An AI class is worth it for a small business owner when the gap is general know-how, and one-to-one help is the better use of time when the gap is one of your own processes. If you can name the job you want fixed, start there; if you cannot yet name it, take a short class first.
Both cost your hours as well as money, and the hours are usually the larger bill. A free class that takes twelve hours of an owner's week and never touches the business can cost more than a paid session that fixes a task in an afternoon. The sums below put both options on one scale.
Two different problems that look like one
People ask "should I take an AI class?" when they have either of two different problems. The first is literacy: you are not sure what an AI assistant can do, how to prompt it, what is risky to paste into it, or how to judge its answers. The second is process: you know roughly what AI does but cannot see how to apply it to the way your business quotes, schedules, invoices or replies to customers.
A class is built for the first problem. It teaches many people the same things at the same pace. One-to-one help is built for the second, because it starts from your documents, your tools and your bottlenecks. Using a class to fix a process problem is slow, and using one-to-one help to teach basic literacy is expensive. Most of the disappointment people report comes from that mismatch.
A two-minute self-test
- Can you name one task you repeat weekly that you would like to hand to AI? (Yes points to process.)
- Have you tried an AI assistant on at least three real tasks? (No points to literacy.)
- Do you know what not to paste into it, such as client records or contracts? (No points to literacy, and to a quick read of how to stop AI tools training on your business data.)
- Has something you tried already worked once but not repeatably? (Yes points to process.)
Mostly "literacy" answers: begin with a class or a self-study plan. Mostly "process" answers: go to one-to-one help or a structured audit, and keep a class for staff later. A mix is common, and the section after next gives a sequence for it.

The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowCosting each option in owner hours
Price a class or a session in three parts: the fee, the hours you spend inside it, and the hours you spend afterwards turning it into something you use. Convert the hours to money using what an hour of your time is worth to the business, which for most owners is more than their notional salary because the work is not easily delegated.
| Cost line | Free or low-cost class | Paid live class | One-to-one help |
|---|---|---|---|
| Fee | $0 to a modest subscription | Varies widely; get the figure in writing | Quoted per session or per project |
| Hours inside | 6 to 12, self-paced | 6 to 16, fixed times | 1 to 3 per session |
| Preparation | Little | Little | 1 to 2 hours gathering your documents |
| Hours to apply it to your business | Often 6 to 10, on your own | Often 4 to 8, on your own | 1 to 3, partly done in the session |
| What you leave with | General skill and notes | General skill, a peer group, a certificate maybe | A working method for your task, written down |
The hour ranges are planning assumptions, not measurements, so replace them with the course outline's stated hours and your own honest guess at the application time. The point is the shape: a class front-loads hours and leaves the application to you, while one-to-one help shifts the work into the session.
Main worked example: a six-person bookkeeping practice
Take a practice whose owner bills clients out at $90 an hour and values her own scarce time at that rate. Her weekly pain is chasing clients for missing receipts and writing the same reminder emails. She considers a ten-hour class, assumed to be free, and a two-hour one-to-one session at an assumed fee of $400.
| Line | Class path | One-to-one path |
|---|---|---|
| Fee (assumed) | $0 | $400 |
| Hours inside | 10 | 2 |
| Preparation | 0 | 1.5 |
| Hours to apply it to receipt chasing | 8 | 2 |
| Total owner hours | 18 | 5.5 |
| Owner time at $90 | $1,620 | $495 |
| Total cost | $1,620 | $895 |
On those assumptions the free class is the more expensive route for this specific job, because her hours dominate. The sum reverses if the owner values her free time at a lower rate, if she wants the class for her own interest, or if the class teaches skills for five different jobs and she will use all five. And the one-to-one route only wins if it really does fix the job: a session that ends with a general chat and no written method is not worth $895. The fee and the hours are assumptions for illustration; swap in your own.
The test of value is the payback. If the reminder routine saves her two hours a week at $90, that is $180 a week and the one-to-one path repays its $895 in about five weeks. If she saves only 30 minutes a week it takes 20 weeks, and a different fix may be better. Run the same payback for any option before you book it.
When a class wins
A class is not a poor choice; it is the right choice for several situations that come up often.
- You have not used an assistant at all. Ten hours of guided practice with prompts, checking outputs and spotting invented facts beats a month of nervous poking. For a free path, a 30-day self-study plan covers the same ground at your pace.
- Your team needs a common baseline. If four staff need the same rules on what to paste and how to check, one class run for all of them gets everyone to the same level. See the time and money AI staff training takes for the sums.
- You learn by watching others. Some people need the pacing of live sessions and the pressure of a group. That is a legitimate reason to choose one.
