A decision meeting: leave with a choice, not another date
Prepare a decision meeting with a clear question, authority, alternatives and conditions. Record the outcome and check that tasks are accepted for execution.
Article language: English

In this article
At 10:00 everyone sits down to decide. By 10:47 no decision has been made, but another meeting is already in the calendar. This scenario is easier to interrupt when, before the discussion, people know what is being chosen, who can choose and what information could still change the result. Calling it a “decision meeting” does not give it those properties. A calendar is not a letter of appointment.
An effective meeting ends with one of three honest outcomes: a choice with conditions, rejection of a proposal, or a specific unknown that blocks the choice. In the last case, someone must own its resolution and a date for returning must be set. “We need to think about it some more” does not satisfy any of these categories.
First check whether a shared conversation is needed
Not every decision needs a meeting. When one person has authority, knows the alternatives and can gather comments in writing, a short note may suffice. A conversation makes sense when conflicting criteria need live resolution, dependencies need clarification or consequences for several teams must be agreed. The invitation should show that reason.
Separate an information update from a shared choice. A useful team status (in Polish) can show progress and blockers in writing; conversation starts where consultation or a decision is needed.
Consider a fictional case: a company wants to automate initial drafting of responses to sales enquiries. Sales wants a faster draft, customer service worries about wrong answers and the process manager has no team time available to check results. This is a good subject for a shared choice about pilot scope. A demonstration of every tool feature would be a different event.
If the problem has not yet been described, start with an automation process audit (in Polish). Choosing a tool before defining where the task ends may merely deliver work to the wrong queue faster. Management's presence alone will not replace missing analysis.
Turn the topic into a bounded question
“Sales automation” is a discussion area. “Do we run a trial drafting answers for one kind of enquiry, subject to salesperson approval?” is a decision question. It allows an alternative, decision-maker and consequence of refusal to be identified. The team can also see what it is not yet approving.
Add the choice's boundaries to the invitation: process scope, trial duration, available resources and exclusions. You need not know every number immediately. If time to check answers has not been estimated, name it as an unknown to investigate. Do not replace an empty field with “decide during the meeting” when feasibility depends on it.
Separate questions accidentally bundled together. Selecting a provider, approving a pilot and permitting a system to send messages independently are three different decisions. The first does not automatically include the other two. Prepare authority boundaries using the principles in responsible AI-agent workflows (in Polish).
Establish roles before opinions appear
The DACI method described by Atlassian separates driving the process, deciding, contributing expertise and being informed of the result. Distinguishing the right to present arguments from the right to make the final choice is especially useful. It is an organisational method that can be adapted to your governance.
If no one present has authority over the choice, the discussion can prepare a recommendation but cannot replace proper approval. In a small company one person may fill several roles. Record them separately so it is clear whether the facilitator collects arguments or also decides. In a larger organisation authority may follow an established approval path. A facilitator should not invent a new hierarchy when opening the presentation.
Invite people whose knowledge changes the assessment and those who must confirm they can deliver their part. Give others access to the result. Attendance alone does not prove broad consultation. Ten silent windows may mean ten busy people, not ten agreements.
Check absences beforehand. If only one person can confirm data availability and the pilot cannot start without it, obtain their position earlier. Do not cover its absence with a vote among those present. A democratic conclusion that a file probably exists still does not create the file.
Give people material to assess, not decipher
Send a short document with the question, recommendation, genuine alternatives and important evidence. A folder link is not yet decision material if nobody knows which of a dozen files contains current findings. A brief prepared by your team, with attachments for people needing detail, can help. The guide to a decision brief for management (in Polish) explains how to prepare one.
GitLab's meeting handbook describes one shared agenda attached to the invitation, with context, expected outcome and room for questions before the conversation. This is a practice of a particular remote organisation. Set your own material-sharing deadline according to decision difficulty and time needed to check arguments.
In the fictional pilot the material might compare three options: retain current work, try one limited enquiry category, or expand automation after further preparation. For each describe expected benefit, checking and maintenance effort (in Polish), and the largest unknown. Do not describe the current process as cost-free: it consumes time too.
Separate an observation from a prediction. “In the sample we found an incorrect product assignment” is a finding. “After changing the instruction the error will disappear” is a hypothesis. If material mixes the two, an AI pilot assessment plan (in Polish) can help. Participants should know what has been checked and what remains to be confirmed.

Lead through differences, not slides
At the start read the decision question and confirm that authority and scope have not changed since the invitation. Then ask about errors in the material. Resolve those before discussing differences in assessment. Presenting the whole document word for word consumes time intended for the choice.
