Melissa Lee
Project outcome report · Work, Reimagined. · Simulated case

We shipped 70% more for 14% more money. Then approvals became the bottleneck.

Trust CountedCalculatedEstimatedJudged 2 wins · 2 mixed · 2 misses · 1 open
01Yes, and more.
Delivery · Did we ship what we planned?

68 assets out the door against 40 planned.

+70%
{C}68 vs 40 planned
 02Late.
Delivery · Did it take longer than planned?

Two weeks past plan. The makers finished. The approvers did not.

+2 wks
{C}10 vs 8 weeks
 03Over.
Cost & effort · Did we stay on budget?

Each asset still cost a third less: $1,346 instead of $2,000.

+14%
{K}$91.5k vs $80k
 04Missed.
Review & quality · Did work pass review the first time?

Only 58% of assets passed on the first try. The rest went back.

58%
{C}target was 75%
 05Mostly.
Review & quality · Was the work good?

Craft was strong and the brief was met. Brand consistency slipped after week 3.

7.6
{J}out of 10, 20 assets
 06Not yet.
AI · Did AI pay off?

It saved making time and created checking time. Net: 16 hours worse.

+16h
{E}net effort
 07Unproven.
Business impact · Did it move the business?

Demand ran 18% over baseline. Nothing shows the campaign caused it.

+18%
{K}demo requests
Why it went this way
1
The message moved after work started.Changed twice in week 3. It caused 41 of 70 revision rounds.
2
The approver list doubled.Three became six. Median wait went from 18 hours to 31.
3
AI outran the people checking it.Output jumped. Review hours went from 88 to 200.
What to keep doing
1
Draft with AI at the making stage.Cost per asset down a third, hours per asset down 30%.
2
Run an approval sprint.Two days in week 9 cleared the backlog. Start one in week 3.
What we change next time
Lock the message before any adaptation starts.

The campaign owner attaches an approved messaging framework and names one final approver.

Chapter 1 of 5 · Delivery

70% more than planned. Two weeks late.

Twenty-eight of the 68 assets arrived after kickoff, and the plan never changed.

Sources · review tool, 68 of 68 assets · project tool, 96% of tasks
Assets shipped
68
40 planned, +70%
{C}
Finished
Week 10
Week 8 planned
{C}
Scope added
+28
In weeks 3 and 5
{C}
Figure 1

Assets shipped, week by week

Running total against the original plan. The plan never moved, even after 28 assets were added.
Figure 2

Schedule

Planned
8 wks
Actual
10 wks

What we found

Finding 01

Scope grew and the plan stayed put.

Two requests added 28 assets. Nobody re-baselined the schedule or budget, so 70% more and two weeks late are both measured against a plan that stopped matching the work.

Finding 02

The extra weeks went to waiting.

Median wait for an approval rose from 18 hours to 31. Chapter 3 has the detail.

What we change
Re-baseline when scope moves more than 10%.

Name who approved the change and what it does to time and money.

Chapter 2 of 5 · Cost & effort

14% more money. 70% more work.

Per asset, the campaign cost a third less than planned. The overrun is real, and it is smaller than it looks.

Sources · finance system, 100% · timesheets and review timestamps, 88%
Budget
$91.5k
$80k planned, +14%
{C}
Cost per asset
$1,346
$2,000 planned, −33%
{K}
Total effort
503h
420h planned, +20%
{K}
Hours per asset
7.4h
10.5h planned, −30%
{K}
Figure 3

Money, week by week

Running spend against plan. Actual crossed above plan in week 5 and kept climbing after the planned finish.
Figure 4

Hours by type of work

Making got shorter. Review more than doubled.
Plan · 420h
Making 332h
88h
Actual · 503h
Making 303h
Review 200h
{K}
Figure 5

Rework

Time spent redoing work after review.
~90h
About 18% of all 503 hours went to revision rounds.
{E}

What we found

Finding 01

The overrun is a scope story.

Budget rose 14% on 70% more work. Per asset, the campaign cost a third less than planned.

Finding 02

Review ate the hours.

Making time fell from 332 to 303. Review time rose from 88 to 200.

What we change
Report cost per asset beside total budget.

A bigger scope should not read as an overrun by itself.

