The pillar where a machine speaks to a person with no one in between — the personalisation wins, the human cost when the metric is wrong, and what “done right” looks like.
July 2026 · Case 5 of 6
As you read — hold this question
When AI can touch every stage of the customer journey, where — if anywhere — does a human still have to be?
45roles
cut at CommBank on a wrong AI metric — then reversed, in writing.
Customer experience is where AI talks straight to your customer. Done well it is Netflix-grade personalisation and a journey mapped end to end. Done by the wrong metric, it is 45 CommBank jobs cut on a bot that hadn’t actually reduced call volumes — and reversed once the union checked. The difference is what you choose to measure.
The third pillar, taught straight
The textbook versionQuiz: AI & customer experience
Answer, personalise, anticipate — and map the journey
Video 4 frames AI in customer experience three ways: it answers (chatbots and voice bots, 24/7), it personalises (tailored offers and content), and it anticipates (predicting needs before they’re raised). The exemplars: CommBank’s “Yello” programme; Netflix (~80% of what people watch comes from recommendations — Netflix, 2015); and Amazon (~35% of purchases from recommendations — a much-cited but dated 2013 McKinsey figure). The tool to know is the customer journey map — awareness → consideration → purchase → onboarding → loyalty — with AI now touching every stage. Notice there is no stage left where a human must be involved. That is the promise, and the problem.
When the metric is wrong
The rest of the story
Kathryn Sullivan, and the cost of measuring the wrong thing
Kathryn Sullivan — 63, twenty-five years at the Commonwealth Bank — spent years helping train the bank’s chatbot. In July 2025 she was made redundant with one hour’s notice, one of 45 roles cut in “Customer Service Direct,” blamed on the “Hey CommBank” voice bot cutting call volumes. Then the Finance Sector Union checked: volumes were rising. On 21 August 2025 the bank reversed all 45 redundancies, conceding the roles “were not redundant.” At the October 2025 AGM, chairman Paul O’Malley said: “We made a mistake. We didn’t adequately consider all the relevant business considerations.”
Klarna did the same in reverse: its AI handled 2.3M chats a month and cut resolution from 11 minutes to under 2 — then walked it back, its CEO saying “investing in the quality of human support is the way of the future for us” (May 2025). The lesson in one line: you can win on cost and lose the customer.
Done right
The other side
What good actually looks like
Sierra — pay for outcomes
Bret Taylor’s AI-agent company charges mainly when the AI resolves the issue — not per conversation. When the vendor only earns on a genuine resolution, “just deflect the customer” stops being the goal.
Lorikeet — humility first
The bot’s first job is knowing what it can’t do: it answers only when sure and in scope, a second AI checks it, and it hands off by default. Human handle time goes up — because only the hard cases reach people. That’s quality, not deflection.
Take this away
The North Star isn’t the cheapest system that still technically answers — it’s a great experience, human or not, delivered honestly. Measure the experience, not just the saving.
Quick recall — without looking back
Test yourself on this case
Question 1 of 3
Name the three ways AI reshapes customer experience, with an exemplar for each.
It answers (chatbots/voice bots, 24/7 — CommBank’s ‘Hey CommBank’/Yello); it personalises (Netflix ~80% of watch-time, Amazon ~35% of purchases from recommendations); and it anticipates (predicting needs before they’re raised). The organising tool is the customer journey map.
Question 2 of 3
What went wrong at CommBank, and how was it resolved?
CommBank cut 45 ‘Customer Service Direct’ roles, blaming the ‘Hey CommBank’ voice bot for falling call volumes. The Finance Sector Union showed volumes were actually rising; on 21 Aug 2025 the bank reversed all 45 redundancies (‘not redundant’), and chairman Paul O’Malley admitted at the AGM: ‘We made a mistake.’ The lesson: measuring the wrong thing (cost/deflection) fails the customer.
Question 3 of 3
Contrast ‘deflection’ with ‘done right’ using Sierra and Lorikeet.
Sierra charges mainly when the AI actually resolves an issue — not per chat — so deflection stops being profitable. Lorikeet leads with humility/handoff-by-default and a second-AI check, which pushes human handle time up because only hard cases reach people. Both optimise resolution quality, not cost-cutting.
Finance Sector Union; ABC / Bloomberg coverage of the 45-role reversal (Aug 2025) and the AGM (Oct 2025); Klarna press releases and CEO statement (May 2025).