The complete guide to AI for small businesses

A cornerstone guide covering where AI belongs in a small business, how to introduce it, what to measure, and what to refuse.

The AI Business in a Box Editorial TeamUpdated 29 Jul 202626 min read

Chapter 1 — What AI is genuinely good at

AI is strongest wherever your business produces language repeatedly: replies, quotes, descriptions, summaries, procedures and follow-ups. It is weakest wherever a wrong answer is expensive and unverifiable — pricing, legal commitments, stock levels and promises about dates.

  • Drafting: first replies, product copy, proposals, SOPs.
  • Transforming: shortening, translating, changing tone, extracting action items.
  • Structuring: turning messy notes into a procedure a new hire can follow.
The rule that prevents most failures

AI drafts, a human decides. Any workflow where nobody reviews customer-facing output within your first month will eventually embarrass you.

Chapter 2 — Find the one bottleneck

Adoption fails when it starts broad. Pick the single place where your business loses money to delay. For most SMEs selling in conversation, it is the gap between an enquiry arriving and a useful reply going out.

  1. 1
    Count

    How many enquiries arrive weekly, and how many get a reply within an hour?

  2. 2
    Cost it

    Multiply missed enquiries by your average order value. That number funds the project.

  3. 3
    Fix one path

    Standardise the answers for that one path before touching anything else.

Chapter 3 — Build your business context file

Your single most valuable AI asset is a plain document describing your products, prices, delivery terms, tone and forbidden phrases. Every assistant becomes competent the moment it reads this file, and useless without it.

  • Products and the outcome each one delivers
  • Price list and what is never discounted
  • Tone rules: how you greet, how you apologise, what you never say
  • Escalation: which questions must reach a human

Chapter 4 — Three workflows worth building first

  1. 1
    First-reply drafting

    Assistant proposes a reply from your script library; the salesperson edits and sends.

  2. 2
    Follow-up scheduling

    A simple list, reviewed daily, with a written message for day 1, 3 and 7.

  3. 3
    Post-sale check-in

    A short outcome question on day 3 that surfaces problems while they are cheap.

Chapter 5 — Measure four numbers, ignore the rest

  • Time to first reply
  • Share of enquiries followed up within 48 hours
  • Conversion rate from enquiry to order
  • Hours returned to the team weekly

If none of these move within 30 days, the problem is the workflow, not the model. Change the process before changing the tool.

Chapter 6 — Risk, privacy and what to refuse

Do not paste customer identity documents, payment details or anything you would not email in plain text. Keep a written list of what your team may and may not share with an assistant, and review it quarterly.

The businesses that win with AI are not the ones that adopt fastest. They are the ones whose standards were already written down.

cornerstoneAISME
Written by
The AI Business in a Box Editorial Team
Research & editorial · Nairobi, Kenya

We research, test and document AI workflows with small businesses across Kenya and the wider continent. Everything we publish is written to be used the same day you read it — scripts, checklists and systems, not theory.

More from this desk →