Skip to main content
••13 min read•By Nathan Ibgui, fractional CIO

Claude Training for Business: Where to Start in 2026

Real plans and prices, training on your content, a four-week training plan: the complete Claude getting-started guide for a small business.

Computer training room: about thirty participants at laptops, facing a presentation screen.
Training a team on Claude is less about demoing a tool than about a way of working. Photo: Raysonho @ Open Grid Scheduler / Grid Engine — CC0, via Wikimedia Commons.

The 30-second answer

Claude isn’t one product, it’s five. And the plan you choose decides as much about what your teams can do as about what happens to your data.

The most important difference doesn’t show up in day-to-day use: on individual plans, model training on your content is on unless you opt out; on team plans, it is off by default. That is the first decision to make, before the very first training session.

$20

Per seat per month on an annual Team plan — the first tier where your content is not used for training by default.

2 to 150

The team format covers this range of users: nearly every small and mid-sized business.

22%

Of French very small and small businesses use generative AI — a share that doubled in one year (France Num 2025 barometer, France).

Prices and plan scopes checked on Anthropic’s public US pricing page as of November 2026. Plans change fast: re-check before you sign.

First confusion to clear up: “Claude” means five things

Most teams discover Claude through the chat window and stop there. That’s the front door, not the product. The same subscription includes tools aimed at different jobs, and each one is learned differently.

One Claude subscription Pro, Max, Team or Enterprise Chat Write, analyze, summarize, decide Everyone Claude Code Reads code, edits files, runs commands Technical team Claude Design Mockups, decks, marketing materials Product, marketing Cowork · Science Collaborative work and research uses Depending on the role Training people “on Claude” without separating these building blocks means training everyone on the same thing — and missing the actual need.

The practical consequence is simple. A single training course for the whole company doesn’t work. The common foundation — how to phrase a request, what to check, what never to paste into a conversation — can be shared. Beyond that, a developer and a communications officer have neither the same tool nor the same habits to build.

The plans, and what they actually change

Criterion
Pro
Team
Enterprise
Price
$17/month billed annually
$20/seat/month billed annually
$20/seat + usage
For whom
One individual
2 to 150 people
Large organizations
Training on your content
You must opt out yourself
Off by default
Off by default
Administration
None
SSO, central billing
SCIM, roles, audit logs
Claude Code and Design
Included
Included (Premium seats: 5x usage)
Included, enabled by the admin

Two clarifications that prevent a budget surprise.

Team seats can be mixed. The standard seat is $20 a month billed annually ($25 month to month) and already includes Claude Code, Cowork, Design and Science; the Premium seat, at $100 billed annually ($125 month to month), unlocks five times the usage. You don’t have to move the whole company to the expensive seat: put Premium seats where usage is heavy, standard seats everywhere else.

The Enterprise plan is not a flat fee. It is billed at $20 per seat per month, billed annually with a 20-seat minimum, plus consumption at API rates. The bill therefore varies with the models used and how heavy the tasks are — it’s a budget line to monitor, not a fixed cost. We walk through the costing method in the real cost of AI agents in business.

Which model for which task

Plans give access to several models, and that confuses people in training. The useful rule fits in three lines: the default model suits nearly all office work, you drop down a tier only for simple high-volume work, and you move up only for hard reasoning.

Situation
What to choose
What you gain
Writing, summarizing, analyzing a document
The default model, no second thought
Time: that’s 90% of real-world usage
Processing 300 rows with the same simple rule
A faster, cheaper model
Speed and usage budget
Complex trade-offs, contract review, difficult code
The most capable model available
Fewer errors where errors are expensive
A 400-page document to process in one go
First check the context window of your plan and model
Avoiding silent truncation

One technical point is worth knowing for whoever leads the rollout, because many contradictory figures circulate: as of November 2026, the context window Anthropic lists for chat on paid plans depends on the model and ranges from 200,000 to 1 million tokens. The size follows the model you pick, not just the plan, and older models stay at the low end. In practice, a very large corpus may still need to be split: better to know that before an employee concludes that “the tool forgot the beginning.”

In training, don’t spend more than five minutes on this. Model selection is a secondary setting; the quality of the context you provide matters far more than the choice of model, and that is where the training hour should go.

The uses that actually stick, by function

Adoption can’t be decreed: it is observed, function by function. Here is what still holds after three months in small businesses, as opposed to demos that impress once and are then abandoned.

Leadership

Preparing a board meeting from scattered documents, rewording a memo for a different audience, laying out the pros and cons of a decision before making it. Low volume but high perceived value — this is often what unlocks the budget.

