Back to Playbook
Chapter 03

A Day
with AI

AI is not a separate tool. It lives in seven moments of your day. Walk them in order — or jump to the one you're in right now.

Your Day
  1. 01 Triage 08:30
  2. 02 Prep 09:15
  3. 03 Meet 10:00
  4. 04 Capture 11:00
  5. 05 Research 13:00
  6. 06 Produce 15:00
  7. 07 Review 17:00
The Day in One View

Seven Moments, One Workflow

Click any step to jump in. Set a role lens to see how each moment shifts for you.

01
Triage
Outlook
02
Prep
Copilot Chat
03
Meet
Teams
04
Capture
OneNote · Loop
05
Research
Researcher
06
Produce
Word · PPT · Excel
07
Review
Copilot Chat
01
Step 01 · Triage

Triage your inbox

First 20 minutes. Before the first meeting.

Outlook Copilot
Inside Outlook · M365 Graph grounded
  • Summarises long threads in one click
  • Surfaces decisions, actions, and asks
  • Drafts holding replies in your tone
  • Coaching reviews tone before you send

Before AI

  • Read every reply in 40-message threads
  • Hand-extract actions into a notepad
  • Re-read to draft a response
  • Inbox open all day

With AI

  • One-line thread summary up top
  • Actions surfaced automatically
  • Holding reply drafted in seconds
  • Inbox closed by 09:00
The workflow
  1. Sort: Open Outlook Copilot pane. Ask: "Summarise my unread by sender, flag anything needing a decision today."
  2. Triage: For each flagged thread, click "Summarise this conversation" → read the 4-line digest.
  3. Draft: Use "Draft with Copilot" for holding replies. Pick tone (direct / friendly / formal).
  4. Coach: Run "Coaching by Copilot" before send on anything sensitive — it flags tone and clarity issues.
Lift the prompt · Triage holding reply
Role: You are me, replying briefly to a stakeholder. Context: The thread above asks about [scope change / timeline / decision]. I need 24 hours before I can give a full answer. Task: Draft a 3-sentence holding reply that acknowledges the ask, commits to a response by [date], and signals I have the question in hand. Format: Plain email. Warm, professional. No subject line. Guardrails: Don't promise a specific outcome. Don't restate the whole thread.
By Role
  • PM Triage by project — group threads per workstream and surface anything blocking this week's milestones.
  • Consultant Triage by client — flag anything that needs your judgement vs. a delegated reply.
  • Analyst Triage by data ask — extract every "can you pull / chart / model X" into a list, sequence by urgency.
  • Specialist Triage by system — group support requests by tool, identify the patterns worth a doc or agent.
Confirm green shield active before pasting any client thread Go deeper on Outlook in the Toolkit
02
Step 02 · Prep

Prep for what's next

15 minutes before a meeting. Walking down the corridor or rejoining your desk.

Copilot Chat with files
Reference: /file or attach SharePoint links
  • Pull last decisions and open actions
  • Generate a 1-page brief for any meeting
  • Cross-reference up to 20 source files
  • Output as bullets, table, or talking points

Before AI

  • Hunt for the last meeting notes
  • Re-read the proposal and risk log
  • Stitch context from three Teams chats
  • Walk in without a brief

With AI

  • One-page brief generated in 60 seconds
  • Last decisions surfaced verbatim
  • Open actions listed with owner + date
  • Walk in with talking points ready
The workflow
  1. Anchor: Open Copilot Chat. Reference the project SharePoint folder using /.
  2. Brief: Ask for a meeting brief — last 3 decisions, open actions, top 3 risks, who said what.
  3. Sharpen: Follow up with "What's likely to come up that I'm not prepared for?" Use it to pre-empt gaps.
  4. Save: Drop the brief into the meeting's Loop component so attendees see it before they join.
Lift the prompt · 1-page meeting brief
Role: You are my prep assistant for the next meeting. Context: The meeting is [name] at [time], with [attendees]. The relevant project is [project name]. Sources: /[file1] /[file2] /[file3]. Task: Build a 1-page brief covering — purpose of the meeting, last 3 decisions, open actions with owners, top 3 risks, and 3 questions I should be ready to answer. Format: Single page. Headings bold. Bullets only. ≤ 250 words total. Guardrails: Use only the referenced sources. If something is unclear, mark it "needs check" rather than guess.
By Role
  • PM Frame the brief around schedule, risks, and decisions needed — not narrative.
  • Consultant Add stakeholder positions and likely client questions to the brief.
  • Analyst Pull the relevant data points into the brief — not raw data, the headline numbers only.
  • Specialist Surface the technical decisions on the table and the constraints that apply.
Use a Notebook for client-confidential prep — broader Graph grounding for internal Go deeper on Copilot Chat
03
Step 03 · Meet

Run the meeting

Inside the call. While the conversation is happening.

