Start with one repeated task, a clear input and an output someone can check. Useful AI workflows can turn an enquiry into a brief, extract fields from a document, prepare a quotation draft, organise meeting notes or explain a small analytics report. Each needs a defined handoff: who checks the result and what happens next?
A prompt is one part of the process. The source material, review and destination matter just as much.
Reviewed on 13 November 2026. All names, records, figures and examples below are fictional teaching material. These workflows have not been presented as completed client implementations or measured time savings.
What makes an AI workflow useful?
Make's workplace automation examples include handling requests, preparing briefs and extracting meeting actions. Those are useful categories to consider; their value in your business still needs to be measured. Source: Make, AI-powered workplace
You can try each workflow manually before connecting tools.
| Workflow | Intended output | Who reviews it? |
|---|---|---|
| Customer enquiry | A brief with missing information | Person handling new business |
| Document extraction | A table linked to source evidence | Person using the records |
| Quotation preparation | A scope and price draft | Person authorised to quote |
| Meeting follow-up | Tasks separated from ideas | Meeting owner and assignees |
| Analytics review | Observations and one next check | Person responsible for the website |
1. Turn a customer enquiry into a usable brief
Input: the enquiry and an approved description of the services you offer. Use only the information needed for the task.
Fictional enquiry:
We need a bilingual service website with a booking request form and editable service descriptions. We want to launch before an event, but our existing copy needs review.
Prompt:
Create a project brief from the supplied enquiry. Separate explicit requirements, missing information and questions for the customer. Do not invent a budget, deadline, language pair or technical requirement. Make missing or uncertain information explicit. Return a draft for review.
Expected output: the brief records a bilingual website, a booking request form, editable service descriptions and copy review. It asks for the languages, event date, content owner and expected booking process. The budget remains “not provided”.
That last detail matters: a booking request form does not automatically mean live availability, payment collection or a confirmed reservation. The brief should preserve what was actually requested.
Check and handoff: the person handling the enquiry compares the brief with the original message, removes assumptions and selects the questions worth asking first. The reviewed brief can then support a customer reply or quotation discussion.
Keep the original enquiry attached for the reviewer.
2. Extract document fields without losing their evidence
Input: a document you are authorised to process, plus the required fields. Begin with a small, readable example.
Fictional delivery note:
Document DN-014. Delivery: 18 November. Quantity: 3 boxes. Total: not provided.
Prompt:
Extract document ID, delivery date, quantity, unit and total. Return the value as written, a short supporting excerpt and any missing information. Preserve an incomplete date. Use “not provided” when a value is absent. Do not calculate or infer a missing total.
Expected output:
| Field | Extracted value | Evidence or issue |
|---|---|---|
| Document ID | DN-014 | “Document DN-014” |
| Delivery date | 18 November | Year missing |
| Quantity | 3 | “3 boxes” |
| Unit | boxes | “3 boxes” |
| Total | not provided | No amount supplied |
Google Document AI's document schema supports entities with text and page references. That provides a useful design principle: retain a route back to the source instead of storing only the extracted value. Source: Google Cloud, Document resource
Check and handoff: verify identifiers, quantities and dates against the document. Resolve the missing year before using the date in a time-sensitive process. Keep extraction separate from any later calculation or system update.
3. Prepare a quotation draft for human review
Input: a reviewed brief, approved service descriptions, an applicable rate card and any agreed commercial terms. If a necessary input is missing, the draft should show the gap.
Prompt:
Prepare a quotation draft using only the supplied brief, service information and rate card. Separate scope, deliverables, exclusions, assumptions, prices and open questions. Do not invent rates, discounts or delivery dates. Mark unsupported amounts “price required”. Label the whole document “Draft — requires review”.
Expected output: for the fictional website enquiry, the draft might list the booking request form and any pages explicitly agreed in the reviewed brief. An unspecified page list remains an open question, along with the language pair and launch date. If payment processing has not been requested, it belongs in a question or an explicit scope discussion, not silently inside the price.
If no rate card was supplied, there should be no convincing-looking invented total. The useful output is still a structured scope that makes pricing easier to complete.
Check and handoff: the authorised person checks scope, workload, rates, arithmetic and terms. Recalculate totals in the business's normal calculation tool. That person approves the final quotation and decides how it is sent.
The draft should make review clearer. Fluent text that hides unanswered questions is a poor result.
