
A project plan can have plausible dates, tidy colours, a structure that looks good in a meeting, and still not survive the first surprise. Not because whoever built it is incompetent, but because in Microsoft Project bad decisions do not produce errors: they produce a plan that seems to work. No message, no warning, no red row. The bill arrives months later, when something slips and the plan does not react.
This is not the usual list of Microsoft Project best practices. Lists say “don’t use constraints”, and nobody follows them, because a rule without a mechanism convinces no one. Here we come in from the opposite side: from the symptom. We start from something you can see, work back to what produced it, and only at the end comes the rule, which by then hardly needs stating. For each of the nineteen symptoms: what you are looking at, why it happens, how to avoid it if the plan is yours, and how to spot it in thirty seconds if the plan was sent to you. It applies to every desktop version, from Project 2010 to Project 2024 and Microsoft 365.
In short
- In Microsoft Project dates are not typed: they are calculated. Every date typed by hand becomes a constraint: the task can no longer move earlier, and if it has no links no delay moves it. The plan becomes a drawing instead of a model.
- A plan is inspected in a precise order, because the checks depend on each other: if the dates are typed there is no point looking at the critical path, and if there is no baseline there is no point looking at the variances.
- Summary tasks are containers, not work: they are not linked to each other and they get no resources. Doing either doubles the budget without a single row looking wrong.
- Without a saved baseline and an explicit status date a plan cannot be measured: any progress figure is an opinion.
- Whoever receives a plan needs different checks from whoever builds it: for each symptom below there is the move that reveals it in thirty seconds.
Part 0: the order in which you look at a plan
The question I am asked most is not “what are the best practices”: it is “where do I start looking?”. And the answer matters more than the list, because the checks are not independent. Each one makes sense only if the ones before it gave an acceptable result: reading the critical path of a plan full of constraints is wasted time, not because the critical path is unimportant, but because what you read is not the critical path of the project, it is the drawing of the constraints somebody put in.
This is the chain, in order. The first three checks tell you whether you can trust the plan; from the fourth on they tell you what it is worth.
- Is it a plan or a list? Structure, levels, milestones. Without a skeleton, everything else has nothing to stand on.
- Are the dates calculated or typed? The check worth all the others put together, and it conditions the next two.
- Does the network hold? Links, dangling tasks, links on summaries. A broken network produces a plan that does not pass on delays.
- Is the critical path telling the truth? It only makes sense after 2 and 3.
- Who works, and when? Resources, workloads, calendar. Here you find out whether the dates can be met by real people.
- Is there a saved baseline? Without one, the plan can never be measured.
- Is the plan being run, or is it stale? Status date and actuals: they say whether what you have in front of you describes today or the day it was born.
Ten minutes are enough for the first three. Half an hour for all seven, on a plan of a few hundred tasks. If you are in a hurry and must choose one, choose the second. A shorter route with the commands is in five checks for a plan you did not build.

Part A: structure
1. The project summary was built by hand
At the top of the plan there is a row named after the project that contains everything else. Somebody wrote it. It was not needed: Project already has one, the row with ID 0, the project summary task, which you switch on with a checkbox and which summarises the duration, work and cost of the whole project by itself, always up to date, with no upkeep.
A hand-made row, instead, is a summary task like any other: it can be moved, it can be linked, and above all someone can assign resources to it. It becomes a container pretending to be the project.
If the plan is yours: switch on row 0 (Gantt Chart Format tab, Project Summary Task checkbox) and remove the hand-made one. It is also the row whose notes are the natural place for information that must travel with the file.
If the plan was sent to you: check whether the first row has ID 0 or 1. If it is 1, bold and containing everything, it was built by hand: check straight away whether it has resources assigned, because that is where the budget swells the most.
2. Task names do not start with a verb
“Electrical system”. “Documentation”. “Supplier Smith”. These are names of things, not of work, and in a plan they raise a question nobody can answer: what has to be done? A task should state an action, Lay cable ducts, Write the manual, and milestones, which are checks, should state a check passed: Acceptance test passed, Design approved.
The twin symptom, more modern and more frequent, is information stuffed into the title: “Lay cable ducts [BLOCKED – ask Bianchi]”. Everything in the name is text, and on text you cannot filter, group or total.