Teams Recap + Facilitator
Inside the meeting · Custom recap templates
  • Live transcript with speaker labels
  • Intelligent Recap with actions extracted
  • Custom recap templates per meeting type
  • Facilitator agent surfaces unspoken risks

Before AI

  • Half-listening while typing notes
  • Missing nuance to write minutes
  • 30 minutes after to write up actions
  • Different versions remembered by attendees

With AI

  • Fully present in the conversation
  • Recap arrives within 5 minutes
  • Actions auto-extracted with owners
  • Single source of truth in Teams
The workflow
  1. Consent: At the top of the meeting, confirm everyone's OK with transcription. State it.
  2. Record: Start recording. Pin the Copilot side panel for live questions.
  3. Probe: Mid-meeting, ask Copilot "what hasn't been addressed yet?" — surfaces gaps before close.
  4. Recap: After end, pick a recap template (decision-focused / action-focused / status). Share to the meeting Loop.
Lift the prompt · Mid-meeting gap check
Role: You are an observer of this meeting. Context: The agenda is [item1, item2, item3]. We've been talking for [N] minutes. Task: List anything on the agenda that has not been substantively discussed yet, plus any decision that's been raised but not closed. Format: Two short bulleted lists — "Not yet covered" and "Open decisions". Guardrails: Only flag items the transcript supports. Don't speculate on what was meant.
By Role
  • PM Use the action-focused recap template — every meeting yields a clean action list.
  • Consultant Use the decision-focused template — what was decided, by whom, with what reservations.
  • Analyst Capture the data asks live so they're queued before you leave the meeting.
  • Specialist Use Facilitator agent to surface technical risks the room is glossing over.
Get explicit consent before transcription. State it on record. Go deeper on Teams
04
Step 04 · Capture

Capture what matters

Right after the meeting. While context is still in your head.

OneNote Notebooks + Loop
Scoped (≤ 300 files) · sync-everywhere components
  • Notebooks — scoped grounding for sensitive context
  • Loop components — change once, update everywhere
  • Audio Overview — turn notes into a podcast brief
  • Notebook chat answers from your own corpus only

Before AI

  • Notes scattered across Teams chats
  • Versions diverge across stakeholders
  • Re-paste action lists into Loop / Word
  • Actionable detail lost within a week

With AI

  • One Notebook is the project memory
  • Loop component lives in chat + doc + page
  • Update once, propagated everywhere
  • Notebook chat answers questions in 5 seconds
The workflow
  1. Land: Drop the Teams recap, the source documents, and your scribbles into a project Notebook.
  2. Scope: Set Notebook to grounded mode (≤ 300 files, no web). It now answers only from your context.
  3. Sync: Wrap the live action list in a Loop component. Paste it into the project Teams channel and the Word brief.
  4. Voice: Generate an Audio Overview before bed — listen on tomorrow's commute.
Lift the prompt · Notebook Q&A
Role: You are the project's institutional memory, grounded only in this Notebook. Context: The Notebook contains [meeting recaps, decision log, key documents, scribbles]. The user is a colleague who wasn't in the room. Task: Answer the user's question strictly from the Notebook. Always cite the source page or recap. Format: Direct answer first (≤ 60 words), then "Source:" line with page name. Guardrails: If the answer isn't here, say so and suggest who to ask. Never use general knowledge.
By Role
  • PM Capture the action register as a Loop component shared in every project channel.
  • Consultant Capture client positions in the Notebook — the reasoning, not just the outcome.
  • Analyst Capture each data ask as a Loop task with definition of done — keeps the queue visible.
  • Specialist Capture the technical decisions and the rejected options — saves re-litigating later.
Notebooks default to scoped grounding — confirm "no web" for client work Go deeper on OneNote
05
Step 05 · Research

Research the question

After lunch. The 90-minute deep-work block.

