How to Figure Out Where AI Actually Fits in Your Business
Ai Getting Started • Updated • 16 min read

How to Figure Out Where AI Actually Fits in Your Business

Nearly 82% of the smallest businesses not planning to use AI say it isn't applicable to them. Here's how to find where AI actually fits in your business, starting with the work you already do every week.

Working out where AI actually fits in your business starts with the tasks you already do every week, and most owners haven't looked there yet.

Among businesses with fewer than five employees that weren't planning to use AI, nearly 82% said it isn't applicable to them. The SBA Office of Advocacy's September 2025 research spotlight reads that as something other than a technology gap. It suggests AI products are designed for larger businesses and that small owners have less time to learn about new technology, which makes it closer to an education gap. Both the tools and the use cases exist; the disconnect is that most small business owners picture AI as enterprise chatbots and self-driving cars, not the three hours they spend every week copying data between spreadsheets.

Meanwhile, the U.S. Chamber of Commerce reported in 2026 that 66% of small businesses now use AI in some form. And in a Salesforce survey, 91% of small and medium businesses using AI said it boosts their revenue. The real gap is between businesses that found the right entry point and businesses that are still looking at the wrong part of the map, more than between businesses that can use AI and businesses that can't. Size still matters: Eurostat found that 17% of EU enterprises with 10 to 49 employees used AI in 2025, against 55.03% of those with 250 or more.

Line chart: EU enterprises using AI, 2021 to 2025, rose from 6.12% to 17.00% for small firms and 28.41% to 55.03% for large
AI use nearly tripled in small EU enterprises from 2021 to 2025, but large ones still used it at about three times their rate.
EU enterprises using AI, by size, 2021 to 2025 (Eurostat, Artificial intelligence by size class of enterprise (isoc_eb_ai), updated June 15, 2026; Statistics Explained, Use of artificial intelligence in enterprises, data extracted December 2025)
Survey year250 or more employees (%)50 to 249 employees (%)10 to 49 employees (%)
202128.4%12.6%6.1%
202330.5%13.1%6.4%
202441.2%21%11.2%
202555%30.4%17%

The short version: to find where AI actually fits, list the repetitive work that eats your week, score each task on time, error cost and data readiness, and test AI on the best-scoring low-risk task for two weeks before you buy anything bigger.

  • List tasks, not tools: what you repeat every day or week.
  • Score each one on hours per week, what a mistake costs, and whether the inputs are already digital.
  • Pick a boring, low-stakes task with clean data for the first test.
  • Measure time and accuracy before and after, including editing time.
  • Keep it, change it or stop it, then move to the next task on the list.

This article is about how to figure out where AI would actually remove friction you've already accepted as normal, rather than which AI tools to buy.

Start with problems, not products

The instinct is to start with a tool. Someone on your team sends a link to a new AI product. You watch a demo. You try to figure out where it fits. This is backwards, and it's the single most common reason AI projects fail inside small businesses.

Instead, start with a pain audit. Sit down (alone, or with your team if you have one) and list the tasks that eat the most time relative to the value they produce. You're looking for work that is:

  • Repetitive: you do essentially the same thing every day or week, with minor variations
  • Rule-based: a set of clear instructions could describe 80% of the decision-making
  • Time-consuming relative to its complexity: it takes 45 minutes but isn't intellectually difficult
  • Error-prone under fatigue: mistakes creep in when volume is high or attention is split

Common examples: invoice data entry, appointment scheduling follow-ups, first-draft content for routine communications, categorizing incoming emails or support tickets, reformatting data between systems, generating standard reports from the same data sources weekly.

None of these are glamorous, and that is why they are good candidates. The root cause of most operational drag is usually unexamined repetition rather than a missing tool.

Pink lit louvered panel and an open notebook on a dark floor, starting with problems to find where AI fits in your business
The first step is mapping the work you've stopped questioning, before picking any AI product, which is where you find where AI actually fits in your business.

Score your opportunities before you spend anything

Once you have a list, score each item across three dimensions:

  1. Time consumed: How many hours per week does this task take across your team? Estimate conservatively.
  2. Error cost: What happens when this task is done wrong? A minor inconvenience, a missed deadline, a compliance issue, a lost client?
  3. Data readiness: Is the input for this task already digital and structured, or does it live in someone's head, a filing cabinet, or 14 different apps?