If the plan is yours: a convention decided at the start and written down somewhere is worth more than any correction made afterwards: a verb for tasks, a check passed for milestones, statuses and owners in dedicated columns.
If the plan was sent to you: sort by task name and scroll: bare nouns and square brackets jump out at once. If there are many, expect the project’s status information to live nowhere but in the head of whoever wrote it.
3. The work breakdown does not hold
Four hundred rows at the same level, or phases that contain a single task, or a branch broken into five levels next to a single row that says “Commissioning, 40 days”. They are all signs of the same thing: the plan was written as a list and organised afterwards, instead of being broken down first.
The cost is not cosmetic. Without real summaries there are no intermediate totals: you cannot say what a phase is worth, where it stands, or show one on its own in a meeting.
If the plan is yours: three or four levels, at least two children per container, tasks of one to ten days, a milestone that closes every phase. How to build the tree is in work breakdown structure: how to build one that survives the project.
If the plan was sent to you: collapse everything to the first outline level. What stays on screen is the idea the planner has of the project: if you take it in at a glance and recognise it, the structure holds. The five signs of a weak WBS are told one by one in a separate article.
Part B: time
4. The dates were typed by hand
The symptom worth all the others. In Microsoft Project dates are not typed: they are calculated. You give durations and links, the engine says when things happen. When somebody types a start date, Project does not complain: it does the only thing it can, it sets a constraint, of the “Start No Earlier Than” kind. From that moment the task can no longer move earlier, and it slips later only if something upstream pushes it: if it has no links, nothing pushes it. We checked it: dragging a bar does the same, with no warning and no indicator.
The result is identical to a healthy plan as long as nothing happens. At the first change the difference is total: the plan built on links reorganises itself and shows you the new finish date, the plan built on typed dates stays still and keeps showing you dates that will not happen. You have a drawing, not a model. The rigid constraints, Must Start On and Must Finish On, do worse: they keep the task on its date even when the predecessor is late. Constraints also damage the calculation: they break the critical path, because they create local deadlines against which slack is recalculated.
There is a case where the constraint is right: the contractual date, the imposed delivery, the plant shutdown window. A constraint set on purpose, knowing why, is a tool. In a healthy plan you can count them on one hand.
If the plan is yours: durations and links, never dates. When you need to mark a date without constraining the calculation, there is the Deadline field, which warns without nailing: it shows an arrow on the Gantt and flags the task if the plan misses it.
If the plan was sent to you: insert the Constraint Type column and sort. The healthy value is “As Soon As Possible”; the values that should make your ears prick up are “Must Start On” and “Must Finish On”. If more than one task in ten has a hard constraint, that plan is not scheduled: it is drawn.
Before you even count the constraints, there is a quicker test that works on a schedule built in Excel too: add one week to a task halfway through and see whether the finish moves. It is the one-week test, walked through step by step in a separate article.
5. Tasks are manually scheduled
It is the modern way of typing dates by hand, and it shows less than a constraint because it leaves no icon in the Indicators column. A manually scheduled task stays where you put it: it does not move when its predecessor changes, it does not take part in the recalculation like the others. In a plan of a hundred tasks where twenty are manual, the engine works on eighty and the other twenty are decoration.
The feature exists for a good reason: at the start, when all you know about a task is that it “takes about three weeks, around May”, manual scheduling even lets you write text instead of a date. It is a note, and like all notes it must be converted before the plan is delivered.
If the plan is yours: use it for the rough draft and switch everything to automatic before saving the baseline. Watch the default for new projects too, which on some installations is manual.
If the plan was sent to you: insert the Task Mode column: it says for each row whether it is automatic or manual. If you find many, redo the first three checks of Part 0 knowing that part of the plan is not calculated.
6. Durations are missing, or all the same
The question mark next to a duration means estimated: it is the value Project sets by itself when you create a task, “1 day?”, and that nobody has replaced yet. A plan with many of them was not estimated: it was listed.
The worse variant is the plan where every duration is identical, five days everywhere, or ten, which is the elegant way of saying nobody looked at them one by one. And there is the opposite case, the sixty-day block: a task that long cannot be followed, and for two months every answer about its progress will be an invented percentage.
