Chapter 9 of 17
AI in Office Productivity: Communication and Documentation
Applies generative AI to everyday office documents: emails, reports, minutes of meetings and proposals. It shows where AI saves time and where human judgement must lead, using the UK government Copilot trial and the Harvard–BCG study.
In this chapter
9.1 Why office documents suit generative AI
Most office writing follows known structures: an email has a greeting, purpose, details and close; a report has a summary, findings and recommendations. Generative AI is good at producing structured, fluent text from rough notes. The person still supplies the facts, the decisions and the final judgement.
A general workflow for any office document
- Collect the inputs: notes, data, previous documents, the recipient and the purpose.
- Prompt with role, task, context, constraints and format (Chapter 5).
- Review the draft for accuracy, tone and completeness.
- Edit and personalise.
- Check for confidential information before sending.
9.2 Email writing
| Email type | Key elements to put in the prompt | Example prompt |
|---|---|---|
| Formal request | Recipient, what is needed, by when, why | "Write a formal email to the Accounts Head requesting approval of ₹45,000 for lab equipment by 15 November. Mention it is for the new IoT course." |
| Follow-up | Earlier communication, what is pending, gentle urgency | "Write a polite follow-up to a vendor who has not sent the quotation requested on 10 October." |
| Apology | What went wrong, responsibility, fix, prevention | "Write an apology to a client for a two-day delay in report delivery due to a data error; the corrected report is attached." |
| Declining | Appreciation, clear no, brief reason, alternative | "Decline a vendor's proposal politely, saying we chose another supplier on price, and invite them for future tenders." |
| Reply to a long thread | The thread pasted, your position | "Summarise this thread in three lines and draft my reply agreeing to option B." |
Illustrative example
Input notes: "tell team – review mtg moved thu 3pm conf room 2 – bring Q2 numbers – Ravi presenting"
Prompt: "Turn these notes into a short, clear email to my team. Friendly professional tone. Subject line included."
A good output has a subject such as "Review meeting moved to Thursday, 3 pm"; one opening line; the new time and place; what to bring; who presents; and a closing line. The writer checks that the day and date are correct before sending.
Built-in tools. Gmail's "Help me write" (Gemini) and Outlook's Copilot can draft, shorten, lengthen or change the tone of an email inside the mail client.
9.3 Report drafting
A standard business report has: title, executive summary, background, method or data, findings, analysis, recommendations and appendices.
How AI helps
- Building an outline from the purpose and audience.
- Turning bullet-point findings into paragraphs.
- Writing the executive summary after the body is complete.
- Interpreting tables: "Describe the three main trends in this sales table."
- Checking consistency: "Do any numbers in the summary differ from those in the tables?"
Where humans must lead: choosing what data to use, checking every number, deciding recommendations, and taking responsibility for the conclusions.
Example prompt for an executive summary
You are a management consultant. Write a 150-word executive summary for the board.
Use only the findings below. Lead with the main conclusion. End with two recommendations.
Findings: """
- Q2 revenue ₹4.2 crore vs target ₹4.8 crore (−12.5%)
- Online channel grew 18%; retail fell 22%
- Two largest retail outlets lost footfall after metro construction
"""
9.4 Minutes of Meeting (MoM)
Minutes record what was decided, who will do what, and by when. A standard MoM has: meeting title, date, time, venue, attendees and absentees, agenda items, discussion summary, decisions, action items (task, owner, deadline) and the next meeting date.
AI-assisted process
- Record the meeting (with consent) or take rough notes.
- Use a transcription tool. Microsoft Teams, Google Meet and Zoom offer built-in transcripts and AI recaps on some plans.
- Prompt: "From this transcript, prepare minutes in the format below. List decisions separately. Put action items in a table with Task, Owner, Deadline. Do not include side conversations."
- The meeting secretary checks names, decisions and deadlines, then circulates the minutes.
Action items table produced from a transcript (example)
| Task | Owner | Deadline |
|---|---|---|
| Finalise vendor shortlist for lab equipment | Dr. Mehta | 30 Oct |
| Circulate revised timetable | Ms. Gupta | 27 Oct |
| Book auditorium for orientation | Admin office | 2 Nov |
Cautions: transcripts mishear names and numbers; AI may present a suggestion as a decision. Always confirm with the chair before circulating.
9.5 Proposal writing
A business or project proposal usually contains: problem statement, objectives, proposed solution, scope, methodology, timeline, budget, team, expected outcomes and risks.
How AI helps
- Turning a rough idea into a structured outline.
- Drafting the problem statement and objectives in clear language.
- Suggesting risks and mitigations to consider.