4. Turn meeting notes into tasks people can accept
Input: meeting notes you are authorised to use. Typed notes are enough; the workflow does not depend on a recording or a particular meeting platform.
Fictional notes:
09:10 — Mari: I will review the language copy by 20 November 2026.
09:15 — Alex: Maybe we add online payments later.
Prompt:
Separate explicit commitments, decisions, suggestions and unresolved questions. For each task, include the stated owner, deadline and supporting note. Leave missing owners or dates unassigned. Do not turn an idea into a commitment. Return a task draft for the meeting owner.
Expected output: Mari has a proposed task to review the language copy by the stated date. Alex's payment idea remains a suggestion. It does not become an assigned implementation task or a promised feature.
Check and handoff: the meeting owner reviews the draft, confirms unclear assignments and adds accepted tasks to the team's usual system. Corrections should update the task record as well as the summary.
If you use Google Meet's built-in note feature, check its current plan, language and access requirements. The workflow here also works from manually written Estonian notes; it does not assume built-in Estonian support. Source: Google Meet, Take notes for me
5. Turn an analytics report into one sensible next action
Input: a small export with the source, date ranges, filters and metric definitions included. Compare equivalent periods before asking for an explanation.
This fictional Search Console example compares two equal-length periods using the same filters:
| Metric | Earlier period | Later period |
|---|---|---|
| Impressions | 1,000 | 1,200 |
| Clicks | 20 | 18 |
| Click-through rate | 2.0% | 1.5% |
In Search Console, click-through rate is clicks divided by impressions. The report also supports breakdowns such as query, page and device. Source: Google Search Console, Performance report
Prompt:
Use only this table and its definitions. Give three factual observations, explain what the table cannot establish and suggest one next check. Recalculate percentages. Distinguish percentage points from percentage change. Do not invent a cause or claim revenue changed.
Expected output: impressions increased 20%, clicks fell 10%, and click-through rate fell 0.5 percentage points. The table alone does not explain why.
Check and handoff: verify the arithmetic and filters. The next action could be to inspect which pages, queries or devices contributed to the change before rewriting a title. The website owner chooses the investigation and records what they learn.
The purpose is a clearer decision.
Test the whole workflow before expanding it
Choose one of these examples and write down what an acceptable result looks like. Test ordinary inputs, missing information, contradictory notes and an unavailable source. Use invented records first when the real material requires access decisions.
Measure the work that remains: preparing the input, checking the output, correcting mistakes and handing it over. Compare that total with the existing process. Keep a record of accepted outputs and recurring corrections, not just the speed of the first draft.
Once the task and checks are stable, decide which steps are worth connecting. A fixed workflow may be sufficient. The AI agent, chatbot and automation comparison explains when varying tool selection is useful.
The accompanying workflow design sheet covers inputs, outputs, permissions, review, exceptions and measurements for your implementation discussion.
Use the workflow design sheet to adapt one example to your own input, review process and destination.
Frequently asked questions
What is the easiest way to start using AI at work?
Choose a repeated, bounded task with an output you already know how to assess. Enquiry summaries or task drafts are useful candidates. Start with a small example and keep a person responsible for accepting the result.
Do these workflows require an AI agent?
No. They can begin as manual prompts or predefined steps. Consider an agent when the system needs to choose among different information sources or tools, and that flexibility solves an identified problem.
How do I know whether AI saves time?
Measure the entire process, including input preparation, review and rework. A fast draft can still create extra work later. Compare similar tasks and record both time and the quality of the accepted result.
Can I use these prompts with confidential documents?
First check whether your organisation permits the document and tool combination. Use the minimum relevant information, appropriate access and the approved review process. A useful prompt does not determine whether a particular upload is authorised.
Sources and further reading
- Make: AI-powered workplace — workplace workflow categories.
- Google Cloud: Document resource — document entities and source references.
- Google Meet: Take notes for me — meeting-note feature availability and access.
- Google Search Console: Performance report — search metrics and dimensions.
Documentation was checked on 13 November 2026. Prompts, examples and recommended review steps are original illustrative material.
Related reading: How much does an AI chatbot for a small-business website cost?
Putting AI to work on your website?
Choosing a model is one part of a working feature. If your business needs a support assistant, content workflow or connection to an existing system, I can help define and build the interface, integrations and checks it needs.