Practical AI for small businessWhere AI helps a small business, and where it does not.

AI is good at drafting, sorting, summarizing and pulling details out of messy text. It is not a substitute for judgment, so the best uses keep a person in the loop and the cost in plain view.

By Shay McGill, CTO of Lagom Tech Solutions · Updated · 5 min read

Where AI earns its keep.

Forget the flashy demos for a minute. In a small business, AI is useful for a handful of plain jobs that involve reading or writing language. Five of them cover a lot of ground.

  • Drafting. A first version of a customer reply, a follow-up after a service call, a job posting or a proposal cover note. A person edits it and sends it.
  • Classifying. Sorting incoming requests by type and urgency. A property manager can tag maintenance requests as plumbing, electrical or general, and flag a leak as urgent.
  • Summarizing. Turning a long call, a meeting or a thread of notes into a short summary. An accounting firm can capture what a client asked for without anyone typing minutes.
  • Extracting. Pulling details out of documents that arrive in different formats. A job shop can read part numbers, quantities and due dates from emailed purchase orders into its order list.
  • Routing. Sending each item to the right person or queue. An HVAC shop can send warranty questions to the office and no-heat calls straight to dispatch.

The common thread: the work is repetitive, it is mostly words, and a person can check the result in seconds. When all three are true, AI is worth a serious look.

Where AI is the wrong tool.

AI writes fluent, confident text, and it can be wrong while sounding right. That makes it a poor fit for some jobs, no matter how good the demo looks.

  • Decisions that need judgment. Approving credit, denying a claim, pricing a large job or handling an upset customer. These need a person who knows the context and owns the outcome.
  • Anything that must be exactly right without review. Invoice totals, payroll, tax figures, medication details or legal filings. If a mistake is costly and nobody checks, do not hand it to AI.
  • Work a simple rule already handles. If every order over a set amount needs approval, that is a rule, not an AI problem. Plain code is cheaper, faster and behaves the same way every time.

Be honest about the alternatives too. If your email, accounting or office software already has an AI feature that does the job, use it. A custom build only makes sense when the task is specific to how your business works.

Keep a person in the loop.

The safest pattern is simple: AI drafts, a person decides. The AI saves the typing and the first pass. The person catches the mistakes and owns what goes out.

Good tools make that review easy. They show the source next to the output, so you can check a summary against the transcript. They make editing quick, and they keep a record of what was approved and by whom.

Start with nothing sent automatically to customers. Once the AI has handled a task well for a while, you can decide whether a low-risk step runs on its own. Make that call with evidence, not on day one.

How to keep AI costs under control.

AI has two costs: building it into your workflow, and running it. The running cost is usage-based. Each run costs a little, and it adds up with volume and with the size of what you feed it.

Three controls keep running costs predictable:

  • Run on a click. AI runs when someone asks for it, not automatically on every record, email or upload.
  • Show the estimate first. The person clicking sees roughly what the run will cost before it starts.
  • Set spending limits. Monthly and daily caps stop a bug or a busy week from turning into a surprise bill.

Per-seat AI add-ons work differently. If one costs $30 per user a month, 15 users is $5,400 a year, used or not. That is fine if everyone uses it daily, and wasteful if only two people do.

The subscription cost calculator adds up what your software subscriptions cost over 3, 5 and 10 years. It is a quick way to see the real price of a per-seat add-on.

If you are pricing an AI consultant or developer, ask for a fixed price for a defined scope. Then ask what the feature will cost to run each month at your volume, and what limits are built in. The guide to what custom software costs covers what drives a quote.

Questions to ask about your data.

Using AI means sending some of your data to an AI provider. Before you do, get clear answers to a few questions.

  • Which AI provider processes our data, and where?
  • Is our data used to train their models, and can we turn that off?
  • How long are prompts and outputs kept, and who can see them?
  • Can we limit what the AI sees to only the fields it needs?
  • Is there a log of what was sent and what came back?
  • If we stop using the tool, what happens to our data?

If you handle patient information, as a dental office or medical practice does, HIPAA is part of the picture. Ask whether the provider will sign a business associate agreement. Have your compliance advisor review the setup before any patient data is used.

What this looks like in our own software.

Lagom Studio builds and runs its own software, so we live with these choices every day. Two examples show the patterns above in practice.

The Lagom Studio client portal records meetings and transcribes them. AI then drafts meeting summaries and specs from the transcript. They are drafts for a person to review, not finished documents.

AI in the portal runs only when someone clicks. The cost is shown first, and there are monthly and daily AI spending limits. That design keeps AI spending a choice, not a background process.

PestelPro is an AI-powered market analysis product that writes PESTEL reports. PESTEL looks at the political, economic, social, technological, environmental and legal factors around a market. It suits AI well: a structured framework, lots of reading and writing, and a reader who judges the result.

How to start small.

The best first AI project is small, boring and easy to check. Here is a simple way to begin.

  • Pick one repetitive task that eats time every week, like sorting inbound requests or writing call summaries.
  • Try what you already have. The AI built into your current software may be enough. If it is, you are done.
  • Run a short pilot with a person reviewing every output. Note where it helps and where it gets things wrong.
  • Set a spending limit on day one, even for a test.
  • Expand only when it earns its keep. Add the next task once the first one is clearly saving effort.

If the task is tied to your own workflow and data, a small custom build may fit better. A paid discovery is a good way to find out before you commit to a build. The guide to how long custom software takes shows what to expect after that.

Lagom Studio builds practical AI into the everyday workflows of small and mid-sized businesses. Discovery starts at $1,500 and is credited against a Build signed within 30 days, and Builds start at $8,000. See full pricing or book a free problem review.

Questions

Common questions.

What are good examples of AI for a small business?

Good first uses are drafting, classifying, summarizing, extracting and routing. Examples include drafting replies to customer emails, sorting maintenance requests by urgency, summarizing client meetings, pulling part numbers and due dates out of emailed purchase orders, and sending each request to the right person. The best tasks are repetitive, mostly words, and quick for a person to check.

What should a small business not use AI for?

Avoid using AI alone for decisions that need judgment, such as approving credit or handling an upset customer. Also avoid it for anything that must be exactly right without review, such as invoice totals, payroll or legal filings. AI can sound confident and still be wrong. If a simple rule already handles the job, plain code is cheaper and more predictable.

How much does an AI consultant or AI automation project cost?

There are two costs: building the AI into your workflow, and running it. Ask for a fixed price for a defined scope, so the build cost is known up front. Then ask what it will cost to run each month at your volume, whether the cost is shown before each run, and what monthly and daily spending limits are in place.

How do we keep AI costs from getting out of hand?

Use three controls. Have AI run only when someone clicks, not automatically on every record. Show an estimate of the cost before each run. Set monthly and daily spending limits so a bug or a busy week does not turn into a surprise bill. For per-seat AI add-ons, multiply the price by your users and by twelve months to see the real yearly cost.

Is it safe to put our business data into AI tools?

It can be, if you know where the data goes. Ask which provider processes it, whether it is used to train their models, how long prompts and outputs are kept, who can see them, and whether you can limit what the AI sees. If you handle patient records, ask about a HIPAA business associate agreement and have your compliance advisor review the setup first.

Tell us the problem

Want a straight answer for your business?

Tell us what eats your team's time. We'll tell you honestly what it would take to fix, and whether custom software is even the right answer.

30 minutes, no pitch deck. Or email info@lagomtechsolutions.com.