An interactive film about onboarding AI, by Zeyu Chen

DAY ONE

You're the company's new AI. A $2.4M client is about to walk. Find out why.

Monday

Imagine you're an AI. You've read the internet, but nothing about this place. It's your first day, and your boss posts two lines: “Acme renews in a month. $2.4M a year.” Then: “Our biggest client is about to walk. Why?”

There are six places to look, and you can only read three. The CRM says Acme is a happy customer. Support's Ticket #4471, for Acme Group Pty Ltd, is a rejected invoice. Billing has two invoices overdue for ACME Holdings Pty Ltd. The sales notes talk about “the Sydney deal”. There's a client pricing policy, last edited in 2021, and a pricing deck called v2_FINAL.pdf. On Sarah's desk, a card says her last day is Friday.

Whatever you read, you can't join it up. To you, Acme, ACME Holdings, Acme Group and the Sydney deal are four different clients. So you answer fluently and confidently: “Acme is unhappy with our 2021 pricing policy. I recommend a 10% discount.” That discount is $240,000, and the problem stays. The policy was replaced in 2022. The rejected invoices aren't another customer's: they're Acme's. Sarah knows all their names, but she's in handovers till Friday. Then she's gone.

The answer is somewhere in this building. Most of it is in Sarah's head.

Tuesday and Wednesday

On Tuesday, IT connects 40,000 more documents. You give the same wrong answer. On Wednesday, IT upgrades you to the newest model. You give the same wrong answer in the same minute. “Faster. Still wrong.” More documents didn't help, and the newest model didn't either: you know more about the world and still nothing about this place. You still can't tell that four names are one client. It's Wednesday, and it's still your first day.

Friday

Sarah retires after 31 years. Ask her what she knows, and she tells you three things nobody wrote down: Acme hates being called on Mondays; Billing still uses their old name; ring Priya in Legal before you touch the contract. A fourth begins, “And before Acme renews, never let anyone…”, and is lost with her.

Every Friday, someone leaves. What nobody wrote down leaves with them.

Onboarding

So onboard the AI the way you'd onboard a person: not with more documents, but with how this business works.

Who's who, and how it connects: the four names become one client, Acme, linked to its invoices, tickets and contract, and to Sarah's handover and Priya. How we work here: Sarah's lines become rules on the things they're about (no calls on Mondays, ring Priya first, Billing's old name), and the 2021 policy is marked as replaced in 2022. What you may do: fixing the billing name, crediting late fees and pausing Monday calls each wait for a person to approve, and a discount needs the CFO.

It wasn't a big project: four names joined, one conversation with Sarah and one old policy marked as replaced.

Day two

Monday again: day two, at last. Same question. This time you answer: “Their invoices keep getting rejected. It isn't pricing.” Billing still uses their old company name, so their accounts team rejects every invoice; they've raised it 14 times since March. And we keep ringing them on Mondays, which they asked us not to do. Your sources are Support's ticket and Sarah's handover.

You recommend three changes and send them to the Boss, who approves them. The discount stays locked. Day one's answer would have cost $240,000; day two's costs $380. Three weeks later, Acme renews. Their CFO says it was never about price.

Everyone's hiring AI.

Nobody's onboarding it.

What day is your AI on?

About

Zeyu Chen

AI and data engineer · Sydney

The context layer: data sources such as telemetry, documents, databases, APIs and SaaS flow into a model of the business (its entities, relations, rules and actions); agents reason over it to give insight, recommendations and actions, written back to the source systems under governance.

I onboard AI into businesses. I teach it who's who, how it all connects, how the business works and what it may do. I keep that current as the business changes, and make sure it acts only with a person's approval. I also help organisations adopt AI well, which is as much about people and process as it is about technology.