Sales

Turning meeting notes into a structured proposal, anticipating a client’s objections based on their industry, reviewing a bid response. Caution: prices and commitments are never invented, they are pasted in.

Administration and finance

Comparing contract versions, extracting deadlines from a batch of documents, drafting a payment reminder template. The gain comes from reading large volumes quickly, not from producing figures — which still need checking line by line.

Marketing and communications

Adapting a message to several formats, producing variants of the same pitch, preparing an editorial calendar. This is the function where adoption is fastest, and where the risk of a bland, generic tone is highest.

Human resources

Writing job postings, preparing interview guides, summarizing internal survey feedback. Sensitive area: no personally identifiable information or health data should be submitted, and screening job applications is subject to specific rules.

Technical

This is where dedicated tooling changes the game, with Claude Code. The topic is different enough to deserve its own treatment: see getting a technical team started.

A four-week training plan

The format that works best in small businesses isn’t a full training day, but four short sessions a week apart, with real work in between. The skill is built on the company’s own cases, not on exercises.

1

Week 1 — the foundation and the ground rules

One hour, everyone together. How to phrase a useful request, how to check an answer, and above all the policy: what must never be pasted into a conversation. Leave this session with the policy written down, not just promised.

2

Week 2 — everyone brings a real case

Each participant arrives with a task they actually do and find tedious. We work on it in the session. This is the moment that decides adoption: someone who leaves with a measurable gain on their own work comes back; the others don’t.

3

Week 3 — job-specific tools

Separate sessions. Claude Code for the technical team, Claude Design for product and marketing, projects and connectors for support functions. This is where training stops being generic.

4

Week 4 — what we keep

Sort it out: which uses have become habits, which were dropped and why. The uses you keep are documented for future hires. Without this session, everything fades within six weeks.

This progression is deliberately close to the one described in our AI training program for teams, which covers the whole setup — budget, metrics and, in France, training funding schemes. This article focuses on one specific tool.

Proving it’s actually useful

It’s the question every leadership team asks after a quarter, and one most rollouts can’t answer — because nobody set a baseline before starting.

Don’t measure the number of logins. It’s the metric dashboards report, and it tells you nothing: one employee can log in every day to reword emails without the company gaining anything, while another uses it twice a month on high-stakes files.

Turnaround time on a named task

Pick two or three recurring deliverables — a bid response, a monthly report — and record how long they took before. It’s the only number leadership will accept.

Tasks that have disappeared

List what nobody does by hand anymore. It’s a better adoption signal than usage frequency, because it’s irreversible.

The 90-day drop-off rate

How many assigned seats are no longer used. If it exceeds a third, the problem isn’t the tool but the training — or the choice of the first use cases.

The baseline to take before the rollout is the one everyone forgets: time two representative deliverables the week before access opens. Without that initial measurement, any later assessment will boil down to impressions — and on this subject, impressions are consistently too favorable in the first weeks and too harsh afterwards.

The full costing method, including the return-on-investment calculation, is detailed in our article on priority use cases and in the business case for AI agents in small businesses.

The four startup mistakes

Rolling out without a usage policy. It’s the only truly irreversible mistake, because it involves data that has already left the building. Before the first access is handed out, one page is enough: which categories of information are banned from conversations — customers’ personal data, confidential contract terms, credentials — and who to escalate to when in doubt.

Confusing the individual plan with the team plan. A small business where five employees took out personal subscriptions reimbursed as expenses hasn’t rolled out Claude: it has five accounts outside any control, on a plan where training is turned off user by user, with no administration or visibility.

Training on the tool rather than on judgment. The skill that matters isn’t knowing how to write a request: it’s recognizing an answer that looks right and isn’t. On figures, regulatory or legal topics, verification remains essential — and that is precisely what needs training.

Starting with the most spectacular use case. Projects that last start from a boring, frequent, low-risk task. We detailed this trade-off in AI in small business: where to start and in the use cases to prioritize.

That leaves the question that will drive your choice of plan as much as your policy: what actually happens to the content your teams submit. It deserves its own treatment, and we gave it one in Claude for Business: What Happens to Your Data.

Nathan Ibgui, IT & AI consultant

Written by

Nathan Ibgui

Independent consultant and fractional CIO, I help small and mid-sized businesses with IT strategy, artificial intelligence, automation and cybersecurity — from assessment to execution.

Ready to transform your business?

Describe your needs in a few lines: I reply within 24 hours with an initial analysis and concrete recommendations.