Why creative estimates are always wrong, and what to do instead

Creative estimates are wrong for a structural reason, not a discipline one. You are estimating how long it will take to arrive at an answer you don't have yet. Nobody gets better at that, because the thing being estimated genuinely doesn't exist at the moment of estimating.

So stop trying to estimate more accurately. Measure how wrong you usually are, and correct for it.

The three reasons the number is low

You estimate the work, not the project. Designing the thing is one part of it. The rest is briefing calls, waiting, re-briefing after the client's boss weighs in, exporting in four formats nobody mentioned, the handover. On most projects the surrounding work is a third of the hours and none of the estimate.

You estimate the version in your head. When you picture the job, you picture it going well: the first route lands, feedback is coherent, the client has their assets ready. You are not being naive — you're estimating the only version you can currently see.

Iteration is invisible until it happens. Nobody estimates "we'll try three directions and the third one works". They estimate the third one.

Add the commercial pressure — you want the work, the client mentioned a budget, the number needs to look competitive — and estimates only ever bend one way.

Measure your error rate

The correction is embarrassingly simple: compare what you estimated to what actually happened, across your last several projects, and derive a multiplier.

A hypothetical agency's last six projects:

Project	Estimated	Actual	Ratio
1	120	168	1.40
2	80	92	1.15
3	200	310	1.55
4	150	141	0.94
5	60	96	1.60
6	240	288	1.20
Total	850	1,095	1.29

Correction factor = 1,095 / 850 = 1.29

The median of the individual ratios is 1.30, which is close enough to confirm that one bad project isn't driving the result. This agency's estimates are consistently about 29% light — and crucially, that's a stable property of how they estimate, not a run of bad luck.

Note project 4, which came in under. There's always one, and it's the reason people resist this exercise: "we're not always over". You don't need to always be over. You need the total to be predictable, and it is.

What the correction is worth

Take a new project this agency estimates at 180 hours, priced at $190 an hour — a quote of $34,200. Applying the correction factor gives a realistic 232 hours, at a loaded cost of $95 an hour.

	Quote	Actual cost	Margin
Quoted at raw estimate (180h)	$34,200	$22,040	35.6%
Quoted at corrected estimate (232h)	$44,080	$22,040	50.0%

Same project. Same team. Same actual delivery. The only difference is which number went on the quote, and it's worth 14.4 margin points.

That's the entire argument. You are not estimating better — you are refusing to price against a number you already know to be optimistic.

Do it by category, not in aggregate

A single agency-wide multiplier is a good start and a bad finish. Error rates differ sharply by work type: routine production is usually close to estimate, while conceptual and strategic work is where the multiplier gets large.

Split your history at least into "work we've done many times" and "work with a novel element". If the first comes out at 1.1 and the second at 1.6, applying an average of 1.29 to both means you're overpricing the safe work you'd win easily and underpricing the risky work you shouldn't take cheap.

What you need to make this possible

Nothing elaborate — but you do need actual hours logged against actual projects. Not perfect hours. Same-day, roughly-right hours against the correct project code, consistently enough that the totals mean something.

Most agencies that can't run this exercise can't run it because time entries were filled in on Friday from memory, or because half the project's work was logged under "internal". If that's you, the fix is a month of honest tracking, not a new estimating framework.

Also record the original estimate somewhere it can't be edited. Estimates that get quietly revised upward mid-project produce a correction factor of 1.0 and teach you nothing.

Two habits that help beyond the multiplier

Estimate in ranges internally, single numbers externally. The team should work in "120 to 320 hours"; the client should see one price. Ranges preserve the uncertainty in the conversation where it's useful and remove it from the one where it just invites negotiation.

Estimate with the person who'll do the work, and separately from the person who wants to win the deal. Commercial pressure and estimation should not happen in the same head at the same time. Get the hours first, then decide what to do about the price.

Run it on six projects this week

Pull your last six completed projects. For each, find the original estimated hours and the actual logged hours. Divide the totals.

Whatever number comes out — 1.1, 1.3, 1.6 — apply it to the next quote you send before you convert hours into money. It's the highest-return hour of admin available to an agency owner, and unlike most pricing advice it doesn't require you to have a difficult conversation with anyone.