If the plan is yours: before you consider planning closed, filter the tasks with estimated durations and empty the list. The estimating techniques are in project estimating methods; here the rule of form is enough: an estimate must be explainable out loud in one sentence.
If the plan was sent to you: count the question marks and look at how durations are distributed. If they are all the same or all round, there is only one question to ask: “what are they based on?”.
7. Links sit on summary tasks
Linking two phases together feels like the most natural thing in the world: it is what you do on a whiteboard. In Project it is a shortcut that makes the plan opaque, because the phase is a container: saying that phase B follows phase A does not say which task of B waits for which task of A. Delay propagates in blocks, the critical path runs through containers instead of real work, and when something slips you cannot tell what pushed what.
Worse: it is easy to create circular dependencies between a summary and its children, which Project rejects with messages that look incomprehensible.
If the plan is yours: links go between detail tasks. The clean way to tie two phases is to link the milestone that closes one phase to the first task of the next: it is readable, and it is also a checkpoint.
If the plan was sent to you: show the Predecessors column and look only at the bold rows. If they have any, the plan is tied in blocks.
8. There are dangling tasks
A plan is a network: each task should know what it depends on and what it starts. In practice you always find some that are disconnected, in two kinds.
- No predecessors: they are free to start on the first day of the project, and that is almost never true. Apart from the first real task, each one is a question to ask.
- No successors: more insidious, because they make no noise. They finish and start nothing: delaying them has no effect on the plan, which in reality is almost always false. Apart from the last, each one is a missing piece of the chain.
It is the first point of every schedule quality checklist, and rightly so: an open network does not pass on delays, so the plan stays optimistic even when the project is not.
If the plan is yours: before delivering, show predecessors and successors in the table and sort by each: the list of dangling tasks reads in a minute.
If the plan was sent to you: same move, and count. A handful on a large plan is normal; dozens say the network was not built but drawn in pieces.
9. Nobody ever looked at the critical path
The critical path is the chain of tasks that decides the finish date: a day late there is a day late for the project. Every other task has slack, and knowing how much is the difference between running a project and being run by it.
When you switch it on, there are two opposite pathologies. Almost everything is critical: the plan has no slack anywhere, and anything that goes wrong moves the finish date. Almost nothing is critical, or the chain is broken into pieces: the most common case, and the direct consequence of constraints. If the dates are nailed, Project has nothing to propagate and the critical path becomes decoration.
If the plan is yours: look at it at every update, not only during planning: slack gets used up, and the chain moves by itself onto another branch. Late tasks are where it usually moves first.
If the plan was sent to you: more than the yes/no of the Critical flag, show Total Slack: it says how many days of breathing space each task has. Negative slack means the plan is already late against itself, and almost always nobody had noticed.
Part C: resources
10. Resources are assigned to phases
Summary tasks are containers: Project rolls up the cost and hours of the tasks inside. When somebody assigns a resource to the phase row, that cost does not replace the sum of the children: it is added to it, calculated over the duration of the whole container.
The arithmetic is brutal. A phase of three 10-day tasks in sequence, with the same person assigned to all three, is worth 240 hours. Assigning that person to the phase row too, which lasts thirty days, makes Project add another 240 hours. The budget doubles, and not one row looks wrong: the three tasks tell the truth, it is the sum that lies. On top of that the person shows at 200% for the whole phase, and resource levelling becomes useless.
If the plan is yours: resources go only where the work is, on the detail tasks. If what you meant was “Bianchi owns this phase”, the place is a renamed custom field; if it was “this cost lasts as long as the phase”, the place is a dedicated task. The full mechanism is here, with the numbers and the three right alternatives.
If the plan was sent to you: show Resource Names and look only at the bold rows. If one of them has a name, the total of that plan is not a total.
11. Resources have no names, or no hours
You open the Resource Sheet and find “Developer 1”, “Resource A”, “Electrical team”. A plan of generic resources has not been agreed with anyone: no real person has said yes to those dates. That is perfectly fine while estimating, indeed correct, and it becomes a problem the moment the plan is approved as a commitment.
