Owners are being told they need AI. Every software company is adding it. Every week a new tool promises a faster, smarter business.
So owners ask a fair question: where should I use AI?
It starts in the wrong place. A better question is: what is costing us time, money, or customers that doesn’t need to?
The difference looks small. It changes which projects you start, which tools you buy, and how you judge whether they worked. AI isn’t the goal. A business with less friction is.
The trouble with starting from the tool
Say you hear that AI agents can answer email, connect your software, and handle routine admin. So you look for places to use them. You automate a few customer emails. You add a chatbot. You connect two apps.
Some of that helps. Some of it works perfectly and changes nothing.
A new capability doesn’t mean you have a problem it solves. “What can we automate?” produces projects. “What should we stop wasting time on?” produces improvements.
What the owner actually wants
Take a business that processes loan applications. Each one is reviewed, typed into a second system, checked for missing details, and partly redone when something is wrong. None of it is hard. All of it takes time.
An automation vendor might pitch an AI agent, a document-processing system, or a workflow across five apps. The owner wants none of those. They want the information to move without someone retyping it: faster turnaround, fewer errors, and a team that isn’t buried in admin.
The fix might be a feature in software they already pay for. It might be a plain automation between two systems. It might be AI, if the documents arrive in too many formats for fixed rules. The right answer depends on the problem, not the tool.
You’re not buying technology. You’re buying a change.
The best opportunities are boring
The AI conversation drifts toward autonomous agents and fully automated operations. The real value is usually in work your team does without thinking:
- Copying data from one app into another
- Chasing customers for missing documents
- Building the same report from three spreadsheets
- Answering the same five questions
- Digging through email for something a colleague needs
Each one looks small. Together they add up. A 10-minute task done 20 times a day is more than three hours of work, before interruptions and mistakes.
Not every minute comes back, and not every repeat task should be automated. But you don’t need an AI strategy to start. You need to notice where work is harder than it should be, and what that costs.
Know the cost before you automate
Repetition alone doesn’t justify automation. Two minutes once a month isn’t worth building and maintaining anything. Fifteen minutes, dozens of times a day, deserves a look.
Run the numbers. If someone spends 45 minutes a day moving information between systems, that’s 165 hours over 220 working days.
Value of time recovered = hours actually saved × cost per hour.
Compare that with the full cost of the fix: setup, subscriptions, maintenance, training, and the time someone spends on the cases it gets wrong.
Then be honest about what the saved time becomes. Five hours a week back doesn’t lower anyone’s salary. It pays off when the team uses it: more customers served, a shorter backlog, a hire you don’t need to make.
You don’t need a perfect ROI model. You need judgment instead of excitement.
Work backward from the outcome
Start with how things work today. Find the friction. Define what better looks like. Only then compare solutions.
Say customer inquiries are slow. Tool-first, you’d build a chatbot. Outcome-first, you’d ask:
- How many inquiries arrive a day, and how many are routine?
- How long do customers wait for an answer?
- Which ones need a person?
- Are people asking because the answer is hard to find on your website?
The answer might be a chatbot. It might be a clearer FAQ page, a reworked support process, or a few saved replies. Those options have different costs and risks. Starting from the outcome lets you tell them apart.
Sometimes a process needs automation. Sometimes it needs simplifying, better software, a clear owner, or nothing at all. The goal isn’t to use AI wherever you can. It’s to make unnecessary work unnecessary.
Five questions for your first AI project
You don’t need to know every model or platform. Pick one process and ask:
1. What work frustrates the team?
Look for tasks that repeat, take time, invite mistakes, or move data between systems. Ask the people who do them. They know where the friction is.
2. Why does the work exist?
Is it required by a customer or a regulation? Or is it there because two systems don’t talk, or because the process was built around a limit that no longer applies? Automating a bad process only makes it run faster. Sometimes removing a step beats automating it.
3. What does better look like?
Be specific. “Use AI to improve customer service” can’t be measured. “Answer routine questions within an hour without lowering quality” can. “Automate our paperwork” is vague. “Stop retyping standard applications, and keep a person reviewing the exceptions” is a target.
4. What is the simplest reliable fix?
Can you simplify the process? Use a feature you already have? Connect two systems with a plain automation? Use AI to read messy documents or free-text messages? Pair it with human review?
AI is good with language, varied documents, sorting, and summarizing. It can also be wrong in ways that fixed rules aren’t. When the rules are clear and mistakes are costly, a fixed workflow is often the better choice. Sensitive decisions stay with a person.
The best fix isn’t the most sophisticated one. It’s the one that works reliably at a cost you can justify.
5. How will you know it worked?
Measure before you change anything: time per task, error rate, response time. Measure again after. Did the backlog shrink? Did people get usable time back? Did the fix create work somewhere else? Count maintenance and exceptions too.
Success isn’t that the system runs. It’s that the business is better off.
Questions to ask an AI provider
Don’t start with “Which tools do you use?” Ask “How would you decide whether this problem is worth solving?” Ask how the fix will improve the process, and how you’ll measure it.
A good provider wants to understand your workflow, your people, and your constraints. They should also tell you when you don’t need AI. A small, low-risk problem can justify a quick experiment. Anything that touches customer data, money, or critical systems needs more care. The depth of the work should match the size of the problem.
That’s how we approach AI consulting: we weigh the options first, and we say so when a simpler step will do.
You don’t need to automate everything
Automation has running costs. Systems need maintenance. Workflows break when apps change. AI output needs checking. Exceptions still need a person.
Some work is valuable because a person does it: judgment, trust, attention. A customer may like an instant reply and still want a knowledgeable person for a hard problem. A team may use AI for research and keep the final decision. The aim isn’t to remove people. It’s to spend their time better.
Knowing when not to use AI is part of using it well.
Your real advantage is understanding your business
AI tools will keep improving. What takes a specialist today will be a standard feature next year, so knowing which tools exist won’t stay an advantage for long.
What lasts is knowing where your business makes money, where it loses time, and which changes matter most. An owner who understands a process deeply will spot a useful AI application faster than someone who knows fifty tools but not the work.
Start with one problem
You don’t need an AI strategy this quarter. Pick one recurring problem. Find out who it affects, why it exists, how often it happens, and what it costs. Decide what better looks like. Try the simplest reliable fix, then measure.
If AI is the answer, use it. If plain automation is better, use that. If the process should be redesigned or removed, start there.
The technology follows the problem.
Stop looking for ways to use AI. Start looking for problems worth solving.
Want a second opinion on your first problem? Our free assessment is 30 minutes with a senior member of our team, and you keep the one-page summary.
