Work Breakdown Structure: Find the Work Nobody Listed
The project was ninety percent complete for five weeks. The build was done. What remained was the testing nobody had scheduled, the training nobody had planned and the data nobody had cleaned. None of it was on the plan, so none of it was in the budget.
What it is
A work breakdown structure takes everything a project must deliver and breaks it down, level by level, into pieces small enough to estimate, assign and finish. The top is the whole project. Under it are the main deliverables. Under each of those are smaller parts, down to work packages that one person or team can own.
Two ideas make it useful. It is organised around deliverables, the things that will exist at the end, and not around activities. And it is meant to be complete: every piece of work in the project appears in it once, and nothing outside the project appears at all. Project managers call that the 100 percent rule.
The point of drawing it is to find the work nobody mentioned.
How it goes wrong
| Failure | What it looks like | What it costs |
|---|---|---|
| A list of activities | Hold meetings, do analysis, write code | Nothing to check off as finished |
| Only the visible work | The build is there, the testing is not | The budget runs out at eighty percent |
| Packages too big | Implement the system, twelve weeks | No early warning when it slips |
| Packages with no owner | Assigned to the team | Everybody assumes somebody else has it |
| Too much detail | Three hundred lines for a small project | Nobody maintains it after week two |
A common rule of thumb is that a work package should take between a day and two weeks. Smaller than that and you are writing a to-do list. Larger and you cannot tell whether it is on track until it is late.
The work that gets left off
Across projects, the missing pieces tend to be the unglamorous ones: testing, training, documentation, data migration, approvals, the handover. In projects that involve AI there is a further set that plans leave out because the people writing them have not run one before.
| Forgotten work package | Why it is needed |
|---|---|
| Cleaning and preparing the data | The tool is only as good as what it reads |
| A test set with checked answers | Without it nobody can say whether it works |
| Designing the human review step | Who checks what, and what happens to uncertain cases |
| Training the people who will use it | Unused tools deliver nothing |
| Monitoring after launch | Quality can drift without any error message |
| A way to switch it off | The fallback when something goes wrong |
Each of these is real work with a real cost, and each has its own guide here: data cleansing, the test set in a proof of concept done properly, and the baseline in the gap analysis. Leaving them off the structure does not remove them from the project. It moves them to the end, unbudgeted.
Where AI earns its place
A first draft from the charter. Given the project charter or the statement of work, it can propose the deliverables and a first level of breakdown for the team to correct.
Checking for gaps. Ask what is usually needed for a project like this and missing from the draft. It will usually suggest testing, training and migration, and the suggestions are cheap to review.
Comparing with your own past projects. If you have earlier plans and the list of things that were added late, matching the new structure against them finds your organisation's particular blind spots.
Tidying. Turning activities into deliverables, flagging packages with no owner and packages that look too large are quick mechanical checks.
What stays with people
Estimates. A model will put hours against every package in a confident column with no basis in how your team works. Estimates come from the people doing the work, and the breakdown exists so that they can give them.
Scope. Whether a piece of work is in the project is a decision for the sponsor. If the draft adds something, somebody has to agree to pay for it or take it out.
Ownership. A name against a package is a commitment, and it has to be made by the person named. The risks that surface while you do this belong on the risk register.
Frequently Asked Questions
What is a work breakdown structure?
A hierarchical breakdown of everything a project must deliver. The whole project sits at the top, the main deliverables below it, and each deliverable is divided into smaller parts until you reach work packages that a person or team can own, estimate and complete.
What is the 100 percent rule?
The principle that a work breakdown structure contains all of the work in the project and none that is outside it. Every piece of work appears once. The rule is what makes the structure a check on scope and not only a diagram.
How detailed should it be?
Detailed enough that each work package can be estimated and assigned, and no further. A common rule of thumb is that a package should take between a day and two weeks. Much smaller and it becomes a to-do list. Much larger and you cannot see a delay until it has happened.
What do AI projects leave out of the breakdown?
Usually data preparation, a test set with checked answers, the design of the human review step, training for the people who will use the tool, monitoring after launch and a way to switch it off. These are real work packages, and omitting them pushes the cost to the end of the project.
How can AI help build a work breakdown structure?
It can draft the deliverables from a charter or statement of work, suggest work that is commonly needed and missing, compare the draft with past projects, convert activities into deliverables and flag packages that have no owner or look too large.
Should AI estimate the work packages?
No. It will produce plausible numbers with no basis in how your team works. Estimates should come from the people who will do the work. Decisions about what is in scope belong to the sponsor, and ownership has to be accepted by the person named.
Plan the whole project
We break AI and automation projects into owned deliverables, including the data, testing and training work that plans leave out.
Related Articles
The Next AI Bet: Agents Trained on How You Work
ChatGPT Work: AI That Acts Across Your Apps
Microsoft’s Own AI and Work IQ: What It Means for Business
Ajan leads the ChatGPT.ca team: 200+ custom GPT builds and automation projects for 50+ businesses across 20+ industries. Based in Markham, Ontario. PIPEDA-compliant solutions.