Where to start with AI and automation in your business

You run a 30-person firm. Everyone tells you to "use AI," two of your competitors just announced they're doing it, and your inbox is full of tools promising to automate everything. So you open a demo, and twenty minutes in you still can't answer the only question that matters: where do you actually start with AI and automation? The honest answer is that you don't start with AI at all. You start with one workflow: the one quietly costing you the most predictable hours every week.

Quick Answer: Start with diagnosis, not tools. Find one workflow that is repetitive, rule-based, high-volume, and digital (client intake, invoicing, or reporting) and automate the handoffs inside it. Pick the process that wastes the most predictable hours, prove one win, then expand. The tool comes last.

Where should you start with AI and automation?

Start with your most repetitive, rule-based, high-volume workflow, not with a model or a platform. The right first project is a single process with clear boundaries: turning a new enquiry into an opened file, or a completed job into a paid invoice. Pick the one that wastes the most predictable hours, automate the handoffs, and prove it before touching anything else.

The reason to start narrow is that the potential is genuinely large, and that makes it easy to over-reach. McKinsey's 2025 Global Institute analysis found that 57% of U.S. work hours involve activities that could be automated with technology that already exists. But "could be automated" is not a to-do list. The point of starting with one workflow is to convert a huge, vague number into a specific hour count you can actually recover.

It helps to know where those hours hide. One analysis of how the workday is actually spent, drawing on Asana's Anatomy of Work Index, found employees give about 62% of their time to repetitive, coordination-heavy tasks and only 27% to the skilled work they were hired to do. Your first automation target is sitting inside that 62%.

Why do most AI projects start in the wrong place?

Most projects fail before they begin because they start with a tool and hunt for a use, instead of starting with a painful process and finding the right fix. Buying software first is backwards: you end up paying for a solution and then reverse-engineering a problem to justify it.

The data shows the pattern clearly. The U.S. Small Business Administration's Office of Advocacy report on AI adoption (September 2025) found small-business AI use climbing fast, from 6.3% to 8.8% in six months, yet about half of the small firms already using AI had invested nothing in it: no training, no integration, no process change. The areas where they lagged furthest behind were hiring help to integrate the tools and training staff to run them. In other words, the gap isn't access to AI. It's knowing which process to point it at and doing the work to connect it.

There's a second trap underneath the first. If a process doesn't make business sense, automating it just makes the mess run faster. A broken intake form that collects the wrong information will now collect the wrong information automatically, at scale. Fix the process, then automate it. Never the other way around.

How do you find which workflow to automate first?

Run every candidate process through four questions. If a workflow passes all four, it's a strong first project. If it fails one, it's either not ready or not worth it yet.

The fastest way to find candidates is to stop guessing and track where a week actually goes. Most owners are surprised. A large share of the day disappears into hidden, low-visibility work: a Quickbase survey found nearly 70% of U.S. workers lose 20 or more hours a week just searching for information across disconnected systems. That data-hunting is invisible on any org chart, and it's often the highest-value thing to fix first. For a deeper ranking framework, see which processes to automate first.

What are the best first automations for a small business?

The best first automations live in the handoffs between systems: the moments where a human copies information from one place to another. Client intake, invoicing and billing follow-up, employee or customer onboarding, and recurring reporting are the usual winners because they're high-volume, rule-based, and already digital.

By workflow, the strongest starting points look like this:

• Client or lead intake. Capturing a new enquiry, running a conflict or qualification check, and opening the file, instead of a paralegal or coordinator re-typing details into the case or CRM system.
• Invoice-to-payment. Generating invoices from completed work and sending the follow-up sequence automatically, so nobody chases payment by memory.
• Onboarding. Turning a signed client or new hire into the full sequence of documents, accounts, and tasks without someone running a mental checklist.
• Reporting. Pulling numbers from the tools where they already live into one weekly view, instead of rebuilding the same spreadsheet every Monday.

This is not theory for us. In the law-firm audits we run, every single one has uncovered ten or more automation opportunities, and the highest-ROI ones are rarely the ones the partners expected walking in. On the operations side, we built a single source-of-truth data layer for a 500-employee New York real estate company, because the first real problem wasn't a fancy AI feature: it was that the same information lived in five places and never agreed. Finding problems like that is exactly what an AI automation audit includes.

