Delegate the Preparation, Own the Call

This course follows one case from start to finish. A national chain offers your company a three-year contract worth nearly a fifth more volume than you handle today. Everyone at the leadership table speaks for their own function:

• Sales wants it signed before a competitor hears about it.
• Operations can only deliver with a second shift that nobody has hired yet.
• Finance knows what the margin looks like at the chain's price.
• The people who would do the work are halfway through a systems rollout you promised the board.

Then the facts change. The chain wants to start a month early, hiring slips, and sales and operations count capacity differently. The board chair wants your view by Friday.

There are three ways to go: sign, phase in or decline. That choice is yours. Everything else is preparation: the contract terms, the capacity data, what the second shift costs and where the rollout really stands. Preparation used to take the time of your best people, and your own. In this course you'll give it to Grok Bot, which you'll set up in the next few lessons.

Move the sliders to see why capacity is the hard part.

A decision is a choice that commits money, people or the company's word, and someone has to answer for it. Preparation is everything that informs that choice: gathering, checking, comparing and drafting.

To tell them apart, ask: if this turns out wrong, who answers for it? If the answer is you, or the person you've made accountable, it's a decision. If it's a finding someone could double-check against a source, it's preparation.

Here is the contract case sorted this way:

• Summarizing the chain's terms, with page references: preparation.
• Collecting capacity figures from sales and operations, and flagging that they define capacity differently: preparation.
• Estimating the second shift's cost, labelled as an estimate: preparation.
• Agreeing to the chain's earlier start date: decision.
• Telling the board chair your view: communication, which stays with you as well.

Preparation can be delegated. The decision, the approval and the communication with the people responsible stay with the leader. The whole course rests on that line.

"Delegate the preparation" needs one more level of detail before you can use it. Decision rights, meaning who may do what, come in three tiers:

1. Prepare. The Bot gathers, checks, summarizes and drafts. Every claim is tied to a source or labelled as an assumption. Example: a table of the chain's terms, with a citation for each line.
2. Propose. The Bot recommends an option, suggests a revised plan or drafts a follow-up, and it is always shown as a proposal, never as something done. Example: "Hiring has slipped six weeks. A phased start fits; a full start does not."
3. Approve. Only you. This covers sending, committing, buying, publishing, signing, hiring and allocating budget. Example: agreeing to the phased start and telling the chain.

Most trouble comes when one tier slides into the next: a proposal gets treated as approved, or a date the Bot inferred gets treated as a promise. Later in the course the approve tier becomes real Grok Bot settings: send tools switched off, and Ask first rules that stop the Bot until you say yes.

Here is a case about ownership. In Moffatt v. Air Canada (2024 BCCRT 149), the airline's website chatbot told a grieving passenger that a bereavement fare could be claimed after the trip. Air Canada's policy didn't allow that. The airline argued, in effect, that the chatbot was responsible for itself. British Columbia's Civil Resolution Tribunal called this a remarkable submission and held the airline answerable for everything on its website. The award was about CA$812.

Two cautions. It was a small-claims decision about negligent misrepresentation, so it carries limited weight as precedent. Treat it as a principle, not settled AI law, and take legal questions to your own counsel.

The principle still applies to you. Whatever your Bot tells a customer, a lender or the board, your company has said. That's why the approve tier covers every message that leaves the building, and why the Bots in this course draft and never send. A small sum was at stake in that case. The same mistake with a lender or a board could cost far more.

Two chief executives explain from different angles where the line falls.

Brian Moynihan, chairman and chief executive of Bank of America, says anything that takes real judgment needs a person in the loop, because in banking the answer has to be right and trusted. The bank kept its assistant, Erica, deliberately narrow so that it gave correct answers as it grew.

Tobi Lütke, founder and chief executive of Shopify, says he doesn't use AI for judgment. He uses it to set up the conditions for his own judgment. Machines can't take responsibility, he says. They inform the person who does, the way a well-set-up Bloomberg terminal informs a trader who still makes the call.

For the contract case, this means some outputs must be right before anyone relies on them: the capacity figure, the margin at the chain's price, and the rollout date in your reply to the board chair. You check those yourself or name the person who does. The Bot can prepare several positions, and you weigh them.

These are summaries of their interviews, not endorsements of this course or of Grok Bot.

AI writes fluently, and that creates a new risk. A recommendation with clean headings, tidy tables and confident wording reads like a decision that has already been made. Signing it can feel like the safe choice.

In IBM's 2026 CEO Study, 64% of chief executives said they are comfortable making major strategic decisions on AI-generated input. That's a self-reported survey figure. Comfort isn't the problem. Comfort without a check is how a company commits to a premise nobody verified.

In the contract case, suppose the pack says "capacity is sufficient for a phased start." Before you accept it, ask:

• Which capacity figure did it use: sales' or operations'?
• Is the second shift's cost a quote or an assumption?
• Does the plan depend on a hire that hasn't been approved?
• What would have to be true for this option to work?

If the pack can't answer these questions with sources, it isn't finished, however good it looks. Every Bot you build in this course will be told to tie each claim to a source, label each assumption, and stop when data is missing.

You'll build every Bot in this course around one real, recurring piece of your own work. Choose it now. Good candidates:

• a weekly operating review
• a pending resource decision, such as a contract, a hire or a budget shift
• a list of commitments you track across the leadership team

Write it down using the template below, splitting it into parts and marking each one. Here is an example for a weekly operating review: pull this week's numbers is prepare, flag metrics whose definitions clash is prepare, suggest three questions for the meeting is propose, send follow-ups to department heads is approve.

STANDING WORK: <e.g. Weekly operating review>
THE QUESTION IT ANSWERS: <e.g. Where are we off plan, and why?>
PARTS:
1. <part> | Tier: PREPARE / PROPOSE / APPROVE | Sources: <files, apps>
2. <part> | Tier: ...      | Sources: ...
3. <part> | Tier: ...      | Sources: ...
APPROVE ITEMS (become approval rules later): <list>
KEEP OFF ANY BOT FOR NOW: <board, acquisition or employee details>

Keep this note. Each APPROVE item will become an approval rule in Grok Bot.