Thread by @darth_na

.
:: We're spoilt - and it's Digitals fault! ::

Unfortunately, digital has kind of treated some of us a little too well.
It's lead to a (somewhat false!) belief that we can tag and track everything,
and attribute appropriately etc.

#DigitalMarketing

>>>

https://twitter.com/SamUnderwoodUK/status/1587010507818549250

>>>

1) Things like TV, Radio and Print manage(d) without such strong attribution for decades

2) More than a few digital attributions are incorrect/shallow/short/misleading (Last Touch!)

3) Annotations and comparative analysis are your friends!

>>>

>>>

4) Using multiple "Models"
(which requires (a) tracking/tagging (b) long/root data)
permits better perspective

The reality is - no method of attribution is perfect.
There's always data loss, confusion, miss-association, contamination etc.

>>>

>>>

You also have to grasp that concept of contributions.

Certain things may not result in immediate action.
Thus the concept of Soak and Top of Mind etc.

You don't watch a TV Advert about laundry detergent,
then rush out and buy some.

But you know that brand!

>>>

>>>

Which leaves you with the problem:
how do you know if those adverts are increasing sales?

You have to be tracking before the ads run.
You need to know figures before the ads run.
You then compare the figures pre/post ad run,
over days/weeks/months.

>>>

>>>

Then it's basic math.
New figure - Old figure = difference.

The problem there is ... you seldom have linear/stable figures.
Sales tend to vary depending on not only the season,
but also the time of the month and day of the week.

So you have to compare like for like...

>>>

>>>

It's no good comparing by Date
(12th of X vs 12th Y),
as more often than not - those are different days!

So remember you may need to nudge one of the value sets along a little to get proper alignment

(2nd Monday of X and Y, last Saturday of X and Y etc.)

>>>

>>>

And similarly, one of the biggest Analytics fails I see,
regularly!,
is people comparing Month to Month,
without looking at the same 2 months from the prior year!

You need to know what the general difference is,
before you look for a specific difference!

>>>

>>>

X = 1,000
Y = 2,000

Your ads did 1,000 (+100%) ?

X 2021 = 300
Y 2021 = 500
X 2022 = 1,000
Y 2022 = 2,000

Now do the math!
(400(ish), not 1,000 (depending on method))

It's not perfect,
but it tends to be more accurate!

>>>

>>>

You can also test/verify ... by running holdbacks/cessations (much easier on Digital than more traditional channels).

Pause a campaign in a specific region,
or via a channel etc.
do it for a few hours/days (depending on size of campaign/volume of actions etc.).

>>>

>>>

If you've got the timing right,
you'll see the blip in your data.

(Do NOT expect the blip to be "live" for most campaigns ... pausing it at 5pm on Tuesday may not result in drops at that time ... but may influence Wednesday-Friday etc.)

>>>

>>>

So - again - none of it is perfect.

But deeper/longer tracking,
with proper analysis,
permits you to see what is more likely to be valuable,
even if indirectly!

>>>

>>>

I learned this due to a poorly converting ad run in a paper.
Diff. Telephone No's ... barely any calls on one of the numbers.

It was decided that they could cut that ad
(they did decide to invest the spend on the other 2 ads though!)

Over the month, calls fell.

>>>

>>>

Put the cut add back.

After a few weeks and bit, calls rose.

That ad did not generate much directly.
But it did get peoples attention!

It indirectly contributed.

You'd be amazed how much that happens with Digital
(esp. SEO and Middle of Funnel content etc.).

>>>

>>>

So ...

Ensure you use UTMs.
Make sure your web tracker captures First as well as Last (and ideally, the in-betweens).
Annotate start/end of campaigns.
Run little holdout tests.
Analyse results with stronger comparisons.
Accept it's never going to be perfect.

:D