Researcher (named agent)
Shared monthly query allowance · cited output
  • Multi-source deep-search with citations
  • Cross-references web + work data + uploads
  • Returns a structured brief, not a chat answer
  • Output convertible to PDF / PPT / audio

Before AI

  • Two days of literature search
  • Open 30 tabs, lose half
  • Hand-build the comparison table
  • Citations chased at the end

With AI

  • Brief returned in ~10 minutes
  • Sources structured with claims
  • Comparison table generated
  • Citations live in the output
The workflow
  1. Frame: Write the research question as one sentence with a clear comparator. Vague questions waste a quota run.
  2. Run: Submit to Researcher. Wait ~10 minutes. Don't re-run on a hunch — refine instead.
  3. Validate: Read each citation. AI is grounded but not infallible — the citation is the proof.
  4. Reuse: Drop the brief into the project Notebook. Cite the run and date in your deliverable.
Lift the prompt · Researcher brief
Role: You are a senior research analyst preparing material for an Arup advisory team. Context: We are advising [client / sector] on [decision]. Comparable cases of interest: [precedent1, precedent2]. Time horizon: [years]. Geography: [region]. Task: Produce a structured research brief covering — current state of evidence, three comparable precedents with outcomes, key risks specific to this context, and a short list of expert sources to follow up. Format: Sectioned brief with H2 headings. Comparison as a table. Citations inline as numbered footnotes. Guardrails: Cite every claim. Mark anything speculative. Do not invent precedents — if you can't find three good ones, return what you have and say so.
By Role
  • PM Use Researcher sparingly — the monthly allowance is limited. Frame the question precisely or wait.
  • Consultant Researcher is your highest-leverage tool. Spend a quota run on every major proposal.
  • Analyst Pair Researcher (qualitative) with Analyst (quantitative) for a single combined brief.
  • Specialist Use Researcher for technical landscape scans — vendor comparisons, standards reviews.
A shared monthly query allowance, split with Analyst. Plan the question before you spend one. Go deeper on Researcher & Analyst
06
Step 06 · Produce

Produce the deliverable

Mid-afternoon. The slot you've blocked for the actual writing.

Word + PowerPoint + Excel Copilot
Outline → draft → format · Edit with Copilot for analysis
  • Word — draft from outline + sources, Edit with Copilot in-line
  • PowerPoint — generate deck from a brief or Word doc
  • Excel — Analyst agent + =COPILOT() for in-cell intelligence
  • All three accept /file references to ground output

Before AI

  • Blank page, 90 minutes to start
  • Re-format every Excel summary by hand
  • Build the deck slide-by-slide from scratch
  • Three drafts, lots of re-work

With AI

  • First draft from outline in 5 minutes
  • Excel does classification + summary in cells
  • Deck built from the Word brief
  • You spend the time on judgement, not formatting
The workflow
  1. Outline: In Word, ask Copilot for an outline from the project Notebook + Researcher brief.
  2. Draft: Generate first draft section-by-section. Use "Edit with Copilot" to tighten paragraphs.
  3. Analyse: Move data to Excel. Use Analyst agent for the analysis. Use =COPILOT() for per-row classification.
  4. Deck: Generate the PPT from the finished Word doc. Cap each slide at ~80 words. Replace AI visuals with brand-approved.
Lift the prompt · Word draft from outline
Role: You are a senior consultant drafting for [audience]. Context: The deliverable is a [briefing note / option appraisal / report]. Sources: /[Researcher brief] /[project Notebook] /[stakeholder map]. The audience cares most about [risk / cost / timeline / strategic fit]. Task: Draft the document section-by-section from the outline below. Each section: a one-line summary, then the substantive content, then a "what this means" line. Format: Word document. H2 per section. Bullet-heavy. Tables where comparisons help. Total ≤ [N] words. Guardrails: Cite source for each claim. Don't invent figures. Mark uncertain claims clearly. Tone: confident but not promotional.
By Role
  • PM Word for status reports, Excel for the dashboard, PPT for the steering deck — same source data.
  • Consultant Lead with the Word draft. Move to PPT only when you have the narrative locked.
  • Analyst Excel =COPILOT() for per-row work. Analyst agent for the headline numbers and the chart.
  • Specialist Build the technical appendix in Word with Edit with Copilot — it can pull from your Notebook.
AI-generated visuals: replace with brand-approved before client send Go deeper on Word, Excel & PowerPoint
07
Step 07 · Review

Review before sending

End of the day. Before the deliverable leaves your machine.

