols, especially if you’re handling sensitive customer or financial information.
- Evaluate total cost. Factor in not just the initial investment but ongoing maintenance, upgrades, and training.
- Train your team. AI tools are only as effective as the people using them. Budget time and resources for onboarding.
How to Get Started
You don’t need to overhaul your entire operation at once. Start by identifying one or two processes where your team is losing the most time to repetitive, low-value work. Map out what that process looks like today, and ask whether AI could handle any part of it.
As with any significant business decision, take a measured approach—research your options, seek expert guidance where needed, and evaluate what makes sense for your specific situation.
Curious How to Use AI for Your Business?
We’re not AI experts—but we are business advisors who work alongside middle market companies every day, helping them find practical ways to operate more efficiently and grow more profitably. The more we explore AI with our clients, the more we see its potential to meaningfully improve how middle market companies operate.
If you’re wondering where to start or whether AI makes sense for your business, that’s exactly the kind of conversation we have. Contact us to learn more about how we can help.
About Tony
Tony is a data-driven financial professional with years of experience in the sell side of investment banking, specializing in risk management and hedging consultation using derivative products, due diligence, valuation, and financial modeling. His expertise extends to corporate restructuring, where he has demonstrated his adeptness at navigating complex financial landscapes.
Beyond his professional achievements, Tony is a lifelong problem solver, constantly seeking opportunities where new technology can enhance business operations and efficiency.
Tony holds a Master of Business Administration from Carnegie Mellon University’s Tepper School of Business, with concentrations in Finance and Strategy, and earned a Bachelor of Arts in Business Administration from Pusan National University.
Combining analytical rigor with strategic vision, Tony drives sustainable growth and success.
