Media effectiveness

What each channel adds, where the next pound should go, and what the evidence can support

156 weeks of sales and 4 paid channels. Paid social earns more than it costs at the margin. The evidence cannot yet call Television, Paid search and Display. The regression alone credited search with 1.9× what a regional test found it adds.

Prepared for a constructed retailer. Every figure below was generated by the model from a constructed population when this report was produced — not any firm’s data or result.

How these figures were produced — the model

Your questions, answered

What a media plan needs to know, in a line each

Which channel deserves more budget?: Paid social. The next pound there returns more margin than it costs — and the whole interval says so.
Which channel is wasting money?: None is proven to. Television, Paid search and Display span breakeven and need a test before budget moves either way.
Is paid search as good as it looks?: No. A regression alone credits it with 1.9× what switching it off in test regions shows it adds.
How long does a campaign take to work?: Television carries 60% of its effect into the next week; half has landed within 2 weeks.
Would moving the budget earn more?: Not proven. The fitted curves say £20.5k a week; the evidence runs from −£25.6k to £42.7k. Trial it in test regions first.
Did stock hold the campaign back?: 6 weeks sold out. In those weeks, sales measured the shelf, not the media.

Background and approach

What the spend bought, read against everything else that moves sales

Weekly sales set against weekly spend on each channel, with the season, price, promotion and stock alongside — the things a spend-against-sales chart otherwise credits to media.

Channels read: Television, Paid search, Paid social and Display.
Each channel’s effect is allowed to carry into later weeks and to flatten as spend rises; both are estimated, not assumed.
Paid search was switched off in a random 20% of 50 regions for 4 weeks, and the model’s search effect is set to what that test measured.
Every figure is in margin — sales after the cost of goods, at 32% — so a channel is judged on what it leaves, not what it turns over.

Campaign statistics

The period at a glance

Weeks156of weekly sales and spend

Media spend£32.7mover the period

Spend a week£209.7kacross 4 channels

Sales a week£4.1mall sources

Margin from media£255.1ka week, estimated

Sold-out weeks6left out of the read

Spend and sales over time

Sales and spend, week by week

Two charts, not one with two scales: sales and spend are different quantities, and a second axis lets any two lines be made to look related. Shaded weeks sold out.

Show as a table

Week	Sales
1	£3.2m
5	£3.4m
9	£3.7m
13	£3.3m
17	£3.8m
21	£3.6m
25	£3.9m
29	£3.6m
33	£4.2m
37	£4.4m
41	£4.4m
45	£4.4m
49	£4.0m
53	£3.9m
57	£3.9m
61	£3.9m
65	£3.9m
69	£3.9m
73	£4.2m
77	£4.5m
81	£4.6m
85	£4.7m
89	£4.6m
93	£4.0m
97	£4.4m
101	£4.4m
105	£4.3m
109	£3.9m
113	£3.8m
117	£4.3m
121	£4.2m
125	£4.0m
129	£4.3m
133	£4.7m
137	£5.0m
141	£4.6m
145	£4.8m
149	£4.9m
153	£4.4m

TelevisionPaid searchPaid socialDisplay

Show as a table

Week	Television	Paid search	Paid social	Display
1	£70k	£11k	£27k	£32k
5	£154k	£15k	£10k	£36k
9	£0k	£52k	£14k	£39k
13	£68k	£34k	£20k	£43k
17	£0k	£44k	£22k	£38k
21	£109k	£32k	£18k	£36k
25	£70k	£39k	£16k	£47k
29	£72k	£47k	£16k	£62k
33	£62k	£37k	£15k	£61k
37	£125k	£43k	£12k	£57k
41	£126k	£57k	£15k	£69k
45	£73k	£49k	£16k	£68k
49	£189k	£46k	£14k	£52k
53	£122k	£17k	£14k	£43k
57	£145k	£40k	£16k	£38k
61	£0k	£48k	£21k	£26k
65	£203k	£55k	£32k	£13k
69	£56k	£20k	£44k	£13k
73	£191k	£20k	£60k	£24k
77	£67k	£51k	£63k	£15k
81	£73k	£68k	£67k	£25k
85	£76k	£41k	£76k	£18k
89	£141k	£38k	£74k	£13k
93	£75k	£42k	£73k	£10k
97	£59k	£47k	£78k	£12k
101	£149k	£35k	£65k	£6k
105	£59k	£18k	£77k	£5k
109	£155k	£34k	£67k	£18k
113	£73k	£21k	£83k	£21k
117	£136k	£28k	£76k	£22k
121	£135k	£24k	£70k	£26k
125	£0k	£20k	£55k	£25k
129	£137k	£18k	£79k	£31k
133	£66k	£31k	£86k	£34k
137	£58k	£42k	£88k	£44k
141	£0k	£45k	£78k	£51k
145	£99k	£27k	£57k	£57k
149	£50k	£35k	£55k	£62k
153	£136k	£31k	£68k	£65k

Diminishing returns

Where more spend stops buying more

Each channel’s fitted response: weekly sales from a steady weekly spend, from nothing to well past today. The dot is today’s spend. Where a curve has flattened, the next pound buys little — however well the channel paid on average.

TelevisionPaid searchPaid socialDisplay

Point estimates of the curve. The calls below are read from intervals, not from the curve’s shape.