The parallel flaw is the plan with durations but no work. Duration is how long a task lasts; work is how many hours it costs. A ten-day task can be worth eighty hours or eight: without work the plan cannot say whether the load is sustainable, and cannot produce any work curve.
If the plan is yours: keep generic resources while you estimate and replace them with people before committing. How the hours of each person are set, even on the same task, is in two resources with different effort on the same task.
If the plan was sent to you: look at the Resource Sheet and at the Work column. Generic names and zero work together say that plan is an estimate presented as a schedule.
12. Nobody ever checked overallocation
The plan’s dates only work if the people who have to meet them are free to work on them. In plans built in a hurry it regularly happens that the same person is booked full time on three parallel tasks: the plan is consistent, reality is not.
Project flags it, the red figure in the Indicators column, red numbers in the Resource Usage view, but it also flags one-hour overloads, so in many plans the warning is so frequent it becomes invisible. Automatic levelling exists and has limited practical value: it moves work by stretching the project, without knowing which tasks really matter. In the end a human decision is always needed.
If the plan is yours: look at the peaks, not the warnings: who is above 100% for more than a few days in a row, and when. There are three solutions, and they are decided, not calculated: move the work, add someone, or reduce the effort on one of the two tasks.
If the plan was sent to you: open the Resource Usage view and look for red numbers. If you find people at 200% for whole weeks, every date in front of you tells a story that will not happen. And if the overallocation involves the very names that appear on the critical tasks, the finish date is already lost.
13. The calendar does not know the holidays
Every plan is born with a calendar that says Monday to Friday, eight hours a day, all year round. Zero holidays: no Christmas, no bank holidays, and still less the company summer shutdown, which in many firms is two full weeks.
The consequence is subtle because it produces no visible error: every task that crosses closure days gets shorter on paper, and along a chain the optimism adds up. An eight-day task that straddles a two-week shutdown finishes two weeks earlier than it really will, and it is better to know in June than to find out in September.
The second layer is people: one person’s leave, a Friday part-time, shifts. Those do not belong in the project calendar but in the resource calendars, which have their own rules of precedence.
If the plan is yours: five minutes in the calendar exceptions, once per project. Better still: an empty file with the calendar already filled in, used as the base for every new plan, and the same for everyone who plans with you, because different calendars produce dates that cannot be compared.
If the plan was sent to you: open the project calendar (Change Working Time > Exceptions) and see whether the exceptions are there. If the list is empty, every date in the plan is optimistic by days or weeks, and nobody will notice until it is too late. It is the fastest and most rewarding check in this whole guide.
Part D: control
14. There is no saved baseline
The baseline is the snapshot of the plan at the moment of commitment: dates, durations, work and costs frozen. It serves one purpose, and nothing else serves it: giving a point of comparison. Without it, a month from now, when the dates have changed, there will be no way of knowing by how much they changed.
No baseline means no variances, no earned value, no measure of the delay: only the feeling that things are getting worse. And it is a five-second check: if the Baseline Start and Baseline Finish columns are empty, that plan is not yet an approved plan, whatever the email that brought it says.
If the plan is yours: save it once, when the plan is approved, and save it again only for an approved change. Project keeps up to eleven: on day one it pays to save it twice, so the original stays untouchable while the working copy follows the replans. The reason is here.
If the plan was sent to you: Baseline Start and Baseline Finish columns. Empty: the plan cannot be measured. Full but identical to the current dates all the way down, on a project under way: the baseline was saved again recently, and the variances you read measure the last decision, not the original commitment.
15. The status date is not set
The status date is the line against which Project decides what is past and what is future. It is not “today”: it is the date the update you are looking at refers to. When it is empty (Project shows NA), Project uses today’s date, which means the numbers change every time you open the file, and two people opening it on different days see different variances. In our test the planned value of the same phase was €4,750 with no status date and €8,750 with the status date set at the end of the month.
All project control rests on this date: progress indices, the line on the chart, the rescheduling of remaining work. Getting it wrong, or leaving it out, silently distorts everything else.
If the plan is yours: set it before entering any progress data (Project > Project Information), and display its line on the chart, so it is visible instead of hidden in a dialog box.
If the plan was sent to you: it is the first of the checks on whether the plan is being run. Empty or months old, you already know what the progress figures shown in the meeting are worth.
