AI for customer service that stays human
Use AI to draft replies from your help centre, summarise long cases, calm upset customers and improve your help articles. For support agents and the people who run support teams.
Who it’s for: Support agents, team leads, CX and contact-centre managers.

AI is already changing support work. Used well, it helps agents answer faster and more consistently, and it tends to help newer agents most. Used badly, it gives confident wrong answers that customers hold you to. This course teaches you to use it well.
You'll draft replies grounded in your own help-centre articles and macros, the saved replies your team already uses. You'll summarise a long, messy ticket into a clear escalation note that sets out the issue, the steps already tried and exactly what you need. You'll write outage updates that never promise a fix time you don't have. You'll use AI to suggest calm wording for an angry customer while keeping every commitment within policy. You'll also find gaps in your help centre from ticket data.
Later lessons are for team leads and managers. You'll design an AI agent, software that answers customers on its own, and write the rules for when it hands over to a person. You'll decide which actions, like refunds, it may take. You'll check every conversation for quality, measure resolution honestly and re-forecast staffing before anyone changes headcount.
By the end you’ll be able to…
- Draft accurate replies grounded in your help-centre articles
- Summarise long ticket threads into a clean escalation note
- Design an AI agent with guardrails and human hand-off rules
- Measure AI resolution, CSAT and reopen rates honestly
Short on time? Start with the essentials
These 4 lessons cover the most useful ideas in this course. Each one takes about a minute.
What AI does to support work, by the numbers
Explain how AI assistance changes agent productivity and why the biggest gains go to newer agents.
Draft replies from your help centre, not thin air
Use AI to draft responses grounded in approved macros and knowledge articles, then edit for tone and accuracy.
Summarise a messy thread into an escalation note
Turn a long, multi-channel ticket into issue, steps tried, customer impact and the exact ask for tier 2.
Calm angry customers with AI-coached tone
Use AI to suggest empathetic, de-escalating wording while keeping every commitment within policy.
Every lesson in this course
This course has 12 video lessons of about a minute each, from the basics to expert habits. Each one ends with a question, three takeaways and its sources.
Foundations 2 lessons
What AI does to support work, by the numbersEssential
Explain how AI assistance changes agent productivity and why the biggest gains go to newer agents.
“AI help raised support agents' output 14%, and 34% for novices, in one big study.”
Draft replies from your help centre, not thin airEssential
Use AI to draft responses grounded in approved macros and knowledge articles, then edit for tone and accuracy.
“Same question, two AI drafts: one invented a policy, one cited it. Spot which.”
Building skills 4 lessons
Summarise a messy thread into an escalation noteEssential
Turn a long, multi-channel ticket into issue, steps tried, customer impact and the exact ask for tier 2.
“Dozens of messages, three agents, one angry customer. Here's the five-line handover.”
Write outage updates and macros customers trust
Turn engineering incident notes into a status-page update, a holding macro and a follow-up message that are accurate, consistent and never promise a fix time you don't have.
“Payments are down and 400 tickets just landed. One clear message, every channel.”
Calm angry customers with AI-coached toneEssential
Use AI to suggest empathetic, de-escalating wording while keeping every commitment within policy.
“'This is the third time I've asked!' Write the reply that calms instead of inflames.”
Fix your knowledge base using ticket data
Cluster ticket topics to find missing or outdated help articles, then draft, review and publish fixes.
“An AI agent is only as good as your help centre. Find the holes first.”
Advanced 5 lessons
Design an AI agent that knows when to hand off
Write the escalation triggers for a customer-facing AI agent, covering topics, sentiment, repeated failures and money, so it hands off before it fails.
“Gartner predicts AI agents will resolve 80% of common issues by 2029. Which 80%?”
Let AI take actions safely: refunds and account changes
Decide which backend actions an AI agent may perform, with limits, customer authentication and audit logs.
“Answering a question is easy. Issuing a refund is a different level of risk.”
Quality-check every conversation, not a small sample
Use AI to score conversations against your QA rubric and calibrate it against human reviewers before using the scores to coach agents.
“Manual QA reads a sliver of tickets. AI can read all of them. Can you trust its scores?”
Your chatbot's promises are your promises
Apply lessons from cases and rules such as Moffatt v Air Canada (Canada), the CFPB's chatbot findings (US) and the EU AI Act's chatbot disclosure duty, and check what applies where you serve customers.
