In a typical week, a team copies data, summarizes conversations, prepares reports and adapts content. Some of that work can be automated. To choose a useful starting point, look at which tasks repeat, how long they take and how easily you can check the result.
Start with a recurring task, clear inputs and an output the team can check.
1. Describe the process before choosing a tool.
Write down what triggers the task, what information goes in, what should come out and who uses it. Include exceptions: incomplete documents, different languages, conflicting data or requests that need a colleague. A process that is hard to explain may need organizing first.
If the steps follow fixed rules, a formula or a simple integration may be sufficient. Generative AI can help with variable text: summaries, preliminary classification or content variations. Report figures deserve reproducible calculations; an interpretation can then be prepared as a draft.
2. Choose an output you can check properly.
A summary can sound convincing while changing the meaning of a conversation. NIST identifies confidently presented false content as a generative AI risk and recommends checking sources in outputs. For a marketing team, that means keeping the original material available and naming the person who validates the draft.
Begin with an internal step: a summary for a colleague or copy options for an editor. Agree on who approves facts, offers and publication. Use the information the process actually needs, and check where it will be processed, who can access it and which permissions the tool requires.
3. Measure the work left after automation.
Test representative examples, including incomplete and unusual cases. Compare outputs with the current process: accuracy, omissions, review time and rework. A successful demonstration is a reason to keep testing; a broader set of situations helps estimate everyday usefulness.
Count the time saved, then subtract checking, corrections and maintenance. Include tool and implementation costs alongside it. For example, 100 tasks taking 5 minutes each total 500 minutes. If review takes 200 minutes and maintenance another 60, the saving is 240 minutes for that period, before recovering the initial setup effort. Agree on who steps in when the workflow stops.
4. Follow a complete workflow, including exceptions.
For an inquiry handover, we would keep the original request beside the AI-proposed summary. The need, timing and missing information would become easy-to-check fields. Separate rules check required information and duplicates. A person approves the next step; the system does not guess budget or availability.
The interactive How we would example compares a clear, incomplete and repeated request. It uses prepared fictional drafts without calling a model or CRM. A real pilot would compare quality and total handling time using authorized examples. We would assess the existing integration and, where needed, candidates such as n8n, Make or Zapier together with the responsibility for maintaining them.
Possible starting points for a marketing team.
Bounded tasks and the checks we would put alongside them.
| Task | Possible AI role | Human check |
|---|---|---|
| Preparing a sales conversation | Summarize the available notes | Facts, sources and missing information |
| Adapting approved content | Draft variations for other formats | Meaning, tone, claims and format requirements |
| Preparing the monthly report | Draft explanations from calculated data | Figures, context and conclusions |
These are possible tests. Time savings and quality need to be measured within the team’s actual process.
Before expanding the automation
The process, inputs and exceptions are described and have an owner.
Representative cases have been tested against an acceptable quality standard.
Review, approval, data access and a return to manual work are defined.
The time saved remains useful after corrections, maintenance and costs.
A good workflow earns its place in the team through usefulness. At NOT MORE., we can start with one task, build a bounded test and decide whether to expand it after seeing it at work.