Reclaim the hours.
The final module is pure utility: pointing everything you've learned at the repetitive, high-friction tasks that eat a team's week. Done right, AI isn't a toy — it's an administrative conduit that gives time back. This class shows you how to find the right targets, prove the ROI, and automate without breaking trust or the law.
Find targets
Spot high-friction, repetitive tasks worth automating.
Prove value
Quantify time and cost saved with a real ROI number.
Pilots that stall
Flashy demos with no metric, owner or governance go nowhere.
True or false?
"The best AI strategy is to automate as many tasks as possible, as fast as possible, across every department at once." True or false?
Find the friction
Successful AI adoption starts with diagnosis, not tools. Walk the value chain — every step of how work actually flows — and mark the bottlenecks: the handoffs, queues and manual re-typing where time leaks. The best first targets are tasks that are high-volume, repetitive, rule-ish and costly in time.
Automating a broken process just makes the mess faster. Fix and simplify the workflow first, then apply AI to what remains. And start small: one narrow, measurable win earns the mandate for the next.
The triage matrix
Sort candidate tasks by two axes: how repetitive they are, and how much judgement / risk they carry. That tells you what to automate, what to assist, and what to leave alone.
High repetition · low risk
Data entry, categorising emails, resizing assets, transcription. AI does these tirelessly — your biggest, safest wins.
High repetition · high risk
Invoice approval, résumé screening. AI drafts and flags; a human approves. Keep a human in the loop.
Low repetition · low risk
Brainstorming, first drafts, research summaries. AI as a thinking partner; you curate.
Low repetition · high risk
Strategy, sensitive HR calls, legal sign-off. Judgement-heavy, one-off, high-stakes — keep these human.
The ROI calculator
Pick a repetitive task and estimate its numbers. The calculator shows the hours and money it could free per year — the figure that turns "let's try AI" into a decision leadership can sign off.
Build the workflow
The leap from "using a chatbot" to "AI in the business" is integration — wiring models into the systems where work happens.
LLMs + your tools
Use automation glue (Zapier, Make) and APIs so AI reads from and writes to the tools you already run — triggered, not copy-pasted.
Your own knowledge base
Retrieval-augmented generation lets staff ask plain-language questions and get answers grounded in (and citing) your internal documents — securely.
Multi-agent for departments
Specialised agents (researcher, drafter, checker) coordinate on bigger tasks — with humans setting goals and approving outputs.
Upskill, don't just cut
Shift staff from doing the repetitive task to supervising and improving the AI that now does it. Change management is the project.
Every deployed AI needs a named owner, a privacy check (GDPR / where data lives / does it train a vendor model), a monitoring plan, and a kill switch. Track KPIs — time saved, error reduction, cost per task — or the pilot is a hobby.
The overloaded coordinator
A community organisation has one coordinator drowning in repetitive admin: sorting 200 inbound emails a week and answering the same policy questions over and over.
"Automate the whole job"
An over-ambitious plan to automate everything stalled — too broad, no metric, no owner, and staff feared for their roles.
One scoped win
They picked one task — email triage — measured the baseline (≈8 hrs/week), deployed AI sorting with a human reviewing flagged-urgent, and added a RAG bot for policy FAQs.
The win came from scope + a metric + an owner — and from framing AI as removing drudgery, not removing people. The coordinator now manages the system instead of doing the grind.
Scope a pilot
Find one tool or pattern for automating a back-office task, and note the privacy consideration that comes with it.
What to look up
"intelligent document processing tool" · "RAG internal knowledge base" · "GDPR AI compliance checklist".
Pitch one automation
Write a 4-line pitch for one automation: the task, why it's a good target, the human checkpoint, and the KPI you'll track to prove ROI.
Success rubric — tap to expand
Needs work: "automate everything", no metric.
Getting there: a task named, but no checkpoint or KPI.
Solid: a scoped task with a human checkpoint.
Excellent: a high-volume, low-risk task, a clear human checkpoint, and a measurable KPI with a baseline to compare against.
Quiz — 5 questions
Common mistakes
Boil the ocean
Automating everything at once. Fix: one scoped, measurable win first, then scale with the lessons.
No metric
"It feels faster." Fix: baseline before, measure after — hours saved, error rate, cost per task.
Ignoring people
Dropping AI on a fearful team. Fix: involve them, upskill, frame it as removing drudgery.
Privacy afterthought
Pasting sensitive data into a free tool. Fix: check data handling and compliance before deploying.
Time reclaimed
If AI gave your team back several hours a week, what would you want them to spend that time on? Automation is only worth it if the reclaimed time goes somewhere meaningful.
From tokens to teams, you can now reason about AI, command it, question it, and put it to work — responsibly. That's the whole point: not passive consumers, but capable, ethical digital creators.