A Chief Operating Officer (COO) at a $5M to $50M company gets AI’s first payback in 5 operating jobs: process mapping, SOPs, scheduling, inventory and quality. A 12-person operations team that gets 23 hours a week back, valued at the BLS rate of $46.89 an hour, gains about $51,767 a year in capacity.
The St. Louis Fed found 37.4% of US workers used generative AI at work by August 2025, but the Census Bureau puts firm-level AI use at just 19.8% in May 2026.
That gap is an operations problem: staff use AI one person at a time, on undocumented processes, with no one counting the hours. The COO can fix that with written processes and a baseline.
AI pays back first in operations on the coordination work around the job: rebuilding schedules, writing SOPs, checking stock, writing handoff reports and answering repeat questions. These tasks are text-heavy and rule-based, they repeat every week, and a supervisor can check each AI draft in minutes before anything goes out.
The physical work is a later, bigger build. Camera inspection and machine-level sensing need clean data and capital, and at this company size they usually come after the paperwork has been sorted out.
The paperwork is where the hours hide. Microsoft’s 2025 Work Trend Index found workers are interrupted about 275 times a day, and 80% of the global workforce say they lack the time or energy to do their work. Operations coordinators live inside that number.
The gain per person is measurable. St. Louis Fed researchers found generative AI users saved 5.4% of their work hours in a week, about 2.2 hours for someone working 40.
Smaller companies are also behind on scale. McKinsey’s 2026 survey found 54% of organizations with more than $1 billion in revenue say AI is scaling across the enterprise, against about one-third of smaller organizations. At $5M to $50M, the COO is usually the right person to close that gap, because most of the recoverable hours sit in operations.
Here’s what the first pass looks like for a 12-person operations team at a $20M company. The hours are planning assumptions, so replace them with your own baseline before you budget against them.
| Operations task | Hours/week now | With AI | Hours given back | Mechanism |
|---|---|---|---|---|
| Rebuilding the weekly schedule after call-outs and rush orders | 8 | 3 | 5 | AI drafts a revised schedule from a written constraint list (skills, hours, due dates); the supervisor approves it |
| Writing and updating SOPs and work instructions | 6 | 2 | 4 | AI turns a screen recording or walkthrough transcript into a draft SOP; the process owner edits it |
| Stock checks and reorder decisions | 7 | 3 | 4 | AI reads sales and usage history and flags items heading below their reorder point |
| Shift handoff and status reports | 6 | 2 | 4 | AI summarizes the day’s job notes into a one-page handoff |
| Quality nonconformance write-ups | 5 | 2 | 3 | AI drafts the report from inspector notes and searches past reports for the same defect |
| Repeat “how do I…” questions from staff | 5 | 2 | 3 | AI answers from the approved SOP library and links the source SOP |
| Total | 37 | 14 | 23 |
23 hours a week is more than half of one full-time role given back to the team. Here’s the math, with every assumption stated.
The math on the operations table
Capacity only becomes money when you decide where the hours go. In operations, the destinations I’d pick are supplier follow-up, root-cause work on repeat defects, and cross-training so one call-out stops wrecking a whole shift.
Run your own numbers in the AI ROI Calculator before you commit a dollar.
You use AI for process mapping by recording the work first and letting AI draft the map from the recording: a staff member screen-records the task, AI turns the transcript into numbered steps with every handoff and copy-paste marked, and the process owner corrects it. A 30-minute recording can become a reviewable draft map the same day.
Interviews and intake forms miss the workarounds. A recording catches the spreadsheet someone keeps on the side, the email that has to go out before the order can move, and the 4 minutes spent retyping a customer number from one screen to another.
Ask for 2 recordings of each task: one normal run and one bad day, such as a rush order or a missing part. The bad-day run shows the exceptions, and exceptions are where AI drafts go wrong.
McKinsey’s 2026 survey shows why this step matters. High performers were roughly 3 times as likely as everyone else to have redesigned workflows around AI, and redesign starts with a map of what people actually do.
Every task on the map goes through 4 questions, in this order. I stop at the first yes.
The 20% rule is the gate on step 4. If AI would give back less than 20% of a task’s time, I recommend cutting or simplifying the task, because build and upkeep rarely pay back below that line.
Every task cut at steps 1 and 2 is capacity you get back before a single license is bought.
Here’s how the ladder plays out at a $22M commercial services company reviewing its job-to-invoice process. It’s an illustrative scenario, so read the numbers as a model of the method.
The process ran 10 steps: completed job, technician worksheet, office re-entry, spreadsheet tracker, missing-detail checks, operations approval, finance recheck, invoice creation, owner approval, then send and file.
4 of those steps went at the kill and fix stages: the office re-entry, the separate tracker, the duplicate finance check and the routine owner approval. The required financial controls stayed with the designated approver.
