Every owner we meet in El Paso, Midland or Las Cruces asks a version of the same question: “What would AI actually do for us?” Not in theory. In our office, with our people, this quarter.

Here is the honest answer we give after walking a few hundred workflows: AI is excellent at a narrow set of things, useful as an assistant on a wider set, and a liability on the rest. Knowing which bucket a task falls into is most of the work.

The three buckets

We sort every task into one of three categories during an AI Opportunity Assessment:

  1. Automate. The task is repetitive, the inputs are predictable, and a wrong answer is cheap to catch. AI does it; a person spot-checks.
  2. Assist. The task needs judgment or context the machine doesn’t have, but a draft or a summary saves real time. AI does the first 70 percent; a person finishes.
  3. Keep human. The task is relationship-heavy, legally sensitive, or the cost of a confident wrong answer is high. AI stays out, or only organizes the information.

What goes in “Automate”

These are the tasks where we routinely see 5 to 20 hours a week come back:

  • Reading documents and typing what’s in them. Invoices into the accounting system, field tickets into billing, purchase orders into the job file, résumés into the applicant tracker. Modern models read PDFs, photos and emails reliably, and the output is a structured record you can check.
  • Classifying and routing. Which inbox does this email belong in? Which job does this receipt belong to? Is this support request urgent? A model can tag and route hundreds of these an hour.
  • First-draft communication at volume. Appointment reminders, status updates, “we received your request” replies, bilingual customer notices. The draft is generated from your data; the tone is set once.
  • Summarizing. Long email threads, meeting recordings, call transcripts, inspection reports. A reliable two-paragraph summary with action items.
  • Answering the phone for routine requests. Hours, directions, “is my order ready,” scheduling a callback. An AI receptionist like the one in our open-source Peal phone system handles the routine and hands off the rest.

What goes in “Assist”

Here AI shortens the task but shouldn’t finish it:

  • Estimates and quotes. AI can count devices on a drawing, pull your pricing, and build the first version. The estimator still owns the number. This is exactly how BidLark is built.
  • Proposals, scopes of work, policies, job descriptions. A good draft in minutes, then your expertise.
  • Research. Comparing vendors, summarizing a regulation, checking a spec sheet. Fast, but verify anything you’ll rely on.
  • Scheduling and dispatch suggestions. AI proposes; a dispatcher who knows the crews decides.
  • Data analysis. “Which jobs lost money last quarter and why?” AI will get you to the chart; a person decides what it means.

What stays human

  • Anything where a confident wrong answer is expensive: final pricing on a large bid, legal or medical determinations, safety sign-offs, firing decisions.
  • Relationship moments: the call to an unhappy customer, the negotiation, the first meeting with a new client.
  • Judgment about people: hiring, performance, discipline. AI can organize notes; it should not decide.

How to find your list

Don’t start with the technology. Start with the hours. For each role in your company, write down the five tasks that eat the most time each week. Then ask three questions about each one:

  • Is the input predictable (the same kind of document, request or event every time)?
  • Is a wrong answer cheap to catch?
  • Does the task need context that only lives in someone’s head?

Two yeses and a no puts the task in “Automate.” That short list is where your first AI project should live, and it is usually worth more than any tool your vendor is pitching.

If you’d like a second set of eyes, that’s what our assessment is for. We do them on-site across El Paso, Midland–Odessa, New Mexico and Southern Arizona. Get in touch and we’ll start with the hours.