How to set project priorities when the team is stretched
Plan priorities around available people, dependencies and work that can be accepted.
Article language: English

In this article
Five projects received the highest priority. The team received the same Mondays it had before. When the list of initiatives grows faster than the capacity to deliver them, another “urgent” label will not solve the problem. A choice is needed: what to start, what to finish and what must consciously wait for a resource that is unavailable today.
Set priorities against the intended result, necessary skills and availability of specific people. A ranking alone is not a work plan. The first project on the list may need a specialist occupied with maintaining a system, while the third can be finished without that person. See the difference before promising dates.
Make one list of commitments, including work without a project name
Imagine a fictional company that wants to launch an AI assistant for salespeople, improve its knowledge base and change how it collects enquiries at the same time. All three ideas rely on the same person who knows the offer. The plan has three coloured rectangles; in practice there is one queue for the expert.
Include routine work, checking results, corrections after earlier releases and agreed tasks for other teams. Work without an impressive name still takes time. Leaving it out of the plan does not free capacity; it hides a commitment.
If you do not know how many initiatives are active, start with your own inventory of AI tools and a prioritised register of uses. The guide to inventorying AI tools (in Polish) and the article on an AI use register (in Polish) explain how to prepare them. The tools list shows what the company has; the use register shows what it is trying to do with it. Planning also needs people, tasks and dates.
Calculate time available for the kind of work
A nominal working week is not a week available for a new project. Subtract known absences and recurring duties, then consider a realistic amount of work requiring concentration. Do not assign everyone the same “productivity” percentage. A service team, an analyst and the person approving an offer work under different conditions. If recurring activities are unclear, use transparent work sampling to understand their share of team time. The guide to work sampling (in Polish) explains the method and its limits.
Azure Boards capacity planning distinguishes estimated requirements from time available for tasks, accounting for different working hours, leave and holidays. It describes sprint planning in a particular tool. A useful general principle is to update availability for the period being considered instead of endlessly copying an old plan.
In our example, the offer expert may be able to review one set of material but not conduct three consultation series in parallel. Record the arrangement of time as well as its total: when the person can enter a project, how long a block is needed and what must arrive first. Two scattered hours do not always replace a joint acceptance session.
Availability to approve one decision and availability to perform analysis are different needs; do not add them as interchangeable hours. Name waiting time separately from active work. A team may spend a day analysing and then wait several days for confirmation. The stage deadline includes both, but the whole waiting period should not artificially increase the workload.
Break projects into the smallest meaningful results
Instead of comparing large labels such as “knowledge base” and “AI in sales”, name stages that can be accepted. One may be an organised set of answers for one category of enquiry; another may be a trial of generating drafts from approved material. Each result needs a recipient, an evaluation criterion and a scope boundary.
This exposes dependencies: before assessing the assistant's answers, you need to know which version of the offer is current. The guide to organising company knowledge (in Polish) can help. It does not mean the entire archive must be organised before every test. A limited, checked collection matching the test scope may suffice.
For each stage record what must be ready as input, who will work and who will accept it. If the recipient has no time to review, finishing execution will not clear the queue. The project merely moves into a waiting room labelled “almost ready”. Such rooms can be surprisingly large.
Do not split stages into individual clicks. The division should help resource decisions and allow a useful result to be checked sooner. If a task brings no independent result, show which next step completes the useful whole.
Choose criteria before seeing the ranking
Agree common questions: does the stage solve an identified problem, what result could it bring, how urgent is the need and can the company maintain the result? Add dependencies and reversibility. Scores may structure discussion, but should not pretend to be measurement when numbers are team judgements.
In the fictional example, organising selected material may come first because it unblocks both the assistant and salespeople. A form change may be quick but will not resolve incorrect answers. A broader assistant may promise more benefit while requiring oversight the team has not provided.
The guide to ordering website, automation and expansion work (in Polish) helps with such dependencies. Do not assume infrastructure always beats an urgent customer need. If a small scope safely tests an important assumption, it may supply information needed to choose the next work.
Also decide who resolves conflicts between teams. If every manager can add their own “highest” item, the shared list is only a collection of wishes. A meeting with a clear decision mandate or an agreed written approval route is needed. The guide to decision meetings (in Polish) also considers when a written decision may suffice. The ranking should end in a choice of work for a period.
Limit starts to reveal the bottleneck
Atlassian describes work-in-progress limits in Kanban, which restrict tasks in process states and help reveal bottlenecks. WIP here means work started but not finished. Adapt the limit to the team's work rather than copying a number from a sample board.