The team also fixed the job form, so technicians capture every required detail at the source. The new process runs 6 steps: one complete job record, an exception check, designated approval, invoice creation, send, and filing that happens on its own.
| Measure | Before | After |
|---|---|---|
| Steps in the process | 10 | 6 |
| Weekly team hours | 30 | 18 |
That’s 40% of the steps removed and 12 hours a week recovered before any AI build, worth about $27,009 a year at $46.89 an hour over 48 weeks. 8 of those hours went to resolving billing exceptions and 4 to customer follow-up.
The AI question comes after that, and it gets asked about a 6-step process.
Yes, AI can write first-draft SOPs from a screen recording, a walkthrough transcript or a photo of the old binder page, and it can keep them current when the process changes. The process owner still edits and signs off every SOP, because AI writes confident steps for exceptions it has never seen.
The workflow is short. Record the task, have AI draft the SOP in your standard template (purpose, roles, steps, exceptions, the “what good looks like” check), then have the person who does the job read it aloud while doing the work.
Every place they stop and say “well, actually” is a missing step. Fix it and publish.
McKinsey’s COO research found the same pattern in documents: one tech company collapsed hundreds of statement-of-work variations into 5 templates that take hours to produce, down from days. SOPs work the same way once there’s one template and one owner.
The common advice puts a scheduling or forecasting tool first, because that’s where vendors sell. For a $5M to $50M company, I put SOPs first, for 3 reasons.
First, every other AI use in operations reads from your written rules. An AI scheduler needs the constraint list, and an AI assistant answering staff questions needs an approved library to answer from.
Second, SOPs are cheap and fast. Claude or Microsoft Copilot can draft from recordings this week, with tools you may already license.
Third, the SOP project is the staff-readiness project. People who record and correct their own SOPs learn what AI gets right and wrong on work they know cold, and that’s the training that sticks.
AI improves scheduling by drafting a revised schedule in minutes when something changes, a call-out or a rush order, using a written list of constraints: who holds which skills, hour limits, machine or truck capacity, and due dates. The supervisor reviews and approves the draft, so the person who knows the floor keeps the final say.
The constraint list is the real work. Write down every rule your best scheduler carries in their head, such as “Maria can’t run line 2 alone” or “no driver goes over 50 hours.” Expect the list to run longer than anyone guesses, because many of those rules have never been written down.
Then measure two numbers for 4 weeks before and after: hours spent rebuilding the schedule, and overtime hours caused by schedule changes. Those two numbers tell you if it’s working.
Keep a person on every send. A schedule touches pay, overtime rules and family plans, so the AI drafts and the supervisor publishes.
If you want software that watches the order queue and reschedules on its own, that’s an agent build with its own permissions and guardrails, covered in AI Agents for Small and Mid-Sized Businesses.
You use AI for inventory management by having it read your sales and usage history, recalculate each item’s reorder point every week, and flag items heading toward a stockout or sitting as dead stock. I’d want about 24 months of clean history before trusting AI demand forecasting over a well-kept spreadsheet, so start with weekly reorder-point updates.
The mechanism is the reorder point formula you already know: average daily usage × supplier lead time in days, plus safety stock. At 40 units a day, a 12-day lead time and 160 units of safety stock, the reorder point is 640.
AI’s job is to keep the inputs honest. It recalculates average usage every week, notices when a supplier’s real lead time has drifted from 12 days to 17, and tells the buyer which 15 items changed, out of 2,000.
The same history flags the other side of the problem: items with no movement in 180 days. Put that dead-stock list in front of your buyer and controller every quarter, with carrying cost attached at your finance team’s rate, and decide what to return, discount or write off.
The buyer still places the order. Start with your A items, the 20% of SKUs that carry most of your sales or spend, and compare stockouts and carrying cost for 13 weeks before and after.
Vendors are moving the other way. Gartner predicts that by 2030, half of cross-functional supply chain management solutions will use AI agents that carry out decisions on their own. At $5M to $50M, I’d want 6 months of a buyer checking AI recommendations before handing any order decision to software.
AI is used in quality control in 2 ways: reading text and spotting patterns, such as drafting nonconformance reports and finding repeat defects across past reports, and camera-based visual inspection on the line. Rockwell Automation’s 2025 survey of 1,560 manufacturing respondents found quality control was the top AI use case for the second year running.
At this company size, start with the paperwork. AI drafts each nonconformance report from the inspector’s notes and photos, tags the defect type, and searches past reports for the same part, supplier or shift.
That search is where the hours come back. A quality lead who can ask “show me every burr defect on this part in the last 18 months, by supplier” gets a root-cause list in minutes, where it used to take a day in a spreadsheet.
Maintenance follows the same pattern. McKinsey reports that a heavy-industry manufacturer’s technician maintenance copilot cut maintenance workload by 40% and raised overall equipment effectiveness by 3%.
Camera inspection is a capital project. It needs a labeled photo set of good and bad parts, consistent lighting, and a person reviewing every reject during the first months, so I treat it as a year-2 decision for most SMBs.
Here’s the same idea at a $35M packaging manufacturer, again as an illustrative scenario. Technicians spent 16 hours a week gathering information and chasing recurring stoppages.
