AI Workflows You'll Actually Use

What Counts as a Workflow

A single prompt is a sentence. A workflow is a sequence. You do the same two or three moves in order, repeatedly, and the output from one step becomes the input to the next. That repeatability is the difference between "I used AI once this week" and "AI is part of how I work".

A prompt saves a minute. A workflow saves an afternoon.

This session walks through seven workflows that land for most NT professionals. They're not magic, they're just the combinations I see consistently save time. Pick one, try it on something real next week, and notice whether it earns its place. Then pick another.

Raw Notes to Report Draft

Turn scattered notes into a first-draft report

Three steps, 15 minutes.

You've got half a page of messy notes from a site visit, a stakeholder conversation, or a week of observations. You need to turn it into a proper report. The AI does the structure; you do the judgement.

1. Brain-dump into one document. Paste or type everything you have into a single Word document or a fresh chat. Don't tidy it. Dates, names, quotes, half-sentences, typos, all in.
2. Ask the AI to propose structure before drafting.

   I've pasted raw notes from [describe]. Before you write anything, propose an outline for a report based on these notes. Tell me what the sections should be and what each would contain. Don't draft yet.

   Adjust the outline until it matches what you actually need.
3. Draft section by section, reviewing as you go.

   Now draft section 1 using the notes. Keep it to about 200 words, NT workplace style, plain language. Flag any claim where the notes are ambiguous or where I'd need to verify a detail.

   Review each section. Rewrite weak bits. Move on. By the end you have a draft you wrote, assisted, not a draft you blindly accepted.

Policy to Plain Language

Translate a policy document into what it means for your team

Three steps, 10 minutes.

A new or updated policy lands in your inbox. It's 14 pages, written in government prose, and you're expected to know what it means for your team. The AI can take the document-to-implications leap you'd otherwise do on your own.

1. Summarise the policy at three levels.

   Summarise this policy at three levels: one sentence (for someone walking past the office), one paragraph (for someone on my team), one page (for someone who'll need to apply it). Cover what it changes, who's affected, and when it takes effect.

2. Translate into "what this means for us".

   Given this policy and the fact that my team does [describe your work], what specifically changes for us day-to-day? What new things will we need to do, stop doing, or do differently? Anything that's genuinely unclear from the document?

3. Draft the update to your team.

   Draft a short Teams channel post or email to my team explaining the policy change in plain English. Tone: clear, not alarmist, one concrete "what you need to do" section. Under 200 words.

   Review, adjust, send. What used to be an afternoon is 15 minutes.

Multi-Document Research Synthesis

Find the common thread across several documents

Four steps, 30 minutes.

You've got five or six reports, papers, or policy docs on a topic and you need to make sense of them for a brief, a presentation, or your own understanding. NotebookLM or Claude earn their keep here.

1. Collect the documents in one place. Upload them to NotebookLM (purpose-built for this) or attach them to a single Claude chat. For Copilot, drop them in a SharePoint folder and point Copilot at it.
2. Ask for the landscape.

   Read all of the attached documents. Tell me: what's the broad topic they share, what positions or findings do they have in common, where do they genuinely disagree, and what's the most important question none of them answer clearly?

3. Drill into what you need.

   Focusing on [specific angle], what do these documents collectively say? Include direct quotes or references where they support something specific. Flag anything I should check against the original source.

4. Produce a usable output.

   Now draft a two-page synthesis of what these documents say, aimed at [audience]. Structure it as: what we know, what's disputed, what we don't yet know. Keep it honest. Include a short bibliography pointing back to which document said what.

Data to Story

Turn a spreadsheet into what it's saying

Three steps, 20 minutes.

You've got monthly stats, a program register, survey results. You need to explain what they mean to someone who won't read the sheet. Copilot in Excel plus a short writing workflow lands this.

1. Get Copilot to analyse the shape of the data.

   Look at this sheet. What are the top three things an outsider should know about this data? Any outliers, notable trends, or quality issues worth flagging? Be honest about what the data does and doesn't support.

2. Turn the findings into a narrative.

   Based on your analysis, write a three-paragraph narrative for a non-technical audience. Paragraph 1: what's happening. Paragraph 2: the one or two things I'd most want a reader to notice. Paragraph 3: what this might mean going forward. Plain language, no jargon.

