Chapter 2

Understand how to work effectively

Define the desired state, assess evidence and choose the next action

Late at night, a salesperson looks pleased with the documents he has produced. The customer still replies, “We are reviewing it internally,” leaving him wondering why finishing all the work has changed nothing.

This is the first of the five stages from Chapter 1. AI barely appears yet. Begin by learning how to work effectively.

Meetings, documents, reports and plans look like separate activities. Every kind of work, however, is an attempt to make progress toward a Goal.

What is work?

Work can look complicated. Sales, hiring, accounting and planning contain hundreds of different activities. Examine their structure and they have the same elements.

Every job has a state someone wants to achieve.

• In sales: the customer has decided to buy and signed a contract.
• In hiring: the people you need have joined and started contributing.
• In accounting: invoices and payments are complete by the end of the month.

That is the desired state. Each example describes a result of the kind discussed in Chapter 1.

There is also a current state: what is actually happening. The customer is still undecided. Too few people have applied. Only half the invoices have been sent.

If the current state already matched the desired state, there would be nothing to do. Work exists because the two states differ. A result occurs when the work resolves that difference and reaches the desired state.

Sometimes one action is enough. If an email resolves the difference, the work is complete.

Most work, such as winning a contract or hiring someone, takes more than one action. Use the following sequence:

1. Compare the desired and current states to identify the main constraint: the most important obstacle to progress.
2. Choose a ToDo that can resolve that constraint.
3. Complete the ToDo.
4. Use the result to update the current state. The remaining difference has now changed.
5. Identify the main constraint in the new situation.

Repeat steps 2 through 5. Eventually the current state matches the desired state, and the Goal is complete. We call the sequence the Goal achievement loop.

The main constraint is not always obvious. Take an action that helps reveal the difference, or choose a likely constraint as a hypothesis and proceed. Acting often tells you more than planning alone.

There is a second loop. Sometimes the difference is so large that you cannot see what to do. Sometimes external circumstances change and the objective itself is no longer appropriate. Revise the desired state when that happens. This is the Goal update loop.

The solid blue lines show the Goal achievement loop. The dashed yellow lines show the Goal update loop. Together they form the Goal Loop: a way to keep working toward a Goal while also reassessing the Goal itself.

The achievement loop may sound familiar as PDCA. PDCA describes work toward a fixed objective. The Goal Loop includes a process for changing the objective, which is often the harder part of work.

A job can contain several Goals. “Win the proposal” includes “Understand the customer,” “Identify the problem” and “Prepare the proposal.” Each has a desired state, a current state and a difference to resolve.

Larger jobs contain smaller jobs, which can contain smaller jobs again. The structure stays the same at every scale: identify the desired state, understand the current state, and act to reduce the difference. A company-wide objective and an email sent today follow the same process.

Effective people are good at running the Goal achievement loop. They keep the Goal clear, establish the current state from evidence, and act on the most important constraint. They repeat the process quickly and produce results.

AI now makes it possible for anyone to run the Goal Loop more effectively.

The practical work of adopting AI is to make this process executable by AI. A person, an AI, or several people and AI agents can run it.

AI already runs similar loops in familiar settings. Social media recommendations aim to keep users watching; actual time spent is the current state. The system adjusts recommendations, keeps changes that increase viewing time, and reverses changes that reduce it. Algorithms were doing this long before ChatGPT appeared. Your work can use the same structure. AI needs access to the desired state, the Goal, and the current state, the context.

Goals, child Goals and ToDos

Use three names for the parts of the work:

Apply the Goal Loop to a real sales proposal

You have seen the structure. Next, follow one job through successive decisions.

The salesperson from the opening example is trying to sell corporate training. He sent a proposal, but the customer has said “We are reviewing it internally” for two weeks. Read the example as the person responsible for the sale.

The salesperson starts adding case studies to the proposal, then stops to ask whether the extra work will help. The customer has not made a decision.

His planned ToDo was “Improve the proposal design and add ten pages of case studies.” Which Goal and which constraint produced that ToDo? He could not answer. Starting from a ToDo can disconnect the activity from the result it needs to produce.

We will reassess the stalled proposal using the Goal Loop, from the beginning until progress resumes. By the end, you should be able to explain each decision in a Goal Loop for one job. Keep one of your own current projects in mind as you read.

Step 1: Turn the result you want into a Goal

Ask the person doing the work, “What is the Goal?” and a common answer is:

“Finish the proposal and send it to the customer.”

The first mistake is treating completion of a deliverable as the Goal. If the proposal is finished but the customer has decided nothing, the situation has not changed. In Chapter 1's terms, the document is a deliverable, separate from a result.

Where should the Goal come from? It should follow the company's objectives, rather than an individual's assumption.

Finishing the document is one possible means of achieving the Goal.

Defining the connection is management's responsibility. Employees who guess the higher-level objective may aim at something different from the company's actual objective. Leadership and the people responsible must define the higher-level Goals and connect the work to them. Without that, an instruction to “use AI” will merely make existing ToDos faster. That is what happened to the company in Chapter 0.

Check whether the connection can be explained in one sentence. Ask, “Why are we doing this work?” An immediate answer such as “To achieve our order target” shows that the connection is clear.

Write a Definition of Done for the Goal

Alongside the Goal, describe what must be true for it to count as achieved. This is the Definition of Done, or DoD.

• ✕ Win Acme's corporate training contract. People may interpret “win” differently.
• ○ The decision-maker has the expected benefits, delivery arrangements and costs needed to decide whether to proceed, and has made that decision.