An AI agent reviewed equipment alerts, fault logs and maintenance history. It grouped related alerts, spotted recurring fault patterns, and drafted work orders with suggested checks and the supporting records attached. (Building the agent itself is covered in AI Agents for Small and Mid-Sized Businesses.)
The maintenance supervisor approved every work order and any planned shutdown. Technicians confirmed the diagnosis and did the repair.
Against an 8-week baseline, an 8-week pilot looks like this:
| Measure | Before | After |
|---|---|---|
| Weekly investigation and work-order prep | 16 hours | 6 hours |
| Average unplanned line downtime per week | 10 hours | 6 hours |
The 10 staff hours recovered each week went to preventive inspections (6 hours) and removing recurring faults at the root (4 hours). At $46.89 an hour over 48 weeks, those hours are worth about $22,507 a year in capacity.
Separately, the line got 4 operating hours a week back. Track machine uptime and staff time as 2 different measures, because they pay back in different places.
Before rolling it out wider, check that production schedules and equipment usage were comparable across both 8-week periods. A quiet month can make any pilot look good.
Ordered people first, then process, then tech. Each step names who does it and what it produces.
Days 1 to 30: people
Days 31 to 60: process
Days 61 to 90: tech
I’m Lee Pomerantz, Certified Chief AI Officer, founder of eMediaAI, Fort Wayne, Indiana. I work with operations and leadership teams at $5M to $50M companies, and every engagement runs in one order: people first, then process, then tech.
The AI Opportunity Blueprint™ is where it starts. It’s a 10-day, $5,000 forensic audit: your staff record the tasks, I map where the hours and the data go, and you get a prioritized workflow map, an ROI projection, an implementation roadmap and governance recommendations.
If a task returns less than 20% of its time, the report says cut it. If the numbers don’t hold, it says don’t build. After the Blueprint, I can build and deploy the workflow and hand it off with SOPs your team runs without me.
AI in operations management is the use of AI tools to plan, document and run day-to-day operating work: drafting schedules, writing SOPs, recalculating reorder points, summarizing shift reports and searching quality records. A supervisor reviews each AI draft before it goes out. At a $5M to $50M company, the first gains usually come from coordination paperwork, where eMediaAI’s planning model for a 12-person team shows about 23 hours a week given back.
AI for operations and AIOps are different things. AIOps means AI for IT operations: monitoring servers, networks and applications and flagging outages. AI for operations, in a COO’s sense, covers the business operating work of scheduling, inventory, quality and SOPs. Searches for “AI ops” mostly return IT monitoring tools, so a Chief Operating Officer researching AI should search by task, such as “AI scheduling” or “AI inventory management.”
eMediaAI’s planning model assumes about $20,000 in year one for a first AI workflow in operations that runs on tools a company already licenses, such as Microsoft Copilot, including setup and training. eMediaAI’s AI Opportunity Blueprint™ is a fixed $5,000 for a 10-day audit that prices each workflow before any build. A team getting 23 hours a week back at $46.89 an hour gains about $51,767 a year in capacity.
A first AI workflow in operations takes a few hours per person over about 60 days. Each staff member screen-records their 3 most time-consuming tasks, joins a 2-hour tools training, and reads their draft SOPs aloud while doing the work. eMediaAI’s AI Opportunity Blueprint™ runs 10 days, and the first workflow build typically follows within the next 30 to 60 days.
The 3 main risks of AI in operations are acting on a wrong AI draft without review, putting customer or pricing data into unapproved tools, and building AI on top of a broken process. Three controls cover most of that exposure: a named person approves every schedule, order and SOP before it goes out, an approved tools list says which data goes where, and every task is mapped before any build.
A COO’s first AI move is a one-page note to the operations team saying what AI is for and where the returned hours will go. Next, name one owner, have each person screen-record their 3 biggest time sinks, and map those tasks. Only then pick 1 workflow that gives back at least 20% of its time and configure a tool for it.
Measure AI ROI in operations against a baseline taken before the build: hours per week per task, schedule rebuild time, stockouts and defect rate. Multiply hours given back by a loaded rate, such as the BLS average of $46.89 an hour for June 2026, and by 48 weeks. Subtract license, setup and training costs, then log what the returned hours were spent on.
AI is unlikely to replace operations managers, because the judgment calls, people decisions and sign-offs stay with them while AI takes over drafting and data checking. McKinsey’s 2026 State of AI survey found 39% of respondents expect AI to reduce headcount, but only 14% saw an AI-driven decline in the past year. At most SMBs, AI gives operations managers hours back for supplier and floor work.
Most operations teams at $5M to $50M companies start with general AI assistants they already license, such as Microsoft Copilot or Claude, for SOPs, handoff reports and schedule drafts. AI features inside existing ERP, inventory and scheduling software come next. Rockwell Automation’s 2025 survey of 1,560 manufacturing respondents found quality control was the top AI use case for the second year running.
Book a free 20-minute call to see if the AI Opportunity Blueprint™ is right for you.
Book Your Call
© 2026 eMediaAI.com. All rights reserved. Terms and Conditions | Privacy Policy