16. The schedule is not being updated
A plan that is not updated stops being a tool after two weeks. Updating does not mean changing the dates: it means recording what happened, actual start, how much has been done, how much work is left, and letting the engine recalculate the rest, rescheduling unfinished work after the status date.
There is a worse variant than not updating, and it is updating by moving the bars: dragging tasks forward until the plan looks like reality again. The result looks updated and is empty: every date moved by hand is one more constraint, and no trace remains of what really happened.
If the plan is yours: the cycle has five steps, always the same, the numbers come from the people doing the work, and what counts more than the procedure is the rhythm: a plan updated every two weeks with approximate data is worth far more than a plan updated once with perfect data, because it is the series that shows the direction.
If the plan was sent to you: the question is not “is it up to date?”, because the answer will always be yes. It is “up to what date, and by what procedure?”. And the proof is in the actual date columns: the last actual date in the file is the last real update, far more reliable than the file’s modified date. The other six signs are collected here.
17. Nobody reads the variances
There are carefully updated plans whose variances nobody looks at. It is the most frustrating form of the problem, because the work was done and the value is not collected: the Variance table applies in one click and says, row by row, how far things have moved from the commitment.
And there is the second step, which almost nobody takes: the direction. A ten percent delay that is widening is an emergency; the same ten percent shrinking is a problem that solves itself. Same number, two opposite decisions, and only the comparison between successive readings tells them apart: a delay has a direction, not just a size. The numbers that measure all this formally are in the complete guide to earned value, with its counterpart on the time side, earned schedule.
If the plan is yours: read the variances right after the update, and report them even when the news is bad: a delay told early is a project problem, told late it is a trust problem.
If the plan was sent to you: apply the Variance table and sort by finish variance, descending. In ten seconds you have the ten tasks pushing the project off course, the only ones worth talking about in the meeting.
Part E: when the plan leaves your computer
18. The file goes out as it is
A Microsoft Project plan contains much more than you can see: people’s hourly rates, full names, notes written in meetings, links to internal documents, custom fields with the builder’s annotations, and the history of the saved baselines, that is, the story of every time you replanned.
None of it is visible in the view you are looking at, and all of it travels with the file when you attach it to an email. When the recipient is outside the company there is a regulatory side too, because some of that data is personal.
If the plan is yours: the original does not go out, a cleaned copy does. By hand it takes a few steps, and the boredom of doing them every time is what led us to build GanttShield, which does the same cleaning in one click on a copy, without touching the original file.
If the plan was sent to you: the reverse applies, and it is an advantage to use with care: the file you received holds rates, notes and baselines that tell the story of the project. Before using them in a negotiation, it is worth asking whether whoever sent it knew they were giving them to you.
19. The recipient does not have Microsoft Project
The client, the boss, the supplier: hardly any of them has a licence, and they are right, the licence is for whoever builds the plan. Every way of getting the plan out of the file loses something, and the choice depends on one question: does the recipient need to read, work on it or decide?
The PDF keeps the look and loses the calculation; the spreadsheet export keeps the data and loses the engine; an image of the timeline loses everything but the message, which for someone who has to decide is exactly what is needed. The worst present for a client is a forty-page PDF with three hundred rows.
If the plan is yours: prepare once the two outputs you will use for the whole project, a print view and a timeline, and save them. The five ways, with what each one loses, are compared here; and to tell each person only their own tasks there is email, without Project Server.
If the plan was sent to you: if you receive a PDF and have to check it, ask for the file: from a PDF you can check nothing that matters. If there is reluctance, ask for a cleaned copy, without costs and names but with structure, links and dates intact. It is all you need to judge a plan.
The inspection sheet
Everything above, in the form I use when I have to give a verdict in half an hour. The right-hand column is the part that counts: the question to ask whoever built the plan.
The same checks as a sheet to keep next to the keyboard: the Microsoft Project plan review checklist, with a one-page PDF to print.
Applying Microsoft Project best practices: where to start
If you are reading with a plan open and have counted ten symptoms, the temptation is to rebuild it from scratch. It is almost always the wrong choice: an imperfect plan that people use is worth more than a perfect plan nobody updates.