Show as a table

Spend a week	Television	Paid search	Paid social	Display
£0k	£0k	£0k	£0k	£0k
£38k	£173k	£83k	£184k	£31k
£76k	£295k	£152k	£284k	£49k
£115k	£386k	£210k	£347k	£59k
£153k	£456k	£259k	£390k	£67k
£191k	£512k	£301k	£422k	£72k
£229k	£558k	£339k	£446k	£76k

Television: at today’s £91.7k a week it brings about £334k of sales a week; at 2.5× the spend, about £558k.
Paid search: at today’s £36.0k a week it brings about £176k of sales a week; at 2.5× the spend, about £339k.
Paid social: at today’s £47.2k a week it brings about £312k of sales a week; at 2.5× the spend, about £446k.
Display: at today’s £34.8k a week it brings about £53k of sales a week; at 2.5× the spend, about £76k.

Channel by channel

Whether each channel pays, and whether to add to it

Two different questions. Margin per pound says whether a channel paid for itself; margin per next pound says whether to give it more. A call is made only where the whole interval is on one side of breakeven.

Margin per pound spent

Show as a table

	Estimate	From	To
Television	1.06×	0.87×	1.33×
Paid search	1.47×	0.76×	2.56×
Paid social	1.90×	1.18×	4.46×
Display	0.45×	−0.20×	1.53×

Margin per next pound

Show as a table

	Estimate	From	To
Television	0.87×	0.78×	1.04×
Paid search	1.30×	0.57×	2.22×
Paid social	1.39×	1.02×	2.17×
Display	0.30×	−0.11×	0.64×

Channel	Spend a week	Margin it adds a week	Call
Television	£91.7k	£96.8k (£80.1k to £122.3k)	Not determined
Paid search	£36.0k	£53.0k (£27.2k to £92.0k)	Not determined
Paid social	£47.2k	£89.6k (£56.0k to £210.7k)	Grow
Display	£34.8k	£15.7k (−£7.0k to £53.3k)	Not determined

The lag

How long a pound takes to work

Part of a week’s spend sells that week and part sells later. A channel judged only on the week it runs is judged on part of what it did.

Channel	Carried into the next week	Half the effect has landed within
Television	60%	2 weeks
Paid search	10%	1 week
Paid social	20%	1 week
Display	0%	1 week

Read a campaign over the window in which it works: Television over 2 weeks or more.

Search, against the test

Search followed demand, and the regression credited it for the demand

Search bids rise when more people are already looking. A regression sees search spend and sales rise together and credits search with sales it only followed. Switching it off in a random set of regions separates the two.

Regression alone£1,030ksales credited to search over the test weeks

The regional test£545k£224k to £867k

Over-credit1.9×regression against test

Every figure in this report uses the test’s value for search. Without it, search would look 1.9× as effective as it is, and the budget would follow the error.

Stock availability

Weeks the shelf ran out are not weeks the media failed

In a week that sells out, sales record what the stock allowed, not what the spend created. Read as demand, those weeks make the media look weaker exactly when it was working hardest.

6 of 156 weeks sold out (shaded in the sales chart above). They were left out of the fit rather than read as low demand.

For the plan: stock for the peak. A campaign that sells a line out spends its last weeks advertising something nobody can buy.

The same budget, moved

Where the budget would go if the next pound always went where it earns most

The same total, reassigned pound by pound to the channel whose fitted curve returns most for the next one. The gain is read with its interval.

TodayPlanned

Show as a table

	Today	Planned
Television	£91.7k	£78.0k
Paid search	£36.0k	£72.0k
Paid social	£47.2k	£57.0k
Display	£34.8k	£2.0k

Margin a week, today£280.3kat a steady spend, on the fitted curves — not the weeks as they ran

Margin a week, planned£300.8kunder the fitted curves

Gain a week£20.5k−£25.6k to £42.7k

not determined

Not proven. The fitted curves say the move earns more; the same evidence is consistent with it earning less. Run the planned mix in a random set of regions first and read it against the rest.

Conclusions and recommendations

What to do with the budget

Television: not determined. Its return spans breakeven (0.87× to 1.33×). Test it — a regional holdout like the search test — before moving budget either way.
Paid search: not determined. Its return spans breakeven (0.76× to 2.56×). Test it — a regional holdout like the search test — before moving budget either way.
Add to Paid social. The next pound returns 1.02× to 2.17× in margin.
Display: not determined. Its return spans breakeven (−0.20× to 1.53×). Test it — a regional holdout like the search test — before moving budget either way.
Keep reading search against a test. The regression alone over-credits it 1.9×.
Plan stock with the media. 6 sold-out weeks were weeks the spend could not be converted.
Judge each channel over the weeks it works in, not the week it runs.
Trial the planned mix in test regions before moving the whole budget — the gain is not yet shown.

Methodology

How this was produced

Weekly sales modelled on trend, season, price, promotion and each channel’s spend, with carry-over and saturation chosen per channel from a grid.
Search’s effect set to the regional test; every other channel estimated from the sales record.
Intervals from 40 bootstrap replicates, each re-choosing carry-over and saturation, at 90%.
Sold-out weeks left out of the fit.
The population is constructed. Every figure was generated by the model when this report was produced.

The model: fit, intervals and the search test

Media mix — sample report