A short discussion might follow this order:
- Confirm facts and identify gaps important to the choice.
- Compare options against agreed criteria.
- Name major reservations and how to handle them.
- Have the authorised person announce the decision and conditions.
- Record action owners and how the result will be checked.
Do not treat this order as a script with an inflexible minute count. If a new constraint rules out every option, return to the question. If someone starts designing a future application screen, decide whether that detail affects today's choice. If not, move it to a separate task with an owner.
Ask participants to write a short assessment before discussion. This makes it easier to see an argument from someone who needs a moment to formulate it. In disagreement ask: “Which premise would have to change for you to choose differently?” That is more useful than repeating a vote after every comment. If some people are in the room and others remote, check participation in a hybrid meeting (in Polish). A remote participant should see the options and be able to comment on the shared material.
Distinguish a reservation, a condition and a blocker
Not every “but” has the same weight. A reservation describes a cost or inconvenience that can be consciously accepted. A condition is something that must be met before a decision is carried out. A blocker means a choice cannot be treated as ready for execution in the given scope without another decision or piece of information.
In our example, several extra checks may be a reservation. Preparing an approved test set may be a condition. A lack of someone authorised to confirm rules for using the material may block a trial on that material. This is a working organisational distinction; establish actual requirements with the appropriate people in the company.
If conditions multiply until nobody can assess their effort, do not announce success “with minor comments”. Return to priorities with a limited team (in Polish). Agreement on the goal does not automatically create time to prepare every dependency. A condition without an owner usually becomes a surprise on launch day.
Record the decision so it survives leaving the room
The meeting note should contain the choice, a short reason, scope, conditions, decision-maker and implementation tasks. Add rejected alternatives and why they were rejected. You do not need a transcript. Keep a record that lets someone distinguish a discussion idea from approved action a month later.
Move the result into your own project decision register and tasks to the place implementers use. The guide to a project decision register (in Polish) explains how to maintain one. Minutes should not become a second, hidden work-assignment system. If a decision hands responsibility to another team, agree the project handoff conditions (in Polish). “We informed them” and “the team accepted the task” describe different events.
A circulated record allows obvious misunderstandings to be corrected. Set a short, reasonable deadline to report errors in the note, not to reopen the whole choice without a new premise. If a genuinely new premise appears, add the change and date. Decision history should explain how the situation developed, not pretend to be infallible.
Example: three honest endings to the same discussion
Return to the sales-response assistant. Suppose the team has prepared material for one product group and can check results, but does not yet know the workload from a wider scope. The decision could be: run a limited trial, a named person approves every answer, and expansion requires a separate choice after assessment. This permits testing a specific possibility with a clear boundary.
A second outcome arises if review reveals outdated material and nobody able to approve it. A sensible ending is then to refuse a trial on the current set and assign preparation of one area. That task must return to resource planning. Do not call this a pilot launch just because an inauguration was already in the calendar. The event's name does not establish process readiness.
A third possibility is to defer the choice pending one specific check: does an export from the source system preserve current offer variants? The note names the person running the test, the material and the date for presenting its result. The team may then decide in writing if that route was agreed. There is no need to recreate the whole meeting when only one closed question remains. A condition should not conceal new, still unidentified dependencies.
For each ending, the recipient should understand what to do the next day. In the first they prepare the trial within approved scope. In the second they organise the material, without a promise of automatic launch after corrections. In the third they test the blocking unknown. This difference is a practical check on decision quality: if all three notes trigger identical tasks, the scope of approval was probably still too broad.
Also record what goes to absent people. Salespeople need to know whether they may use the solution, the process owner which checks apply, and the sponsor when the material for the next choice will arrive. Not everyone needs the full discussion history. Everyone should find the consequence relevant to them and where to check current status. Thus the meeting has one outcome even if messages vary in detail.
Check one meeting from invitation to execution
Choose the next decision about a process change. Ask the organiser and future implementer independently to write what is to be chosen, who has authority and what document will confirm the result. Compare differences before sending the agenda. This short exercise reveals whether the meeting has a shared goal or merely a shared time.
Afterward check one task arising from the decision: has its owner accepted it, do they know the conditions, and do they have a deadline and necessary resources? Judge the meeting by that transition, not only by the number of recorded conclusions. Ten tasks can be a good outcome or ten ways to postpone a choice.
If AI and automation decisions regularly end with unclear responsibility, the Aeneia Audit scope can help identify a starting point. Prepare an invitation, a note and an example task that stalled after a decision. That material makes it easier to reconstruct the role of tools, sources and human approval.