High time, high error cost, and high data readiness is your best starting point. High time but low data readiness means you have a systems problem to solve first, and that's fine. Knowing that saves you from buying an AI tool and then spending three months cleaning data before you can use it.

This scoring method is deliberately simple. You don't need a framework from a consulting firm. You need fifteen minutes with honest numbers.

The data readiness question matters more than you think

Gartner predicts that through 2026, organizations will abandon 60% of AI projects that aren't supported by AI-ready data, and in its July 2024 survey of 1,203 data management leaders, 63% said they don't have, or aren't sure they have, the right data management practices for AI. In those projects the model was fine and the information it was working with was the problem. This is why understanding your own technical landscape comes first.

If your customer records are split across a CRM, a spreadsheet, and an email thread, no AI tool will magically reconcile them. The first project might be "get our data into one place so AI can eventually work with it" instead of "deploy AI," and that is foundational progress.

Glowing pink cables with bright copper tips on black, scoring which business problems AI should connect to before spending
AI is only as good as the data infrastructure underneath it. Clean inputs produce useful outputs.

Run a small experiment, not a transformation

The word "transformation" should be a red flag when anyone uses it in the same sentence as AI and small business. You don't need a transformation. You need one task to work better than it does today. In a Census Bureau study of US firms using AI from November 2025 to January 2026, 57% used it in three or fewer business functions, most often sales and marketing (52%).

Bar chart: of US firms using AI, 52% use it in sales and marketing, 45% strategy, 41% IT and 40% research and development
Even the most common function reaches only about half of AI-using firms, and 24% of them use AI in a single function.
Where US firms that use AI are using it (U.S. Census Bureau, CES Working Paper 26-25R, The Microstructure of AI Diffusion (Bonney, Breaux, Dinlersoz et al.), October 2026 revision of the April 2026 paper; BTOS AI supplement, reference period Nov 2025 to Jan 2026)
Business functionFirms using AI in any function (%)
Sales and marketing52%
Strategy and business development45%
Information technology41%
Research and development40%

Pick the highest-scoring item from your audit. Find the smallest possible way to test whether AI improves it. That might mean:

  • Using a generative AI tool to draft the first version of routine client emails, then editing for accuracy and tone
  • Setting up a simple automation that categorizes incoming inquiries by type before they reach your inbox
  • Having an AI summarize your weekly reports so you spend ten minutes reviewing instead of forty minutes compiling

Set a timeframe: two weeks is enough. Measure the before and after honestly. How much time did you save? Did accuracy change? Did the output require so much editing that the savings disappeared?

This is the automate-or-hire question applied to AI specifically. Sometimes the answer is that the task needs a human with better tools. Sometimes it's that a $20/month subscription handles 80% of the work. You won't know until you test with your actual work, not a demo dataset.

The skills gap is real, and it's not what you think

In the World Economic Forum's Future of Jobs Report 2025, 63% of employers named skill gaps as a major barrier to transforming their business between 2025 and 2030, the most cited barrier in the survey. Prompting ChatGPT is the easy part of that gap; the harder part is that most teams can't evaluate whether AI output is correct for their specific domain.

An AI can draft a contract amendment. Can anyone on your team identify when it's introduced a clause that conflicts with your state's UCC provisions? An AI can generate a financial summary. Does someone on your team know enough to catch when it's misclassified a revenue category?

The skill that matters most is knowing your domain well enough to validate the output, more than operating the tool. This is why AI amplifies expertise rather than replacing it. The people who benefit most from AI in their workflow are the ones who already know what good output looks like; they just don't want to produce it from scratch every time.

If your team doesn't have that domain expertise yet, building operational culture around quality checks matters more than which AI platform you choose.

Don't start with the hard problems

The temptation is to throw AI at the thing that's causing the most pain. Resist this. The most painful problems in your business are usually painful because they involve ambiguity, judgment calls, incomplete information, and high stakes. These are exactly the problems AI handles worst.

