If I could level with you.... most of what you've heard about AI is either trying to sell you something or trying to scare you. Funny enough, sometimes it's both at the same time.
I'm sure you've seen the headlines... AI is going to replace your workforce. AI is going to 10x your revenue. AI is going to write your emails, run your meetings, manage your projects, and heck, probably make your coffee soon.
And the part I'm not supposed to say... every software company you've ever bought from has practically slapped "AI-powered" onto their product page and raised their prices.
Here's what nobody is saying at those product release conferences: AI is a tool. That's it. A very capable tool, but a tool. And like every tool before it, it's only as useful as the problem you point it at and the process you wrap around it.
So what is it, actually?
AI, in the way most people encounter it right now, is software that recognizes patterns and generates output based on those patterns.
When you talk to ChatGPT or Claude, you're talking to a language model. It was trained on an enormous amount of text and learned the patterns of how words follow other words. It doesn't "think." It doesn't "understand" your business. It predicts what comes next based on what it's seen before, and, frankly, it's gotten VERY good at that prediction.
Pattern recognition at scale is genuinely powerful. It means a computer can now do things that used to require a human sitting there reading, writing, sorting, analyzing, deciding... much of the repetitive, pattern-based things that eat up hours of your week.
What it's good at right now
AI is good at tasks that involve processing language, recognizing patterns in data, and generating structured output from unstructured input. In plain terms:
Reading and extracting. You can hand it a stack of vendor invoices and it pulls out the line items, quantities, and totals. You can give it a contract and it finds the key dates and dollar amounts. Work that used to take someone an hour of careful reading takes seconds.
Drafting and formatting. It's able to write first drafts of emails, proposals, reports, SOPs. A human still needs to review, edit, and approve. But going from blank page to 80% draft in two minutes instead of forty is a real time savings. (And yes, we know who you are when you send that email without editing it)
Sorting and categorizing. Incoming support tickets routed to the right department. Expense reports categorized by type. Materials matched to catalog entries. Anywhere you have a person looking at something and deciding which bucket it goes in, AI can handle the straightforward cases and flag the edge cases for human review.
Answering questions about your own data. If your data is structured and accessible, AI can sit on top of it and answer questions in plain English. "What did we spend on lumber across all projects last quarter?" Instead of running a report or digging through spreadsheets, you can ask and get a quick answer.
What it's not good at
AI does not understand your business out of the gate. It has no context for why you made the decisions you made, what your relationships with vendors look like, what your team's strengths and weaknesses are, or what happened last time you tried something similar (unless you provide it that context).
It also cannot replace judgment. It can give you information faster, draft things faster, sort things faster. But the decision about what to do with that information still requires someone who knows the business - and it's telling when someone just trusts AI's judgement without applying their own.
AI hallucinates. That's the technical term for when it generates something that sounds right but is completely wrong. It will cite sources that don't exist, make up numbers, and present fabricated information with total confidence. Crazy enough, this is not a bug that will be fixed next quarter. It's a fundamental characteristic of how these models work. Everything AI generates needs verification.
And it can't fix a broken process. This is the part that matters most.
AI applied to a broken process just speeds up the broken process
If your materials tracking lives on notepads and in someone's head, adding AI doesn't fix that. What you'll get is just wrong answers, faster. You get automated confusion. You get a very expensive tool generating reports from data that was never accurate in the first place (this is literally the biggest mistake and best fix once identified).
The companies getting real value from AI also aren't the ones who bought the most AI tools. They're the ones who had clean data, documented processes, and structured systems before AI showed up. AI became an accelerant for something that was already working.
If you can't explain your process to a new hire succinctly, AI can't follow it either. If your data lives in five different places with five different naming conventions, AI will struggle with it the same way your team does. If the answer to "how do we handle 'X'?" is "ask Steve, he knows," then no amount of AI changes that dependency.
Systemization is needed first. Then we can actually apply technology to that system.
How to start
Don't start with "we need AI." Start with "where are we spending time on work that follows a pattern?"
And more importantly, answer the question "WHY would I apply AI 'here'?"
Look for the repetitive processes. The tasks where someone is doing the same decision-making steps over and over. Reading a document, pulling out certain fields, putting them somewhere else. Looking at incoming information and sorting it. Drafting similar communications repeatedly with minor variations.
Those are your first AI opportunities that will move the needle. Not because AI is magic, but because those tasks are pattern-based, and pattern recognition is exactly what these tools do well.
Then ask three questions before you buy or build anything:
Is our data clean enough? If the input data is messy, inconsistent, or scattered across systems, fix that first. AI trained on bad data produces bad output confidently.
What does the human review step look like? Never set up AI to take action without a human checkpoint. Draft and review. Suggest and confirm. Categorize and verify. The human has to stay in the loop.
How will we measure whether this actually helped? Time saved per week. Error rate before and after. Dollars recovered. Pick something concrete. "We feel more efficient" is not a metric.
The honest version
AI is going to change how most work gets done. It already is in the offices of every business that uses it. But the change looks less like a robot replacing your estimator and more like your estimator finishing bid prep in half the time because a tool pre-populated 80% of the quantities from the plans.
It looks like your office manager spending thirty minutes on vendor invoices instead of three hours. It looks like your project managers getting weekly materials reports generated automatically instead of someone building them by hand every Friday afternoon.
Small, specific, measurable improvements to processes that were already structured enough to improve. That's what AI does well right now. You don't need it to do everything. You need it to do the right things, consistently.
The companies that will get the most from AI over the next few years are not the ones buying the flashiest tools. They're the ones getting their processes and data in order today, so that when they do plug in AI, it has something solid to work with.
Start there.
-DC