- You want a credential. If a client, tender or insurer asks for evidence that staff are trained, a class with a certificate gives you a document. Check what your own client requires rather than assuming.
- You are exploring before committing. The big AI vendors run their own learning sites, such as OpenAI Academy and Anthropic's Claude Academy, with courses and events aimed at work use. Claude Academy describes its courses as free; check OpenAI Academy's terms for any course or event you join, since they change.
How to judge a class before you enrol
Ask for the outline and check these five points. If an organiser cannot answer them, that is information.
- Date. When was it last updated? A class built on last year's tool screens teaches you buttons that have moved.
- Practice share. What proportion of the time is you typing prompts, versus watching slides? Aim for at least half hands-on.
- Business tasks. Does it use realistic tasks, such as customer replies, quotes and summaries, or only toy examples?
- Data handling. Does it teach what must not be pasted, and how to switch off training on your data?
- Checking output. Does it show how to verify facts and catch invented details, with real examples of errors?
When one-to-one help wins
One-to-one help wins when the problem is concrete and tied to your own documents. Five situations stand out.
- A repeated job with a clear start and end. Quote follow-ups, invoice chasing, rota messages, product descriptions. The session builds the prompt, the template and the check, and you leave with them.
- A tool you already pay for. If your accounting, CRM or booking system has AI features switched off, a session can work out whether any of them are worth turning on.
- Sensitive material. Client files that cannot go into a shared class exercise. A private session can work on structure and anonymised samples.
- A stalled attempt. You tried something, it worked once, and now results vary. A second pair of eyes finds the missing instruction or the messy input quickly.
- Time pressure. A fixed deadline, such as a seasonal rush, makes a class's fixed schedule a poor fit.
What to bring so the session is not wasted
Bring:
1. The one job, in a sentence: "Chase missing receipts from 40 clients each month."
2. Three real examples, anonymised (an email, a spreadsheet row, a reply).
3. How long it takes now, per run, and how often you run it.
4. What a good result looks like, and what counts as a mistake.
5. The tools you already use and who can approve changes.
A session that starts with those five items can end with a method you run the next morning. A session that starts with "tell me about AI" ends with a general chat. If you would rather work this out with someone, my AI implementation consultation covers finding the jobs where AI pays off and handing over something your team can run; you can start from the contact page.
The decision rule, with three owners
Put the self-test and the costing together into one rule: if you can name the job and have a working understanding of the tools, buy help on the job; if you cannot name the job, or you have not yet used an assistant for real work, learn first and cap the hours. Three illustrative owners show how the rule plays out.
A physiotherapy clinic owner with no experience
She has never used an assistant and worries about patient information. She takes a short class on basics and safe use, spends two weeks trying it on non-patient tasks such as rota emails and a newsletter, and only then books one-to-one help on appointment reminders. Class first, then a session, because the rule says to learn when you have not used the tool.
An IT support firm owner with a specific bottleneck
He uses assistants daily and wants ticket summaries and customer replies drafted consistently. A class would teach him what he knows. He books a session on his ticket workflow, brings five real tickets with client details removed, and leaves with a template and a checking step. One-to-one only.
A dental practice manager training a team of eight
The need is a shared baseline and shared rules about patient data. She picks one group class for everyone, then asks for one session on a single admin job, such as recall messages. Class for literacy, session for the process, in that order. For more on staff time, read what your staff need to know about AI.
What goes wrong with each route
Classes
- Tool tourism. You see six tools in six weeks and use none. Fix: pick one assistant and use it daily during the class.
- Toy examples. Writing poems and recipes teaches little about invoices. Fix: bring your own tasks and try them in the practice slots.
- No follow-through. Notes pile up and nothing changes. Fix: before the class ends, schedule the first real task for the next morning.
- Stale content. A recorded class with old screens. Fix: check the date, and prefer material that names its tool versions.
One-to-one help
- Vague scope. "Help me with AI" produces a vague session. Fix: bring the five items above.
- Dependency. You cannot run it without the adviser. Fix: ask for the method written down and run the first repeat yourself in the session.
- Over-promising. Be wary of fixed saving claims before anyone has seen your process. Ask how success will be measured and who measures it.
- Wrong person. Someone who knows tools but not your type of business. Fix: ask for an example of a similar job they handled, described generically.
A sequence that suits most owners
If the rule leaves you in the middle, use a staged plan that spends little until something has worked.
- Week 1: time the job you want to improve and write down the steps. Four hours at most.
- Weeks 1 and 2: do a short class or the self-study plan, using that job for every exercise. Eight hours at most.
- Week 3: time the job again with the new method. If it is at least 30 percent faster with no loss of quality, keep going on your own.