3. Ask Copilot for one supporting chart.

   Suggest the single most useful chart from this data to support the narrative above. Describe what it would show, what chart type, and generate it if you can.

   Three paragraphs plus one good chart is almost always more effective than ten charts and no words.

Meeting Chain: Agenda to Follow-Up

From agenda to follow-up email, with the meeting in the middle

Four steps, across one meeting.

A full-cycle workflow that uses AI before, during, and after a meeting. Especially useful for recurring meetings you organise.

1. Agenda, before the meeting.

   I've got a meeting next Tuesday about [topic] with [who, roughly]. Help me draft a clear agenda. Maximum 5 items, with a time allocation and an owner for each. Start with the goal of the meeting in one sentence.

2. Prep brief for yourself.

   Based on the agenda and our recent conversations (from emails and Teams), what should I expect people to raise? What's the one outcome I should push for? What's the biggest risk of this meeting going sideways, and how do I pre-empt it?

3. Recap, after the meeting.

   Generate a recap of this Teams meeting: discussion points, decisions, actions with owners. Then draft a follow-up email to attendees with the recap and any asks that came out of it. Warm but brief.

4. Status update, a week later.

   Based on the meeting recap from last Tuesday, draft a short status update for the team showing progress against each action. Include anything I've since received by email that's relevant.

   The cumulative effect: meetings that stay on track, land decisions, and follow through.

Grant Application Support

Structure a grant application without the panic

Five steps, over a few days.

Grant applications are where AI can legitimately save days, if you use it for the lift-and-shift parts and keep your judgement on the specifics. The AI doesn't invent your program, you do. It just helps you articulate it in the form's vocabulary.

1. Decode the application.

   Read the grant guidelines. In plain English, tell me: what they're actually funding, what they care about most (priorities, keywords), what's a dealbreaker, what word counts apply to each section, and what evidence or documents I'll need.

2. Gather your raw material. Pull together what you already have: program descriptions, past evaluations, stats, quotes, letters of support. Drop into one folder or chat.
3. Draft each section with context.

   Draft the "Need" section using the attached background. Audience: the assessment panel. Tone: evidence-based but human. Word limit: 400. Use the framing the funder cares about (point back to their priorities).

4. Critique your own application.

   Now read this draft as a sceptical assessor. What's weak? What's unsupported? What would make you score this lower than competitors? Suggest three specific improvements.

5. Final polish and verification.

   Final pass: flag any statement in this application that sounds specific but isn't backed by evidence in my background material. Flag anything I may have overclaimed. Tighten wording where it's wasting words.

Stakeholder Comms Loop

Keep stakeholders informed without drowning them or yourself

Three steps, ongoing.

For anyone with external partners, community organisations, or senior stakeholders who need regular updates, this workflow replaces the "oh dear I haven't updated them in a month" panic.

1. Gather the month's updates.

   Based on emails, meetings, and Teams chats in the last four weeks involving [stakeholder or project], what's changed that they'd want to know about? Exclude anything routine or internal-only.

2. Draft the right update for the right audience.

   Draft a short update email for [stakeholder]. Tone matches our past correspondence with them. Include: what's happened, what's coming next, anything I need from them, and any open questions. Keep it under 300 words.

3. Schedule the next one. Set a calendar reminder to do this again at the same cadence. The AI makes it fast, the habit makes it work.

Building Your Own Workflow

The seven above are starting points. The real win is spotting your own repetitive work and turning it into a workflow. A template for doing that:

How to build a workflow

• Pick a task you do at least monthly. Not a one-off. Something recurring.
• Break it into 2–4 steps. What do you actually do in order? Write them down in plain English.
• For each step, ask: "is this AI-able?" Drafting, summarising, analysing, rewriting are usually yes. Personal judgement, external relationships, or decisions that need your authority are usually no.
• Write a prompt for each AI-able step. Include the context the AI will need. Save the prompts in your library (see the prompting session).
• Run the workflow next time the task comes up. Notice what doesn't work. Refine. By the third run it's smooth.
• Document for handover. If you're building workflows for your team, write them down. A good workflow survives the person who built it.