A service team might start with visible states: preparation, execution, awaiting review and completion. Agree how much can actually be handled in each. When too much material waits for evaluation, investigate unblocking the review before starting more tasks.
A limit should not be used to hurry people or hide work under another name. A task started and then “temporarily put aside” still needs attention and a return to context. If it vanishes from the board, the list looks better than reality. That is a poor trade even if the presentation gains breathing room.
Look also at incoming and completed everyday work. The guide to too much work in progress (in Polish) explains how to make the queue visible. Choosing important projects does not remove the queue already in progress. You need to see what is waiting, why and for how long.
If a limit is continually exceeded, investigate why. Acceptance criteria may be unclear, an expert unavailable or information from the customer missing. Automatically raising the limit stops it limiting anything; keeping a poorly chosen value without discussion is no help either. Treat it as a work-design hypothesis and assess it through observed flow.
Reserve room for surprises explicitly
Teams supporting live solutions can face matters requiring a response. Use your own observations of their type and frequency rather than a universal buffer from a guide. Without data, record the uncertainty and observe an initial period with fewer new commitments.
Name what can interrupt the plan: an outage, an error in customer material or an urgent change to an approved offer. Specify who can decide to interrupt and which task moves. The exception should visibly change the plan; otherwise it becomes an indefinite addition to an already full week.
In automation, remember the checking, correction and maintenance costs discussed in the article on the total cost of an AI process (in Polish). A tool may shorten one action while increasing the burden on the person receiving its output. Resource planning should cover the entire path of the task, not just the most impressive demonstration stage.
Give a project permission to wait or end
For topics that do not fit the period, state why and the condition for return: a skill becoming available, a dependency finishing or a need being confirmed. “Later” without a condition creates a warehouse of hope. After a while no one knows whether an idea is still current or merely comfortable on the list.
If evidence challenges the goal, consider project stopping criteria. The guide to stopping a project (in Polish) distinguishes pausing, changing scope and closing. A pause for lack of capacity and closure of an ineffective direction have different consequences. Both require communication with people who planned their own work around the project.
Preserve priorities and their reasons in a project decision register. The guide to project decision records (in Polish) explains how to retain the reasoning and conditions for review. When someone asks weeks later why another initiative did not start, the answer should identify the constraint and choice. “There was no time” is true too often to explain a particular decision.
Compare two plans using the same resources
In the fictional team all three initiatives need the offer expert to approve material. The first plan starts them at once: the knowledge owner collects documents, the assistant builder prepares instructions and marketing redesigns the form questions. At first everyone has work. Then all three streams request decisions from the same expert. Without a shared order, each person may believe their own item is merely waiting for a small signature.
The second plan first agrees one material scope useful to all three efforts. The expert reserves time to review it and the others prepare inputs. Then the team chooses the stage that best tests the most important unknown. The other ideas have a start condition. They are not rejected, but neither receive a fictional start date just so every sponsor sees movement.
You cannot conclude in advance that the second arrangement is always better. The form may not need the shared material and might be accepted independently. The comparison should therefore reveal real dependencies, not create a mandatory queue for everything. The goal is to reduce resource conflict, not require the whole company to work on one topic at a time. Independent work can proceed in parallel if its acceptance is available.
Add the cost of changing the plan to both options. Moving the expert to an urgent case may stop the review without stopping all material preparation. Identify what can continue and what loses its purpose without that resource. This helps the team respond without resetting priorities every day and exposes when an apparently small change shifts several dependent stages.
After choosing, tell sponsors which results they will receive in the period and what is outside it. A numbered ranking alone does not say whether work can start. The plan should show its entry conditions: dependencies met, resource available and acceptance agreed. If one is missing, name the preparatory state instead of presenting the topic as started. That small wording change exposes the difference between intention and commitment.
Test the plan against one real week
Take the next three stages and map the involvement of everyone needed, including approvers. Add routine duties, dependencies and known absences. Then ask each person whether the arrangement is feasible and what they would have to set aside. The difference between the spreadsheet and the team's answer is material for improving the plan.
After the period check what finished, where work waited and what unplanned tasks entered. Do not draw a universal norm from one week. Look for a specific correction: fewer parallel stages, an earlier review or a shorter acceptance scope. The next plan should use this knowledge.
If several AI initiatives compete for the same people and their order is hard to decide, see Aeneia Audit. Prepare a list of work in progress, people's availability and an example of a stage awaiting review. Priority then becomes a decision about execution.