What should you not automate yet?

Avoid automating anything that needs human judgment on every instance, happens too rarely to repay the setup, or is still visibly broken. These are the processes that look automatable in a demo and turn into a maintenance headache in production.

Three to leave alone for now:

1. Judgment-heavy calls. Negotiation, sensitive client conversations, hiring decisions, anything where the "rule" changes with context. AI can draft and prep here, but it shouldn't decide.
2. Rare exceptions. A process you run four times a year usually costs more to automate than it saves. Do it by hand and spend the effort on the weekly grind instead.
3. Broken processes. If the workflow is unclear, inconsistent, or produces the wrong output today, automation locks in the error. Redesign first.

Naming what not to touch is half the value of a good starting plan. It keeps your first project small enough to actually finish.

How do you go from idea to a working automation?

Once you've picked the workflow, the path from idea to running automation is short and deliberate. The goal of the first project is one clean win you can measure, not a full overhaul of how the business runs.

1. Baseline the hours. Measure how long the current process takes, how often it runs, and where it breaks. This is your "before."
2. Scope one workflow inside 30 days. Draw hard boundaries: "automate the intake step," not "automate sales." A tight scope is what makes it finishable.
3. Pilot for six to ten weeks. Run the automation alongside the manual process, against your baseline, so you catch the edge cases before you rely on it.
4. Measure against the baseline. Compare hours, errors, and turnaround to the "before" number. If it moved, you have proof. If it didn't, you learned cheaply.
5. Then expand. Use the first win to fund and justify the next one. Confidence compounds faster than any single tool.

That sequence is deliberately boring. Boring is what ships. For worked examples in a professional-services setting, see what law firms can automate.

---

Not sure which of your workflows to automate first? That's the exact question our AI assessment answers. A few questions, a couple of minutes, and you get a personalized preview of where automation would pay off in your business, before you spend anything or commit to a single tool. Start assessment.

Frequently asked questions

What is the first thing a business should automate with AI?

The first thing to automate is a single, repetitive, rule-based workflow where data already lives in your systems: commonly client intake, invoicing follow-up, onboarding, or weekly reporting. Pick the one process that wastes the most predictable hours each week, automate the handoffs inside it, and prove one measurable win before expanding.

How do I know if a process is worth automating?

Run it through four tests. Is it repetitive (daily or weekly, not rare)? Is it rule-based (you can describe the steps clearly)? Is it digital (the data lives in your tools already)? Is it measurable (you can baseline the hours or errors first)? A process that passes all four is a strong candidate. If it fails one, it's usually not ready.

How much does it cost to start with AI and automation?

The software subscription is almost never the real cost. Scoping and integration are: connecting the systems that need to talk to each other and making sure you're automating a process worth keeping. Most first projects are modest by design: one workflow, a fixed scope, a measurable result. The expensive mistake is paying to automate a process you should have fixed or eliminated first.

Do I need technical staff or a developer to start?

No. The hardest part of getting started is not technical: it's choosing the right workflow and mapping how it actually runs today. That's a business decision, not a coding one. Most 10-to-70-person firms bring in outside help to scope and build the first automations, then run them in-house once the process is proven.

What's the difference between AI and automation?

Automation follows fixed rules: when X happens, do Y. It's ideal for predictable, repeatable steps like moving data or sending a follow-up. AI adds interpretation: reading a message, classifying a request, drafting a reply. Most useful first projects combine both: automation handles the flow, AI handles the one step in the middle that used to need a person's judgment.

How long before an automation pays off?

A well-scoped first automation usually shows results within its six-to-ten-week pilot, because you measured the "before" and can see the "after." The payoff is fastest on high-frequency workflows: a task you run fifty times a week recovers its setup cost far quicker than one you run monthly. That's why frequency is the first test.

Will automating a process mean cutting staff?

Usually the opposite of what people fear. The point is to move people off the repetitive 62% of the day and onto the skilled work they were actually hired for. In practice, most first automations remove the tedious handoffs and re-typing, so the same team handles more volume without more headcount, not fewer people doing the same low-value work.