Start with the boring problems. The ones where the answer is almost always the same, the inputs are clean, and the consequences of a mistake are low. Build confidence and measurement habits on low-stakes work. Then graduate to harder problems as you learn what works and where the edges are.

This is counterintuitive, and it matches how firms actually use AI: in the Census data above, 57% of AI-using firms use it in three or fewer business functions. Small, proven starts are the norm.

Hot pink dandelion seed head shedding white seeds against black, starting AI with small easy wins instead of hard problems
Small, deliberate starts outperform ambitious rollouts. One proven use case builds momentum for the next.

What this looks like in practice

Illustrative example: a business owner runs a five-person professional services firm. Three hours a week go to reformatting client data from intake forms into their project management tool. The process is manual, the fields are consistent, and errors (wrong dates, misspelled names, missing fields) cause downstream problems that take longer to fix than the original entry took.

That's a high-score opportunity: repetitive, rule-based, time-consuming for its complexity, error-prone, and the data is already digital. An AI-assisted workflow (or even a structured automation without AI) could reduce those three hours to twenty minutes of review time.

That recovered time adds up to 130 hours a year. For a five-person firm, that's meaningful capacity returned to work that actually requires human judgment.

The useful question is "where would AI remove friction we've accepted as normal," rather than "should we use AI." The answer is almost always more mundane and more valuable than the pitch decks suggest.

Why so many owners can't see where AI fits

The "not applicable" answer is common for a reason. The U.S. Chamber of Commerce Foundation's Main Street AI Monitor, an Ipsos survey of 1,070 people working at small businesses in May 2026, found that half of them already use AI at work. Yet 41% said it's unclear how AI applies to their specific business, and only about one in ten had been offered formal AI training.

How people use it explains a lot. In the same survey, 64% of small business AI users said their main use is personal productivity, such as drafting, summarizing and brainstorming, 26% mainly use it to help with recurring tasks, and only 6% use AI to automate workflows with minimal human involvement. Most people meet AI as a chat window. Very few see it attached to the invoice queue, the intake form or the weekly report, which is where it usually saves the most time in a small business.

Bar chart: 64% of small business AI users mainly use it for personal productivity, 26% recurring tasks, 6% automation
Only 6% of small business AI users have it running workflows, so most have not yet tried AI on the repeated tasks where it often saves the most time.
How small business workers mainly use AI (U.S. Chamber of Commerce Foundation, Main Street AI Monitor (with Ipsos), June 17, 2026; survey of 1,070 small business workers, May 2026)
Primary useAI users naming it their primary use (%)
Personal productivity64%
Recurring tasks26%
Automating workflows6%

That's why the method in this article starts with your work instead of with the tools. Once you have a list of real tasks in front of you, the question of where AI actually fits stops being abstract.

A one-page worksheet for finding where AI actually fits in your business

You can run the whole exercise on one sheet of paper or one spreadsheet tab. Give each task a row and these columns:

  • Task: written as a verb and an object, such as "enter supplier invoices" or "answer pricing inquiries".
  • Hours per week: across everyone who does it, estimated low.
  • Error cost (1 to 5): 1 is a typo nobody sees, 5 is a compliance problem or a lost client.
  • Data readiness (1 to 5): 1 is "it's in someone's head", 5 is "it's already in one system in the same format every time".
  • Checkable (yes or no): can a person tell in under a minute whether the output is right?

Sorting the rows to see where AI actually fits in your business

Then sort. The tasks that belong at the top have high hours, clean data, and output a person can check quickly. If a mistake would be expensive, the task can still go on the list, but it needs a review step built in from day one. Tasks that score low on data readiness move to a separate list of systems to fix first.

Here's how three rows might look for a small accounting practice (an illustration, not survey data):

  • Entering client receipts into the bookkeeping system: 6 hours a week, error cost 3, data readiness 4, checkable yes. A strong first test.
  • Drafting the monthly client update email: 2 hours a week, error cost 2, data readiness 5, checkable yes. An easy second test.
  • Advising a client on whether to restructure their business: 3 hours a week, error cost 5, data readiness 2, checkable no. Not a place to start.

Qurtoo's task inventory playbook uses a similar scoring sheet for teams (how often a task happens, the risk of a mistake, how easily the output can be checked) and ends with three pilot tasks and a "never" list, which is a useful format if you're running this with a group.

