How to rescue a project that's already gone wrong

The instinct when a project goes wrong is to work harder and say nothing. Push through the weekend, absorb the overrun, hope it comes back. It almost never comes back, and the silence turns a recoverable commercial problem into a relationship one.

Rescuing a project is three steps in order: work out how bad it actually is, pick the lever, then have the conversation. Skipping to step three without step one is why these conversations go badly — you can't negotiate from a feeling.

Step one: project it forward

Stop looking at what you've spent and calculate where you'll land.

A hypothetical project: quoted at $60,000 for 400 hours, loaded cost $95/hour, so a planned margin of $22,000 — 36.7%. It's week 6 of 10. The team has logged 310 hours, and the honest estimate of completion is 55%.

Projected total hours = hours used / fraction complete
                      = 310 / 0.55
                      = 564 hours

	Hours	Cost	Margin	Margin %
As quoted	400	$38,000	$22,000	36.7%
Projected	564	$53,580	$6,420	10.7%

The project is 77.5% through its budget and 55% through its work. Left alone it delivers $6,420 instead of $22,000, and that's the optimistic version — projects in this state usually get worse in the last stretch, not better.

There's a second number worth having. The remaining 254 hours have to fit in four weeks: 63.5 hours a week, against a current burn of about 52. So the plan isn't just over budget, it requires the team to go 22% faster than they have managed so far. Whenever a recovery plan needs a speed-up, name it out loud, because that assumption is usually where the second overrun comes from.

Run this projection before you talk to anyone. It takes ten minutes and it changes the conversation from "we're a bit behind" to a specific, defensible number.

Step two: diagnose the cause honestly

The right lever depends entirely on why you're here, and there are only four real answers.

Scope grew. The work being done is genuinely more than the work quoted. This is the most common and the easiest to address, because the fix is legitimate and the client can see the additions.

The estimate was wrong. Same scope, more effort than you thought. Your problem commercially, whatever you feel about it.

Delivery was inefficient. Rework, wrong person on the task, waiting on internal reviews, a false start. Also your problem.

The client blocked you. Late approvals, changed direction mid-stream, extra stakeholders. Legitimate to raise — if you documented it at the time. If you didn't, raising it now sounds like blame-shifting, because from their side it is indistinguishable from it.

Most bad projects are a mix, and the split matters. If 60% of your overrun is scope growth you have a strong position. If it's mostly a bad estimate, asking for more money will damage the relationship more than absorbing the loss will.

Step three: pick the lever

Four exist. Everything else is a variation.

Reduce scope. Usually the best outcome for both sides. Find the deliverables with the worst effort-to-value ratio and propose cutting or deferring them. Most projects have something the client agreed to without wanting much.

Raise a change request. Only for documented additions. Needs a specific list and a specific number, not "it's grown a bit."

Change the team mix or the approach. Sometimes the remaining work can be done differently — a simpler execution, a template instead of bespoke, a mid-level person instead of a senior. Requires client agreement if it affects the output.

Absorb it and manage the finish. Sometimes right: the fault is yours, the relationship is worth more than the margin, or the client is strategic. Absorbing deliberately is a decision. Absorbing by default is a habit, and it's how agencies with full order books end the year with nothing.

Whatever you pick, the deadline is a lever too. Extending the timeline rarely recovers margin on its own, but it removes the overtime that makes an overrun worse.

The conversation

Have it on a call, not by email, and have it early. The cost of this conversation roughly doubles every two weeks you delay it.

[OPENING — the fact, immediately]

"I want to flag something on the project before it becomes a
 surprise. We're at 77.5% of the budget and about 55% of the work.
 If nothing changes we'll run roughly 40% over on effort.

 I'd rather tell you now, in week 6, while we can still choose
 what to do about it."

[CAUSE — specific, no blame, no hedging]

"Two things got us here. The extra language variants and the
 second round of stakeholder review added work we hadn't scoped —
 that's about 60% of it. The rest is on us: we underestimated
 the integration work.

 I'm not asking you to cover our estimating error."

[OPTIONS — always more than one, always with a recommendation]

"There are three ways forward.

 One: we reduce scope. The two secondary templates are the
 lowest-value items left. Cutting them brings us close to the
 original number and doesn't affect the launch.

 Two: we extend the budget by [amount] to cover the scoped
 additions only, and we absorb our own overrun.

 Three: we hold everything and I'll absorb the difference — but
 I'd be doing that quietly and I don't think that's a good basis
 for us working together on the next thing.

 I'd recommend one. It costs you nothing and it protects the date."

[CLOSE — a decision with a deadline]

"Can we decide by Thursday? After that the team starts on the
 templates and the option to cut them is gone."

Four things make that script work. The number comes first. The cause is specific and includes your own share. There are options rather than a demand. And there's a date attached, so the decision doesn't drift for two weeks while the budget keeps burning.

After the rescue

Two things, once the fire is out.

Write down what actually happened while it's fresh — where the estimate broke, which decision cost you, at what point it was first visible in the numbers. Almost always it was visible weeks before anyone reacted, which is a monitoring problem rather than a delivery one.

Then check whether the same pattern is on your other projects right now. Bad estimates cluster by work type. If the integration work was underestimated by 40% here, the two other projects with integration work are also wrong, and you can find that out this afternoon.

Do this week

Take your most worrying active project and run the projection: hours used divided by fraction complete, times your loaded rate, against the fee. Ten minutes.

If the answer is bad, book the conversation for this week rather than next. Nothing about the situation improves with age, and the version of this conversation where you still have four weeks left is dramatically easier than the one where you have one.