Don't Ask What AI Can Do: Discover Where it Truly Adds Value

TL;DR

• Any AI initiative starts with understanding your business.

• Most botched AI initiatives don’t fail because the models are bad. They fail because the business problem was never clearly defined. 

• Before you think about tools, vendors, or algorithms, break your business down into how value is actually delivered.

• Find where time and money get stuck, and let that bottleneck define the AI use case. The technology comes last.

Discover Where AI Adds Value

If you start an AI initiative by asking, “What can AI do for us?” you are already behind. The most critical step in identifying a strong AI use case has nothing to do with artificial intelligence. It starts with a clear, unvarnished understanding of how your business actually operates day to day. You don’t need data scientists. You don’t need machine learning engineers. And you certainly don’t need to start by shopping for tools.

What you do need is the discipline to break your business into its essential moving parts: how customers enter the system, how work flows through it, and where time, money, or attention quietly leaks out along the way.

The framework below applies across industries. Whether you run a field service company, a manufacturer, a professional services firm, or a software business, the logic is the same. Five steps. Business first. Technology second.

Step 1: Define the One Metric That Matters

Every process exists to deliver value to a customer. The first question is simple: How do you know you’re doing that effectively?

The answer should be a single operational metric. The metric isn’t a vanity KPI nor a dashboard full of averages. Pick one number that captures whether the system is working. Think of this as your operational North Star. It should be:

• Directly tied to customer experience.

• Measurable in real time or near real time.

• Influenced by day‑to‑day decisions.

Any discussion of AI is premature until that metric is clear.

Step 2: Chart the Customer Journey

Work backward once you’ve defined that North Star metric. Map the major stages a customer moves through to reach that outcome. Stay at the 10,000‑foot level. This isn’t a detailed workflow yet. You are identifying the value chain: the sequence of activities required to deliver the service or product from the customer’s perspective. At this stage, clarity beats precision. Four to six stages is usually enough.

Step 3: Break Activities Into Real Processes

This is where things start to get productively uncomfortable. Expect each high‑level activity to contain a handful of repeatable processes. These processes are gears that make the machine work. The phone calls, handoffs, lookups, checks, approvals, and follow‑ups that actually move work forward.

List these processes explicitly. If a step depends on someone’s memory, experience, or intuition, write that down. These details matter later.

Step 4: Measure Time and Resources

Now ask the most revealing operational question: Where does the time actually go? For each process, estimate:

• Cycle time (from start to finish)

• People involved

• Cost or opportunity cost

• Variability

Don’t aim for perfect data. It’s a red herring, and perfect data isn’t useful at this stage. Interviews, shadowing, and rough time studies are often enough to surface the pattern. What you’re looking for is not the longest step, but the most variable one. That’s usually where systems break under growth.

Step 5: Identify the Bottleneck

At this point, the AI use case often reveals itself. The bottleneck is the process that:

• Is the most variable.

• Relies heavily on human judgment or memory.

• Scales poorly as volume increases.

• Directly degrades your core metric.

This is where technology can create leverage.

A Practical Example: Riverbend Mechanical Services

To make this concrete, consider a hypothetical field service company: Riverbend Mechanical Services.

Riverbend provides residential and light commercial plumbing and HVAC services. It employs roughly 40 people and generates about $12 million in revenue annually. Phones ring constantly. Business is steady. Yet margins are flat, and customers increasingly complain about slow response times.

The Metric

After the discussion, ownership agrees that they are in the business of speed. If customers wait too long, they call someone else. Riverbend’s North Star metric is Time‑to‑Resolution: the elapsed time from the initial customer call to the signed invoice.

The Activities

To move a customer through Time‑to‑Resolution, four major activities occur:

1. Intake: Receiving and logging the request.

2. Dispatch: Assigning the job.

3. Execution: Completing the work on‑site.

4. Close‑out: Billing and payment.

The Processes

Each activity breaks down further:

1. Intake: Answering calls, entering data into the CRM, and verifying addresses.

2. Dispatch: Reviewing technician skills, checking locations and traffic, confirming job status, and assigning work.

3. Execution: Driving, diagnosing, checking inventory, and installing parts.

4. Close‑out: Final paperwork and payment processing.

The Time Audit

When Riverbend reviews cycle times, a clear patterns emerge:

• Intake is fast and standardized.

• Execution is time‑intensive but predictable.

• Close‑out is quick.

The Dispatch stage’s cycle time, however, ranges from 45 minutes to over two hours. One senior dispatcher mentally juggles technician locations, certifications, traffic patterns, and job complexity. Nothing is written down. Everything lives in one person’s head.

The Bottleneck

Dispatch is not just slow. It is fragile. Growth means more calls, more technicians, and more complexity. All without increasing decision capacity.

The problem statement becomes obvious: We need a way to match incoming jobs to the nearest qualified technician without relying on one person’s memory. No one asked for “AI.” The business asked for speed and reliability.

From Bottleneck to AI Use Case

Only now does technology start to enter the conversation. Riverbend doesn’t need experimentation or massive technological investments. It needs an AI‑enabled scheduling system that can read job requirements, evaluate technician skills and availability, and optimize routing in real time

Instead of hiring another dispatcher at roughly $60,000 per year, Riverbend could use a scheduling optimization tool for a few hundred dollars per month.

Why This Framework Generalizes

This approach is not specific to field services or scheduling algorithms. It doesn’t start with models. Because it starts with operations and metrics, it applies broadly:

• E‑commerce: Identifying friction in product discovery or support routing.

• Manufacturing: Reducing downtime or defects through better monitoring.

• Software: Accelerating deployment or ticket resolution.

• Professional services: Automating document review or routine analysis.

The pattern is consistent. AI creates the most value where human judgment is overused in systemic tasks.

Final Thought

We’ll close as we started: If you start an AI initiative by asking, “What can AI do for us?” you are already behind. The better question is: Where is our business slow, fragile, or expensive, and why?

Answer that honestly, and the AI use case tends to write itself. In our next post, we’ll dive into the different types of friction small- and medium-sized businesses face and how AI tools relate to them.

Recap

• Business First: AI initiatives must be grounded in business realities.

• Define to Succeed: Precision in the problem statement prevents failure more than the quality of the code.

• Value Before Tools: Understand your workflow before you pick your toolkit.

• Tech is the Final Piece: Use your biggest bottlenecks to guide your AI strategy, not the other way around.

We value your insights: If you're interested in joining our AI community of practice for small- and medium-sized businesses, please provide your information via the link below.

I'd Like to Learn More about the Community of Practice

Dan Lindberg is the founder and principal of Applied Economic Insight ® LLC. His insights have been featured in Discover Artificial Intelligence, Business Economics, and Homes.com. He served as the first Director of the Applied AI Lab at Waukesha County Technical College and currently teaches business analytics and data science courses at Marquette University and Alverno College. In 2022, Dan won the Contributed Paper Award from the National Association for Business Economics. Dan has a Master's Degree in Applied Economics from Marquette University and has over a decade of experience consulting in predictive analytics.

Raj Muthupandiyan is an applied technology advisor and the founder of Arivu LLC, where he is dedicated to turning the complexity of AI into practical results for the SMB community. With over two decades of experience across manufacturing, financial services, and service-based industries, Raj understands that for a small or mid-sized business, technology is only valuable if it drives growth and operational efficiency. He focuses less on the "hype" of emerging tools and more on the real-world constraints that leaders face, designing solutions that are secure, sustainable, and realistic. Raj holds a Master’s degree in Computer Science from Marquette University.