Where AI tends to fit by type of work

Every business is different, but the same kinds of work come up again and again when owners run the worksheet. In the Census Bureau study above, sales and marketing was the most common place firms used AI (52% of AI-using firms), followed by strategy and business development (45%) and IT (41%). In a small business, the useful tasks inside those categories tend to look like this:

  • Front office: sorting incoming inquiries by type, drafting replies to common questions for a person to approve, sending appointment reminders and follow-ups.
  • Sales and marketing: first drafts of routine emails, proposals built from a standard template, summaries of call notes, turning one article into several social posts.
  • Finance and admin: pulling fields from invoices and receipts, categorizing transactions for review, flagging duplicates, preparing the same weekly report from the same sources.
  • Operations: summarizing long documents, turning meeting recordings into action lists, moving data between systems that don't talk to each other.
  • Customer service: searching your own policies and past answers to suggest a reply, with a person sending it.

Where AI usually doesn't fit yet is just as useful to know: decisions that depend on judgment and context, situations you haven't seen before, anything you can't check quickly, and anything where a wrong answer reaches a customer or moves money with nobody looking. I wrote a longer list in the tasks you should never automate.

Decide how much the AI should do: suggest, draft or act

Finding the task is half the decision. The other half is how much of it the AI does on its own. Brian's five-rung ladder for agentic AI is a helpful way to think about it: a chatbot answers when asked, a copilot suggests and a person acts, a workflow runs a multi-step task inside one area, and the higher rungs reason across systems with far less oversight. He argues the workflow rung delivers the most practical value right now, and that skipping rungs is where costly failures happen.

For a first project in a small business, the copilot or workflow level is usually right. The AI drafts the reply, fills in the fields or prepares the report, and a person approves anything that goes to a customer or touches money. LTFI's write-up on agentic operations shows how that split works in practice, with routine work running on its own and approvals required for anything customer-facing or financial.

Check the data and the rules before the pilot

Before you start a two-week test, answer four questions about the task you picked:

  • Where does the input come from, and is it in the same format each time?
  • Does the task involve personal information about customers or employees?
  • Which tool will see that information, and what does its vendor do with it?
  • Who checks the output, and where is the check recorded?

The second and third questions matter more in Massachusetts than people expect. If the task involves personal information about state residents, 201 CMR 17.00 already requires you to oversee service providers, including "requiring such third-party service providers by contract to implement and maintain such appropriate security measures for personal information." An AI tool that reads client records is a service provider. My guide to Massachusetts WISP requirements covers what that means for your written program, and a short AI acceptable use policy tells your team which tools are approved and what data can go into them.

How to judge a vendor demo against your own work

If you start with problems, you'll still end up watching demos. The difference is that you can now test them against a task you've scored. A demo is built to look good on the vendor's sample data, so bring your own:

  • Hand the vendor three real examples of the task (with personal details removed), including one messy one, and ask to see the tool work on them live.
  • Ask what happens when the tool is wrong or unsure. Does it say so, or does it produce a confident answer anyway?
  • Time the whole task, including the minutes a person spends checking and fixing the output.
  • Ask where your data goes, how long it's kept and whether it's used for training.
  • Ask how you'd get your data and settings out if you stopped paying.

A tool that handles your messy example well, admits when it isn't sure and keeps editing time low is worth a pilot. One that only shines on its own sample data isn't ready for your work yet. I go deeper on this in how to evaluate AI tools without getting sold.

After the pilot: keep, change or stop

At the end of two weeks, compare the before and after numbers and make one of three calls. Keep it if time went down, accuracy held and the person doing the work wants to continue. Change it if the time savings are real but the output needs too much editing; that usually means a better prompt, a template or a different task. Stop it if the numbers didn't move, and write down why, so the next idea doesn't repeat the same mistake.

Then go back to the worksheet and pick the next task. Over a few months, the list of places where AI actually fits in your business gets longer and more specific, and so does your sense of where it doesn't. Once you have several tools running, a quarterly review keeps them honest; I lay out a simple version in how to audit what AI is actually doing in your business.

Finding where technology (AI or otherwise) fits is the advisory work we do at Kief Studio, and a system that has to run every day is our AI automation work.


