An example I made up
Everything in this example is made up. Illustrative figures and dates.
The situation. Hannah Wijaya is head of customer operations at an office furniture importer with its head office in Singapore. A customer in Kuala Lumpur is opening a new office on 3 March. It has ordered 300 chairs and wants them delivered by 27 February, so its team can set up the office before the opening. The chairs are in a container on its way to Port Klang. It is 17 February.
What the AI produced. Hannah asks SittellaAI when the chairs can be delivered. Its answer reads: "The container clears customs at Port Klang by 24 February, and delivery by 27 February is possible. Based on the shipping line's arrival notice and the port's published clearance times. I am 70 percent sure." She can't tell whether it's right.
The check. She asks herself what would make the date wrong. A customs inspection would add days. So she phones the customs broker her company uses. The broker tells her that inspections this month are adding days to clearance.
The decision, and who was told. Hannah doesn't promise 27 February, and she doesn't wait either. She tells the customer that 27 February is likely, not certain, and that she will confirm on 25 February. She books the delivery lorry for 27 February, with free cancellation until 25 February. She holds 60 chairs from the showroom in case the container is late. Then she tells the sales lead what has been promised to the customer and what hasn't.
What happened after. On 25 February the container still hadn't cleared. Hannah told the customer, as she had said she would, and kept the lorry. It took the 60 showroom chairs on 27 February, so the customer's team could start setting up for the opening on time. The container cleared on 26 February, and the chairs from it arrived on 28 February. The office opened on 3 March with all its chairs. Three months later, the customer ordered chairs for a second floor.
What I do when I can't tell
Hannah's position is a normal one. Most of the AI answers I get arrive faster than I can check them. Some of them I can't check fully at all. The answer gives me a date or a figure, with a number for how sure the tool is. That number is the tool's. How sure I should be is still mine to decide.
So the first thing I do is say out loud that I'm not sure. It sounds like a small thing, but it changes the conversation. People stop treating the answer as settled, and they start telling me what they know. If I stay quiet, the tool's 70 per cent turns into certainty the moment I repeat the date to someone else.
Then I ask what the answer is based on, and who would know whether that still holds. SittellaAI gave Hannah its basis: an arrival notice and the port's published clearance times. That was useful, because it told her where to look. The person who knew how clearance was going now was the customs broker, and Hannah could reach the broker by phone in minutes. The person who does the work every day often has the evidence I need, so I ask that person before I decide.
Next I ask what it costs to be wrong, and who pays. If the chairs had been late with nothing in place, the customer would have paid first. It would have had a new office to open and nowhere for people to sit. Hannah's company would have paid later, if the customer stopped ordering. Once I can see who pays, I can see who has to be told, and how early.
Then I split the decision. Some parts of it can be undone if I'm wrong, and some can't. I decide the parts that can be undone now, and I name a date for the parts that can't. Hannah booked a lorry she could cancel and held chairs she could put back in the showroom. The firm promise to the customer was the part she couldn't take back, so she gave it a date: 25 February. Waiting without a date is a slower way of not deciding.
Last, I tell the people who must know what has been decided and what hasn't. Hannah told the customer, and she told the sales lead. The sales lead is the person people forget. Suppose someone in sales had written to the customer "27 February, no problem". That one friendly message would have turned her "likely, not certain" into a firm promise.
None of this needs me to be sure whether the AI is right. I have a line for this: "Sounding right is a style, not a signal." A percentage works the same way. A tool that says it's 70 per cent sure still leaves me to decide what to do about the other 30 per cent.
SittellaAI did the fast part well. It read the arrival notice and the clearance history in seconds. Hannah did the part that was hers, which was to check the basis with the broker and decide what to commit to now. AI can speed up existing work. People still decide what needs to change.
One thing I won't claim: that this example shows how well anyone decides with AI. It is made up, and nothing on this page measures or scores a person.
I give keynotes, briefings and workshops on this.