Chapter 3 of 5 · Review & quality

Making went fine. Checking is where we lost time.

First-pass approval missed its target in every week but one, and the queue peaked at 22 assets in week 6.

Sources · review tool, 68 of 68 assets · calendar, 78% · creative director scoring, 20 assets
Approved first try
58%
Target 75%
{C}
Revision rounds
70
50 planned, +40%
{C}
Wait for an answer
31h
18h planned, +72%
{C}
Approvers
6
3 planned, +100%
{C}
Review hours
200h
88h planned, +127%
{K}
Figure 6

Approved on the first try, by week

The low point followed the week 3 message change.
Figure 7

Assets waiting for review, by week

Red weeks had more than 12 assets stuck.
Figure 8

Was the work good?

Scored by the creative director against the brief, on a sample of 20 assets. A person's judgment, not a count. {J}
CraftExecution quality of the finished asset
8.4
Fit to the briefDoes it say what the campaign meant to say
7.8
Brand consistencySlipped after the message changed in week 3
6.6

What we found

Finding 01

Adaptations started before the message was locked.

The framework changed twice in week 3. Social work had begun, so it was redone.

Evidence · 41 of 70 revision rounds · brief v2 and v3 · drafted by AI, confirm before sharing
Finding 02

Approvals doubled in size with no turnaround rule.

Five of the six approvers came from legal and regional teams. Nobody owned a response time.

Evidence · approver roster · queue timestamps, 68 assets · drafted by AI, confirm before sharing
What we change
Lock the message first. Cap approvers at three.

Any added approver gets a stated turnaround time.

Chapter 4 of 5 · AI

AI sped up the making. It did not speed up the approving.

The 96 hours it saved is an estimate. The 112 hours of extra review is not.

Sources · asset tags and AI tool logs, 85% tagged · review tool timestamps
Production time saved
−96h
Against the same work without AI
{E}
Extra review time
+112h
Created by AI-assisted assets
{K}
Net effect
+16h
Slightly worse overall
{E}
Figure 9

From plan to actual, in hours

Each bar starts where the last one ended. Hatched means estimated.
Figure 10

AI-assisted against hand-built

29 AI-assisted assets against 39 hand-built. This is counted, not modeled, so it is the strongest evidence in the chapter. {C}
MeasureAI-assistedHand-builtGap
Approved first try45%68%23 points lower
Revision rounds per asset1.50.7About double
Review hours per asset3.7h2.4hAbout 1.5 times
Figure 11

Where AI sits in the process

1

Intake

People set the plan and tag which assets are AI-assisted.

2

Make

AI drafts. This is where it saved time.

AI works here
3

Review

People check. More output landed here.

Bottleneck
4

Approve

People decide. Six approvers, 31 hour wait.

What we found

Finding 01

AI assets need more checking, not less.

They were approved first try 23 points less often and took about 1.5 times the review hours each.

Finding 02

The saving was real. The review bill was not planned for.

Making time fell. Nobody added review capacity to match.

What we change
Set a review budget per batch before AI drafting starts.

Lock the message first. The fix is a limit on what enters review. Less AI would not solve it.

Chapter 5 of 5 · Business impact

Demand rose 18%. Nothing shows the campaign did it.

The demo count is real. What caused it is not settled.

Sources · marketing analytics, 4 weeks after launch · baseline set by Marketing
Demo requests vs baseline
+18%
Average over 4 weeks, from counted requests
{K}
Baseline
8 wks
Trailing average before launch
{K}
Caused by the campaign
Unproven
No control audience
{E}
Figure 12

Demo requests after launch

Percent above baseline, by week after launch. The bars are counted requests. The hatched band marks the email push that overlaps weeks 2 and 3.
Figure 13

How sure are we?

✓
The baseline is definedTrailing 8-week average, set before launch.
✓
Tracking was the same before and afterSame form, same counting rules.
×
The campaign is not proven as the causeThere was no control audience to compare against.
×
Other activity ran in the same windowAn email push overlapped weeks 2 and 3 after launch.

What we found

Finding 01

The trend is encouraging and unproven.

Demand climbed every week, which fits a campaign ramp. It also fits the email push.

What we change
Hold out a control audience next campaign.

That is what moves the verdict off Unproven.