Set Quarterly Goals and OKRs With AI: Clear Targets, Weekly Review

Set Quarterly Goals and OKRs With AI: Clear Targets, Weekly Review

To set quarterly goals and OKRs with AI, ask it to question every goal before it enters the quarter: what changes, for whom, measured how and from what starting point. Then use it to separate results from activities and to prepare a short weekly review built on data and questions. AI organizes and provokes, but the leader defines the goal, makes the cuts and changes course.

Key takeaways

• AI is more useful as a skeptical sparring partner that asks uncomfortable questions about vague goals than as a writer of polished OKRs.
• A key result describes a change in the business, the customer or the process. An activity describes what the team will do. AI helps tell the two apart.
• Without a baseline (the measured starting point), no key result is reliable. Mapping the current state comes before writing the target.
• A weekly review works when it is short, follows a fixed format and ends with a decision recorded by the leader.
• Sensitive data, evaluation of people and cutting decisions stay out of what you delegate to AI.

Why vague quarterly goals fail and where AI helps

Most problematic quarterly goals do not fail in execution. They fail in the writing. Phrases like improve customer service, grow digitally or organize sales sound like objectives, but in the last week of the quarter they do not let you say whether they were met. Each person on the team understands something different, and the review turns into a conversation about impressions.

The cause is human: whoever writes the goal knows the context and fills the gaps from memory. The team does not have that context. A language model that receives the goal with no context plays the reader who knows nothing, and that is exactly the reader a leader needs to test against. If the AI cannot say how it would know the goal was reached, the team probably cannot either.

The right use, then, is as a reviewer. You write the draft, the AI asks questions, you answer, and the final version comes from your answers, not from the machine's suggestion. That keeps the goal tied to what the business actually needs, not to what sounds good in a document.

How to ask AI to challenge a vague goal

The aim of this step is to generate questions, not answers. Paste in the draft objective and ask the AI to act as a skeptical colleague who does not know the company. Tell it explicitly not to rewrite anything until you have answered.

The questions most worth forcing the AI to ask:

• What is the current situation, and how was it measured? If nobody measured it, what is the simplest way to start measuring this week?
• What will be different for the customer, the team or the cash position if this goal is met?
• Who is responsible for this result? One person, not a department.
• How would you know, without arguing, that the goal was met in the last week of the quarter?
• What could improve the number without improving the business? (This question exposes metrics that are easy to game.)
• What is deliberately left out of this goal?

The last question is the most underrated. A quarter has limited room, and a goal with no stated exclusions tends to grow until it fills everything. Asking the AI to point out what the goal leaves out helps the leader decide consciously what will not be done.

One practical caution: AI tends to agree with whoever is writing. Instruct it to disagree, to point out ambiguities and to say when a question cannot be answered with the information provided. If it invents a reference number, discard it. The baseline comes from your systems, spreadsheets or reports, never from the model's imagination.

How to separate results from activities

The most common OKR mistake is writing activities as if they were key results. Launch the new page, train the team, implement the CRM and hold weekly meetings are work deliverables. They may be necessary, but they do not prove anything improved. A result is what changes because of them.

A simple test that AI applies well to a list of items is to ask three questions about each candidate key result:

• Can you complete this item and have the business stay the same? If so, it is an activity.
• Does it describe a work verb (implement, launch, create, train) or an observable change (reduce, increase, keep above, eliminate)?
• Would someone outside the team, looking only at the data, notice the difference?

Ask the AI to classify each item as a result or an activity and, for each activity, to propose the result it should cause. You choose what makes sense. Often the activity becomes a supporting initiative listed under the key result, and the result itself becomes what the team tracks every week.

Example of a rewrite: the activity implement a quote flow on WhatsApp can become the result reduce the time between a quote request and the first proposal sent. The implementation still happens, but it becomes a means, not the goal. If your team serves customers on WhatsApp, it is also worth reading about customer support KPIs with AI, which helps you choose what to measure without losing quality.

Step by step: building the quarter's OKRs with AI

The flow below fits into a few short sessions. It assumes the leader already has an idea of what they want to change this quarter.

• Step 1, map: write a few lines on how the process works today and which numbers you already have. If there is no number, note that as a gap, not a detail.
• Step 2, draft: write the objective in one sentence and list the candidate key results, even if imperfect.
• Step 3, question: ask the AI the questions from the previous section and answer in writing. Stop when your answers start repeating.
• Step 4, separate: ask for the classification between result and activity and rewrite whatever is an activity.
• Step 5, limit: reduce to a few key results per objective. More items mean less attention for each one.
• Step 6, set the owner and the source: for each key result, record who answers for it and where the number comes from (system, spreadsheet, report).
• Step 7, decide: the leader reviews, cuts and approves. The AI approves nothing.

