AI transformation for a small sales and marketing team: where to start

“AI transformation” sounds like a company-wide programme. For a small team it is smaller and more useful: make one repeated job faster, then do it again.

Start with one job, in five steps

1. InputPick one jobRepeated, boring, and easy to describe.
2. RuleWrite the rulesWhat happens today, step by step.
3. AI stepAdd one AI stepOnly where the work is fuzzy: reading, sorting, drafting.
4. PersonKeep a deciderA person approves what matters.
5. OutputMeasure itTime saved or errors avoided, on one page.

1. Pick one job. Look for work that repeats every week and that people dislike: sorting inbound leads, writing campaign reports, chasing follow-ups. If you cannot describe it in a few lines, pick another.
2. Write the rules down. Much of the job is often rules already. Write them as if for a new hire. This list is the automation, and it shows what is left over.
3. Add one AI step for the fuzzy part. Use AI for reading, summarising, sorting or drafting, where a rule cannot capture it. Keep the rest as plain, checkable rules.
4. Keep a person on the decision. Approvals, prices, and anything a customer reads as a promise stay with someone who owns it.
5. Measure one thing. Time per task, or mistakes per week. Note the number before you start.

What to leave for later

• A company-wide AI policy. Write two lines about what data (customer contact details, prices, contracts) must never be pasted into an AI tool, and move on.
• Buying a tool suite. Try the smallest tool that does the one job.
• Full autonomy. Start with drafts and suggestions; widen only where the results have been checked.

Signs it is working

The task takes less time, nobody had to learn a new system to use it, and a person can check any result against the written rules. If not, shrink the job.

Where this shows up

Hoàng Huy Đức describes the work as the automation layer under marketing and sales: lead scoring and routing, CRM pipelines, telesales workflows, campaign reporting and internal agents built with Claude Code (LinkedIn, September 2026). See the services for how a first project is scoped.