The order that works is the order of dependencies, the same as Part 0. Constraints first, because while they are there the plan does not react and every other fix is provisional. Then the network: dangling tasks and links on summaries. Then the calendar, which takes five minutes and moves every date. At that point, and only then, save the baseline and start measuring. Structure and names are fixed along the way, without stopping anything.
And if the plan is not yours, the goal is not to correct it: it is to know how far you can trust it, and for what. A plan with a solid network and no baseline is very useful to understand how the work is made and useless to say where we stand. They are two different things, and mixing them up is why some meetings end with decisions taken on March figures.
Frequently asked questions
How many of these best practices apply outside Microsoft Project?
Almost all of them, because they are not about the software but about the critical path method, which is the same in any scheduling tool. The names of the commands change, and a few details of behaviour. The two points most tied to Project are manual task scheduling and the way multiple baselines are handled.
The plan I inherited has almost all these symptoms: do I rebuild it?
Only if it is small. On a large plan, follow the sequence above: constraints first, then the network, then the calendar, then save the baseline and start measuring from today. Rebuilding from zero takes weeks and produces, in the meantime, the only thing worse than a wrong plan: no plan.
Is there a threshold beyond which a plan should be considered unreliable?
The thresholds that circulate, “no more than ten percent of tasks constrained”, “no more than five percent without successors”, come from the schedule quality checklists used in large contracts, and as orders of magnitude they are reasonable. They are not measurements made on real plans, though: they are conventions. The criterion I use is more prudent and more useful: what matters is not how many symptoms there are, but whether the plan reacts. Change a duration in the middle of the network and see whether the finish date moves. If it does not, it is not a model, whatever percentage you counted.
Where do I start if I only have ten minutes?
Three checks, in this order: the Constraint Type column, the Successors column, and the project calendar. They tell you, respectively, whether the dates are calculated, whether the network passes on delays and whether time is measured on a real year. A plan that passes these three may have many flaws, but it is a plan.
How is this guide different from a schedule quality checklist?
Formal checklists count: how many tasks are constrained, how many have no successors, how many have negative slack. They produce a score, they are useful in contracts and they explain nothing. Here the order is reversed: you start from what you see, work back to the mechanism that produced it and arrive at the question to ask. The score says the plan is sick; the mechanism says of what, and what happens if you do not treat it.
Do I need to apply all of them on a small project?
No, and insisting on it is the surest way to make people abandon the tool. On a project of twenty tasks and two months, four things are enough: no dates typed by hand, a closed network, a calendar with the holidays, a saved baseline. Everything else becomes important as the number of people and the duration grow, that is, when the plan has to speak to someone other than you.
Further reading on this blog, part by part
Reading a plan
- How to review a Microsoft Project plan you did not build: five checks
- Microsoft Project plan review checklist (with PDF)
Part A: structure
- Work Breakdown Structure: how to build one that survives the project
- The WBS that doesn’t hold: five signs you can see in the plan
- One WBS, more than one way to see it: outline codes in Microsoft Project
- The WBS code is not an address: what happens when the plan changes
- The columns you are missing already exist
Part B: time
- Task dependencies in a project schedule: the four link types
- Project estimating methods
- Rolling wave planning
- Late tasks in Microsoft Project
Part C: resources
- Don’t assign resources to summary tasks (the budget shows why)
- Two resources with different effort on the same task
- The work S-curve in Microsoft Project
- Microsoft Project calendar and holidays
Part D: control
- How to update a Microsoft Project plan: baseline, status date and progress
- How to tell a Microsoft Project plan is not up to date
- What is a baseline in Microsoft Project
- Baselines after replanning
- How often to update a project plan
- How to collect actuals from your team
- The superintendent moved the crew: don’t rewrite the links
- Earned value: the complete guide
- What is earned schedule
- Your project delay has a direction, not just a size
- Don’t hide project delay from the client
Part E: sharing the plan
Disclosure: this is the blog of a company that makes software for Microsoft Project files, including GanttShield mentioned above. Every check in this guide is done with Project desktop alone; the measured figures come from tests we ran in Project Professional.
#MSProject #ProjectManagement #Scheduling #ScheduleQuality #PMO