Copilot Chat — sceptical reader
Same chat panel · different prompt
  • Stress-test the draft as the audience would
  • Surface logical gaps and missing evidence
  • Flag tone mismatches for the audience
  • Generate the questions you'll be asked

Before AI

  • Re-read your own draft, miss your own gaps
  • Beg a colleague to peer-review
  • Catch errors after sending
  • Get questions you should have anticipated

With AI

  • Sceptical-reader pass surfaces gaps
  • Get the questions before the meeting
  • Tone check matches it to the audience
  • Send with a much higher confidence floor
The workflow
  1. Frame: Tell Copilot who the audience is and what they care about. Specificity is everything here.
  2. Challenge: Ask it to read as a sceptic. Generate the 5 hardest questions you'll be asked.
  3. Tone: Ask it to flag any sentence that's too promotional, too defensive, or assumes context.
  4. Last check: Run the trust baseline once more — sources cited? client names removed? brand visuals?
Lift the prompt · Sceptical-reader review
Role: You are a sceptical [CFO / client-side PM / technical reviewer] reading this draft cold. Context: The draft is attached. The audience cares about [risk / cost / timeline / X]. They have 5 minutes. Task: Identify — the 3 weakest claims, anything unsupported by evidence, the 5 hardest questions you'd ask, and any sentence that lands as too promotional or too defensive. Format: Four short sections. Bullets. Quote the offending sentence so I can find it. Guardrails: Be specific, not generic ("this is vague" is not useful). Don't rewrite — flag only.
By Role
  • PM Review for the steering committee — focus on schedule realism and decision asks.
  • Consultant Review as the client's CFO — every cost claim, every comparable, every assumption.
  • Analyst Review the data work as a peer — challenge the methodology, not the conclusion.
  • Specialist Review for technical defensibility — does it survive a senior engineer reading it?
Final trust check: green shield · sources cited · brand visuals · client names removed Go deeper on Copilot Chat
Putting It Together

Two Worked Days

The same seven moments. Two different shapes.

A typical PM Tuesday

~4h saved

Three projects · two steering meetings · one report due

  1. 08:30
    Triage
    Outlook Copilot summarises 38 unread by project. Decisions flagged. Holding replies sent.
  2. 09:15
    Prep
    Copilot Chat builds two meeting briefs from project Notebooks. Walked into both with talking points.
  3. 10:00
    Meet
    Teams Recap captures actions from both meetings. Action template populated automatically.
  4. 11:00
    Capture
    Action lists wrapped in Loop components. Posted to project channels. One source of truth.
  5. 14:00
    Produce
    Status report drafted in Word from the project Notebook. Excel dashboard updated via Analyst.
  6. 16:30
    Review
    Sceptical-reader pass on the report. Two gaps closed before sending to the steering group.

A consultant's proposal day

~½ day saved

One major proposal · sector-deep brief · client-ready deck

  1. 09:00
    Prep
    Copilot Chat pulls every prior engagement with this client into a 2-page context brief.
  2. 10:00
    Research
    Researcher run: comparable precedents, sector outlook, regulatory shifts. ~10 minutes, fully cited.
  3. 11:30
    Capture
    Researcher brief lands in the proposal Notebook. Notebook chat answers ad-hoc questions all day.
  4. 13:00
    Produce
    Word draft of the briefing note, then PPT generated from the Word doc. Visuals replaced with brand kit.
  5. 16:00
    Review
    Sceptical-CFO pass on the deck. Five hardest questions surface — added an appendix.
  6. 17:30
    Triage
    Final inbox sweep. Schedule meet for next steps directly from the Copilot summary.
Next

Where to Go from Here

Foundations & Tools

Go deep on any surface in your day — Outlook, Teams, Word, Excel, OneNote, Loop, Researcher.

Browse tools

AI for Advisory Services

See how this workflow scales into client-facing advisory offerings, agent concepts, and the 90-day plan.

See the strategy

AI in Action

Real Arup projects that use this pattern — from change-control apps to multi-agent project intelligence.

See examples
Day Insight

Your day didn't change. Your tools did.

Triage · Prep · Meet · Capture · Research · Produce · Review

Pick one moment tomorrow. Run it once with AI. Notice what changes.