“Air Canada said its chatbot was responsible. The tribunal made the airline pay.”
Measure AI resolution honestly
Track true resolution, reopen rate, CSAT and escalation quality instead of vanity containment rates.
“High 'containment' can just mean customers gave up. Measure what really matters.”
Expert 1 lessons
Re-forecast staffing when AI takes the easy tickets
Recalculate agent requirements using the volume AI truly resolves, the longer handle time of the tickets left and reopen rates, before anyone changes headcount.
“AI resolves 40% of tickets. Do you need 40% fewer agents? Usually not.”
Apps we recommend
These are the tools this course uses, and what each one is best for. We have no partnership with any of them.
Fin
Best for: AI agent that resolves customer conversationsIntercom's AI agent works across chat, email and voice on major helpdesks, priced per resolution with testing tools.
PaidZendesk AI
Best for: AI agents and copilot inside ZendeskAdds AI agents, agent copilot, ticket summaries and quality assurance to the widely used Zendesk helpdesk.
Free trialSierra
Best for: Enterprise AI agents that take actionsBuilds branded AI agents that handle returns, bookings and account changes, with guardrails for large companies.
PaidShip one AI-assisted fix for a top contact reason
Take 50 anonymised tickets from one contact reason. With approved tools, cluster them, fix or write the help articles, generate grounded draft replies and test them on 20 tickets. Then write hand-off rules, a metrics baseline and a short note on what changes for the team's roles and staffing.
What you’ll make: Updated articles, a 20-ticket test log with pass/fail, hand-off rules, a metrics baseline and a one-page team impact note.
Check your own work
- Every AI draft in the test log cites the help article it relied on
- Each failure is tagged (wrong policy, missing info, tone) with the fix made
- Hand-off rules cover angry customers, money, account changes and repeated failures
- The baseline reports resolution, reopen rate and CSAT, not only deflection
- All tickets were stripped of names, emails and account numbers first
Try a ready-made prompt
Copy a better prompt for a real task, and see why it works better than the version most people type.
Further reading and viewing
We’ve chosen these articles, research papers, videos and guides from other trusted sources.
Generative AI at Work
Landmark study of 5,000+ support agents: AI raised productivity 14%, and 34% for novices.
Chatbots in consumer finance
Regulator's view of how badly built chatbots that block human help can harm customers and break the law.
Air Canada found liable for chatbot's bad advice on plane tickets
The tribunal case that established a company owns what its support chatbot tells customers.
Article 50: Transparency Obligations for Providers and Deployers
The EU rule that people must be told when they are talking to an AI system.
Questions about this course
How can I use ChatGPT for customer service replies?
Give it the customer's message and the help article or macro that answers it. Ask for a reply based only on that source, in your team's tone. Then check every fact, price and promise before you send it. If you paste in customer messages, use a tool your company has approved and remove personal details first. The course shows you how to draft from your help centre, not thin air.
When should an AI chatbot hand customers over to a person?
It should hand over before it fails, not after. Good hand-off rules cover certain topics, an upset customer, repeated failed answers and anything involving money. Tools such as Fin, Zendesk AI and Sierra let you set rules like these. The course shows you how to write those triggers and test them on real tickets. You'll also decide which actions, such as refunds or account changes, an AI agent may take, with limits and checks.
Are companies responsible for what their chatbot says?
Generally, yes. In Moffatt v Air Canada, a Canadian tribunal held the airline responsible for wrong refund advice its chatbot gave. In the US, the CFPB has published findings on chatbots in consumer finance. The EU AI Act requires people to be told when they're talking to AI. The course uses these as examples. Check what applies where you serve customers, and keep customer data in approved tools.
How long are Tau's customer support lessons?
Each lesson takes about a minute. You watch a short video, answer one question, then get three takeaways and the sources. This course has 12 lessons, from drafting grounded replies to measuring AI resolution honestly. Your first 5 lessons in Tau are free, and Tau Pro gives you every course and lesson.
Get early access to this course
Tau opens soon. Join the waitlist and we’ll email you as soon as you can start this course.
We’ll only email you about Tau, and you can unsubscribe any time. Privacy policy.
We’ll email you as soon as Tau opens.