- Week 4: if it is not, book one-to-one help with the timed evidence in hand, and ask for a method that targets the slowest step.
- Month 2: train staff on the method that worked, not on the whole of AI. See course, 1:1 call, audit or project: which AI help you need if the next step is not obvious.
That plan caps the downside: you spend about twelve hours before you know whether a class was enough, and you reach one-to-one help with evidence instead of a vague wish. The thirty percent threshold is a suggestion, not a research finding; choose a number that makes the change worth the effort for your business.
A timing log you can fill in this week
Whichever route you choose, the evidence of whether it worked is a before-and-after timing. Keep it to one task and three columns. This filled-in version is illustrative, for a bookkeeping practice chasing missing receipts.
| Step | Before (minutes) | After (minutes) | Note |
|---|---|---|---|
| List clients with missing receipts | 25 | 10 | Spreadsheet filter, no AI needed |
| Write 40 reminder emails | 90 | 20 | Template with [first name] and the missing items |
| Check drafts for wrong amounts | 0 | 15 | New step; this is the cost of using AI |
| Log replies and follow up | 35 | 35 | Unchanged, the next thing to look at |
| Total | 150 | 80 | Saves 70 minutes a month |
Notice two things. A new checking step appears, which is the honest cost of AI output. And the saving is 70 minutes a month, about $105 at $90 an hour, which would not justify the one-to-one fee in the worked example above unless the same method also covers other jobs. A timing log stops you buying help for a job that was never big enough, which is a valid result: sometimes the right decision is to do nothing and look for a larger job.
Example: a before and after of the reminder email
Before (typed from scratch each time):
"Hi, just checking if you have those receipts for the last quarter, thanks."
After (template filled by the assistant from the client list):
"Hi [first name], for the quarter ending 30 September we are still missing
three receipts: the 14 August fuel receipt, the 2 September equipment
purchase and the supplier invoice from 20 September. Could you reply with
photos by Friday? [your name]"
The after version names the items, so fewer clients reply asking which receipts. That detail, not the AI itself, is where most of the time saving comes from, and it is the kind of thing a session on your own process finds and a general class rarely does.
Example: when the log says stop
Say an owner of a small physiotherapy clinic logs her appointment-reminder job and finds it takes 40 minutes a week. Even a perfect fix saves under three hours a month, worth perhaps $120. She decides not to buy a class or a session for that job, and instead uses the timing habit on her bigger one: writing up exercise plans, which takes six hours a week. The log redirected her money to the task where it could actually pay back, and cost her an hour of logging.
Questions to ask any provider, class or personal
- "Show me the outline, and tell me when it was last updated."
- "What will I be able to do on Monday that I cannot do today?"
- "How many hours should I expect to spend outside the sessions?"
- "What do you advise me not to paste into these tools?"
- "What happens if it does not work for my business? Is anything refundable?"
- "Do you recommend a free route for the basics, and where does paid help start to matter?"
A provider who answers the last two honestly, including by sending you somewhere free, is usually the one worth paying for the rest.
Questions owners ask before choosing an AI class
Should I start with a free class?
Usually yes, if you have never used an AI assistant for real work. A free or low-cost class costs you hours, not money, so cap it at about eight hours and stop when it stops teaching you something you can use that week. If the examples never touch your own business, stop early and book help on your own process.
What should a good AI class include?
Live practice on realistic business tasks, guidance on what not to paste into a tool, and a way to check output for mistakes. Be wary of classes that are mostly tool tours, since tools change monthly. Ask to see the outline, and check that it was updated recently and states which tools it uses.
Is one-to-one help only for bigger firms?
No. A small firm with one repeated job, such as quoting, invoicing queries or customer replies, gets the most from one-to-one help, because the session can work on that job directly. The question is whether your problem is specific, not how many people you employ.
How do I know it worked?
Pick one task before you start and time it. After the class or session, do the same task again with the new method and time it. If it is not faster or better within two weeks, you have bought information, not a change. Write down which step still takes the time.
Further reads
- Is a 1:1 AI Consultation Worth It for a Small Business? — The one-to-one side of this decision in more detail.
- How to Build a Staff Training Matrix With AI — Track who has learned what once a team joins in.
- How to Train Front-of-House Staff to Work Alongside AI — Training customer-facing staff to work beside AI.
- How to Classify Business Data Before Using AI Tools — Sort what can and cannot go into AI tools.
- How to Add AI Steps to Zapier: Classify, Summarise, and Draft — A practical first automation once you have a method.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI Academy and Claude Academy home pages (checked November 2026). Class prices vary widely and are not quoted here.