Neural pathways splitting into glowing decision branches on black, what finding where AI fits looks like in practice
The first step in AI adoption is identifying the tasks that are manual, repetitive, and error-prone, before choosing a tool. Start where the evidence is strongest and the risk is lowest.
Laptop keyboard macro with hot pink LED backlight on black, the everyday mechanics of fitting AI into business operations
Most AI implementations that fail had a problem that wasn't well-defined before the solution was purchased; the technology itself usually worked.

Related reading

Frequently asked questions about where AI actually fits in your business

How much does it cost to start using AI in a small business?

Many useful AI tools start at $20-50/month. The bigger cost is the time spent figuring out where AI fits and ensuring your data is ready for it. A focused pain audit (described above) costs nothing and prevents you from spending money on tools that don't match your actual workflow. Start with free tiers and trials applied to your highest-scoring tasks before committing to paid plans.

What if our data is a mess? Should we wait to explore AI?

No, but your first project might be a data cleanup rather than an AI deployment. Gartner predicts that through 2026, organizations will abandon 60% of AI projects that aren't supported by AI-ready data. Knowing your data isn't ready is useful information. It tells you exactly what to fix first, and some AI tools can actually help with the cleanup process (deduplication, format standardization, gap detection). The audit step above will surface this before you waste money.

Is AI going to replace my employees?

For small businesses, AI is far more likely to give your existing team capacity back than to eliminate roles. In the Census Bureau study of firms using AI from November 2025 to January 2026, 66% used it only to augment tasks, and AI-related employment decreases happened in just 2% of firms. The tasks AI handles well (repetitive, rule-based, high-volume) are rarely the tasks your best people want to be doing. Redirecting their time to judgment-intensive, relationship-driven, or creative work is where the return shows up.

How do I know if an AI tool is actually working for my business?

Measure before and after on three metrics: time spent on the task, error rate, and whether the output required significant human editing. If an AI drafts something in two minutes but you spend thirty minutes fixing it, the net savings are negative. Set a two-week evaluation window with honest tracking. If the numbers don't improve, the tool isn't the right fit for that specific task; try it on a different one or try a different approach.

How do I figure out where AI actually fits in my business?

List the tasks you and your team repeat every week, then score each one on hours spent, what a mistake costs and whether the inputs are already digital and consistent. The best first fit is a task with high hours, clean data and output a person can check quickly, such as entering invoice data, drafting routine emails or summarizing a weekly report. Test AI on that one task for two weeks, measure time and accuracy before and after, and only then decide whether to keep it, change it or move on.

How do you evaluate an AI vendor demo beyond the surface?

Bring your own work. Give the vendor a few real examples of the task you scored, including a messy one, and watch the tool handle them live. Ask what it does when it's wrong or unsure, time the full task including checking and fixing, and ask where your data goes and how you'd get it back out. A demo on the vendor's sample data tells you what the tool can do on a good day; your own examples tell you what it will do in your business.

AI Getting Started Apr 11, 2026 • 18 min

How to Evaluate AI Tools Without Getting Sold

Every AI vendor demo looks impressive. The question is whether it solves your problem or theirs. Here's how to evaluate AI tools before you commit, when only 13% of CIOs say AI has delivered significant value (Gartner, 2026).

AI Getting Started Sep 22, 2026 • 14 min

An AI Audit for Your Business: What It Covers and Costs

An AI audit for a business is an inventory plus an evidence check on every AI system in use. IBM's 2026 report found 92 percent of organizations hit by an AI-related breach lacked proper AI access controls. The inventory is where the surprise is.

AI Getting Started Aug 19, 2026 • 8 min

Why AI Pilots Stall Before Production

AI pilots stall before production for boring reasons: no owner, no definition of done, no data path, no kill switch. CISA 1 May 2026: start low-risk. NIST AI RMF: Govern, Map, Measure, Manage.

Work With Us

Need help building this into your operations?

Kief Studio builds, protects, automates, and supports full-stack systems for businesses up to $50M ARR.

Newsletter

New writing, straight to your inbox.

Strategy, psychology, AI adoption, and the patterns that actually compound. No spam, easy to leave.

Subscribe