AI for small business in 2026: a practical playbook for teams under 20 people

Start with the bottleneck, not the tool

List the five tasks that consume the most hours in your business this month. For each, note whether the output is text, a decision or a physical action. AI helps enormously with the first, moderately with the second and not at all with the third.

Pick the single most expensive text-heavy task and automate that one. Teams that buy tools first and look for uses later almost always end up paying for shelfware.

Customer communication

Drafting replies is the fastest win. Feed the assistant your five best past responses as examples, your tone rules and your refund policy, and have it draft while a human sends.

Build a short knowledge base of your policies and product details in plain text. Every assistant answers dramatically better when the facts are supplied rather than remembered.

Keep a human in the loop on anything involving money, complaints or legal wording. The time saved on drafting is real, the risk of unattended sending is not worth it.

Marketing and content

Use AI for the first draft, the outline and the repurposing, and use a person for the point of view. Content that reads as though nobody had an opinion performs poorly with both readers and search engines.

Build a reusable brief template: audience, goal, tone, three must-include facts, banned phrases. Pasting that brief before each request removes most of the generic output problem.

Batch production weekly. Fifteen minutes of setup and one hour of editing regularly produces a month of social posts and two solid blog drafts.

Back office and admin

Invoice extraction, expense categorisation, meeting notes and CRM updates are all bounded, checkable tasks, which is exactly where automation is reliable.

Connect the pieces with an automation tool once a process repeats weekly. Below that frequency the maintenance cost is higher than the saving.

Reconcile automated output monthly against the source records for the first quarter. After that, spot checks are usually enough.

Data protection and policy

Check whether each tool trains on your inputs by default and turn that off. Business and team tiers usually exclude training, while free tiers often do not.

Write a one-page policy: which tools are approved, what data may never be pasted, who to ask when unsure, and the requirement to disclose AI-assisted work where a client expects human authorship.

Keep a short register of the tools in use and who owns each subscription. It takes an hour and prevents both duplicate spend and orphaned accounts.

Budget and expected return

A realistic starting budget for a ten-person team is one assistant seat each plus one automation subscription, which lands in the low hundreds per month.

Expect three to six hours saved per person per week within two months if the rollout targets real bottlenecks. If you cannot name the hours saved, the rollout is not working.

Review quarterly and cancel anything below thirty percent seat usage. Subscription drift is the largest hidden cost of AI adoption in small teams.

A 90-day rollout

Days 1 to 30: assistant seats for everyone, one shared prompt library, weekly thirty-minute sessions where people demo what worked.

Days 31 to 60: automate the top two repeating processes and write the usage policy. Days 61 to 90: measure hours saved, cut unused tools and add one specialist tool against a bottleneck you can now quantify.