- Adapting one proposal for different funders or clients.
- Checking the draft against a funder's guidelines: "Compare this proposal against the call guidelines below and list any missing sections."
Example prompt
You are an experienced grant writer. Draft the "Objectives" and "Expected Outcomes" sections
of a proposal to set up an AI skills lab in our engineering college for 240 students a year.
Objectives must be SMART. Outcomes must be measurable. Maximum 250 words.
Caution: budgets, past results, staff names and institutional commitments must come from the organisation, never from the AI.
9.6 Tools for office productivity
| Tool | Where it works | Typical office uses |
|---|---|---|
| Microsoft 365 Copilot | Word, Excel, PowerPoint, Outlook, Teams | Drafting from files, meeting recaps, email summaries, slides from a document |
| Gemini in Google Workspace | Gmail, Docs, Sheets, Slides, Meet | "Help me write", summarising Drive files, meeting notes |
| ChatGPT, Claude, Gemini app | Web and apps | General drafting, analysing uploaded files |
| Grammarly and similar | Browser and apps | Grammar, tone and clarity checks |
9.7 Worked numericals
Numerical 9.1: Annual time saved from a reported average.
The UK government's trial found that users of Microsoft 365 Copilot saved 26 minutes a day on average. With 220 working days a year and a 7.5-hour working day:
This matches the report's description of roughly two weeks a year.
Numerical 9.2: Is a paid AI licence worth it? (illustrative figures)
An executive earns ₹60,000 a month and works 22 days × 8 hours = 176 hours. A licence costs about ₹2,550 a month (US$30 at ₹85/US$). The executive saves 26 minutes a day.
The licence just pays for itself at this salary. For a senior manager at ₹2,00,000 a month, the same time saved is worth about ₹10,800, a much stronger case. This assumes the saved time is used productively.
Numerical 9.3: Readability of a draft (Flesch Reading Ease).
An AI-drafted circular has 120 words, 8 sentences and 180 syllables.
A score of 60–70 is considered plain English, easy for most adults. If the score were below 50, you could prompt: "Rewrite with shorter sentences and simpler words."
9.8 Industry use cases
- UK government Copilot trial (2024–25). About 20,000 UK civil servants used Microsoft 365 Copilot for three months. Users reported saving 26 minutes a day on average, mostly on drafting documents, managing email, scheduling and preparing presentations. 82% said they would not want to go back, though 17% reported no time saving (GeekWire report).
- Consultants at Boston Consulting Group (2023). A Harvard-led field experiment with 758 BCG consultants found that, on tasks within AI's capability, consultants using GPT-4 completed over 12% more tasks, worked over 25% faster and produced over 40% higher-quality output. On a task outside AI's capability, AI users did worse, which the authors called the "jagged technological frontier" (Harvard Business School summary).
- The lesson from both studies. AI gives large gains on routine drafting, but the user must know which tasks it handles well and must check its work on the rest.
9.9 Practical exercises
- Convert these notes into a formal email: "inform students – internal exam postponed to 4 Nov due to fest – syllabus same – timetable on notice board."
- Paste a two-page article and prompt for a 100-word executive summary for a manager. Compare two prompt versions.
- Write a mock meeting transcript of 15 lines and use AI to produce minutes with an action items table. Mark any errors.
- Draft the objectives of a proposal for a college placement training programme using SMART criteria.
9.10 Review questions
- Describe a five-step workflow for producing an office document with AI.
- What elements should a prompt include for (a) a follow-up email and (b) an apology email?
- Analyse which parts of a business report AI can draft and which a person must own.
- List the standard sections of minutes of a meeting. What errors are common in AI-generated minutes?
- Compare the findings of the UK Copilot trial and the BCG study. What does "jagged frontier" mean?
- An employee saves 40 minutes a day for 230 days. How many 8-hour working days is that? (Answer: about 19.2 days.)
Knowledge sources
- Bishop, T., Microsoft AI tools saved British government workers 26 minutes a day, new study shows, GeekWire, June 2025 (reporting the UK Government Digital Service evaluation).
- Dell'Acqua, F. et al., Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality, Harvard Business School Working Paper 24-013, 2023 (HBS summary).
- Microsoft, Microsoft 365 Copilot.
- Google, Gemini for Google Workspace: Prompting Guide 101.
- Flesch, R., "A new readability yardstick," Journal of Applied Psychology, 32(3), 1948.
Independent study notes for learning purposes. Not affiliated with any university or course provider. Facts and case-study figures are linked to their original sources. ChatGPT, Gemini, Copilot, Jasper, Synthesia, Runway, Lumen5 and other product names are trademarks of their respective owners.