A useful check is whether someone other than you would reach the same conclusion about completion. “Try hard” and “Do it properly” do not provide a condition someone can verify.

A Definition of Done also makes the next action easier to identify. Compare the current state with the completion criteria: the missing elements become visible. In Chapter 3, the same clarity will affect how well AI can carry out delegated work.

Step 2: Establish the current state with evidence

Next, assess the situation against the Goal of winning the contract. The salesperson used to report it like this:

“How is the Acme project going?”
“It is progressing. The proposal is almost finished, so we are just waiting for their reply.”

The exchange tells you how complete the document is. It tells you nothing about whether a purchase decision is closer. “Going well” and “Not done yet” are similarly impressions, rather than evidence.

Use evidence to describe the current state. Put it alongside the desired state.

• Desired state, the Definition of Done: The decision-maker has the expected benefits, delivery arrangements and costs needed to decide whether to proceed, and has made that decision.
• Current state, based on evidence:
  • The customer has received the training content and pricing. The sent email confirms delivery.
  • The expected benefit is described only as “higher productivity.” There is no basis for calculating return on investment, or ROI.
  • The internal arrangements needed to start in three weeks have not been confirmed.
  • The reason the decision-maker is holding the proposal is unknown.

Comparing the desired and current states using evidence reveals what is missing. Four areas to check are now visible.

Step 3: Choose the most important constraint

Should you address every missing element immediately?

“I will cover everything” is a common response from conscientious people. For two weeks, the salesperson had done just that: an ROI document, a staffing table, and answers to security questions. Today he planned ten extra case-study pages. The decision-maker was still “reviewing it internally.”

Choose the most important reason the desired state has not been reached. One question is enough:

“Which issue is most critical to the Goal right now?”

Compare three ways to choose:

• ✕ Start with case studies because they seem easiest.
• ✕ Start with the design because it is most visible.
• ○ Identify the main reason the customer has not decided to buy.

The strongest candidate was not knowing the decision-maker's real reason for waiting. Until that was understood, producing more material was guesswork.

The choice also determines what you will leave for another day. There is no evidence that ten more case-study pages would resolve the delayed decision, so they are unnecessary today. Constraints are not a fixed priority list. They change as results change the current state. Choose the most important one for now.

Step 4: Turn the constraint into one action with an observable result

“Find out why the decision-maker is waiting” can lead to an executable ToDo.

If one action returns a result, it is a ToDo. If it requires several lines of work, create a child Goal and break that into ToDos.

• “Read the last two sales meeting records and extract the decision-maker's criteria and reasons for waiting.” One action returns a result: a ToDo.
• “Confirm delivery arrangements for a start in three weeks.” Instructor availability, effort required and the customer's staffing all need attention: a child Goal, with ToDos inside it.

The arrangement of Goals, child Goals and ToDos is called a Goal Tree. It represents the same relationship described earlier: larger jobs contain smaller jobs.

Do not plan the Goal Tree far beyond the next result you can get from reality. It is tempting to specify every future step. Stop where an action will return useful evidence, because the next correct action may change when the result arrives.

One attempt followed by observation and adjustment can be faster and more accurate than an elaborate plan based on assumptions. Apply that principle to work: take a useful action sooner instead of spending time on plans you may never use.

Step 5: Update the current state from the result and repeat

Finishing a ToDo is followed by another responsibility: update the Goal's current state with the result.

Then return to Step 2. Compare the new state with the desired state, choose the main constraint again, and define the next ToDo. Continue until the current state matches the desired state.

At each review, there are four possible decisions: complete the Goal when evidence satisfies the DoD; continue toward the same Goal with the next constraint; investigate when information is insufficient; or hand over when someone else needs to act. Delegating to AI in Chapter 3 is an extension of that last option.

The salesperson extracted the decision-maker's statements from the last two meeting records. They asked, “Can we really start in three weeks?” and “Who will we need from our team, and for how much time?” Delivery arrangements were holding up the decision, rather than document design or ROI.

The evidence changed the current state, so it changed the next action. He checked the availability schedule, prepared an answer on delivery arrangements and sent it. Two actions restarted a project that had been stalled for two weeks.

Evidence changed the decision about what to do next.

Follow the achievement loop through to completion

Continue the same proposal until the current state matches the desired state.

• Goal: Win Acme's corporate training contract.
• Desired state, the Definition of Done: The decision-maker has the expected benefits, delivery arrangements and costs needed to decide whether to proceed, and has made that decision.

First iteration

Second iteration

Third iteration

Fourth iteration

The current state now matches the desired state. The Goal is complete.

Across four iterations, the only new document the salesperson produced was the final one-page summary. The other actions were reading meeting records, sending answers and revising the terms. Those actions completed a sale that two weeks of document production had failed to advance.

The salesperson looks at the one-page summary, the only new document he needed to create. Four iterations and four actions have led to the customer's decision.

Review

• Define work Goals from higher-level Goals, rather than from a deliverable or an assumption.
• Describe the current state using evidence. Compare it with the desired state to identify the difference.
• Ask “Which issue is most critical to the Goal right now?” to choose the main constraint.
• Break the work into a ToDo that returns a result from one action.
• Update the current state with the result. The next action may change.

A person ran the entire process in this chapter. The next chapter gives parts of the same loop to AI. You do not need to learn a new structure: delegate part of the Goal Loop you already understand.

• Chapter 3: Delegate work to AI