The final quality test is practical: can a new person on the team read the OKR and explain in their own words what needs to change and how it will be measured?

Illustrative scenario: a distributor that wants better sales service

This is an illustrative scenario with no real numbers. A distributor of electrical supplies writes the objective: improve sales service. The owner pastes the sentence into the AI and asks for questions.

The AI asks what improve means, for whom, and how it would show up in the data. The owner realizes the real problem is quote requests that arrive on WhatsApp and take too long to get an answer. The owner also discovers that nobody measures that interval today, so the first job is to record response time for a few weeks to build the baseline.

The draft key results included items like train the salespeople and create a quick-reply table. The AI classifies both as activities. The owner rewrites:

• Key result: the time to the first proposal sent drops from the measured baseline, down to a limit the owner sets after seeing the data.
• Key result: requested quotes that go without any reply to the customer stop happening, or stay below a ceiling the owner defines.
• Supporting initiatives: salesperson training and the quick-reply table.

The AI also flags a gaming risk: replying quickly with an empty message would improve the first number without helping the customer. The owner adds a review of a sample of conversations each week, done by a person. What was left out on purpose: changing the pricing policy this quarter.

How to run a short weekly review

A good weekly review is short and ends with a decision. The AI's role is to prepare the material before the meeting, so the time goes to discussion, not to building slides. If you already have a planning ritual, the routine in weekly planning with AI for business owners fits well with this format.

An agenda template the AI can fill in with the data you provide:

• For each key result: this week's value, last week's value and the distance to the target, with the source of the number.
• Confidence from the result's owner: on track, at risk or off track, with one sentence of justification.
• Blockers: what is holding progress back and who can clear it.
• AI questions: points where the numbers and the owner's account do not match.
• Proposed decisions: continue, adjust the initiative, ask for help or stop. The AI suggests options, the leader chooses.

After the meeting, the leader records in a few lines what was decided and why. That record is what gives the cycle its value: at the end of the quarter, the retrospective rests on written decisions, not on memory.

As for the numbers, the AI should receive the data and only summarize it. When it calculates or interprets on its own, check a sample. A spreadsheet reading error that goes unnoticed can lead to a wrong decision that looks rigorous.

Do not let the review turn into surveillance. OKRs measure business results, not the behavior of each person. To see where that line sits, read how to measure AI productivity without surveillance.

Data, limits and what the leader does not delegate

Quarterly goals often involve sensitive information: margins, people's performance, customer data. Before pasting anything into an AI tool, define what may go in. Personal data of customers or employees calls for attention to Brazil's data protection law (LGPD), and it is worth consulting the guidance from the national data protection authority (ANPD) and setting a company policy for employee ChatGPT use. For OKRs, working with aggregate numbers and process descriptions, without names, is normally enough.

Some decisions stay with the leader no matter how good the AI's answer is: approving or cutting a goal, evaluating individual performance, deciding pay or dismissals, and changing strategy mid-quarter. The article on what not to delegate to AI in business details that boundary. The same logic appears in guidance on agent projects: keep a person in the loop at decision points, as described in Anthropic's Building effective agents, and handle risks explicitly, as the NIST AI Risk Management Framework proposes.

The working method we apply in AI projects fits here without effort. Map: describe how goal management works today and which numbers exist. Limit: decide which data the AI sees and which actions it can take, which in this case are only asking, classifying and summarizing. Pilot: use the flow for one quarter, with one team, recording what the AI got right and wrong. Transfer: hand the question script and the review template to the leader to use alone, then decide whether to expand, adjust or stop.

A goal the team cannot measure on its own is still an intention, not an OKR.

Frequently asked questions

Can AI write my company's OKRs? It can produce drafts, but the risk is getting generic goals that do not reflect your business. The safer use is to ask it to question your draft and classify items, and to leave the final wording and approval with the leader.

How many key results should I have per objective? Few, enough that the team remembers all of them without opening the document. If neither the AI nor the team can say which one matters most, there are probably too many items or mixed objectives. Cut before the quarter starts.

What if I have no data for the baseline? Start by measuring. Spend the first weeks of the quarter recording the number in a simple way, even in a spreadsheet, and only then set the numeric target. A goal set without a starting point tends to be too easy or impossible, and you only find out at the end.

Does the weekly review need a specific tool? No. A spreadsheet, a shared document or the management system you already use is enough. What matters is the fixed format, a clear source for each number and a record of the decision. Automate the preparation of the material only after the manual routine has worked for a few weeks.

Sources

• NIST: AI Risk Management Framework
• Anthropic: Building effective agents
• ANPD: Autoridade Nacional de Proteção de Dados