Winning with small data in PPC

Working with the data you have (not the data you wish you had).

If I asked you to draw a picture, pixel by pixel, on this grid…

(have a go if you like!) 

What would you draw?

A house seems doable… 

A face… 

Boat maybe… Stick man…

At a push you could try a tree.

(What do you reckon…?)

You’d be wise not to try anything much more intricate, if you want to make it recognisable. 

You probably wouldn’t try to capture the enigmatic almost-smile of the Mona Lisa, for example.


The Point?


The same thing applies to the pictures and patterns we try to paint with our data.


When you don’t have many data points (either in your whole account, or whatever component you’re looking to evaluate) you have to accept that you’re working with a broad brush if you want those patterns to mean anything at all.

That means being selective about the variables you use for analysing performance, and sticking to those dimensions that don’t offer too many different values.

For example, Device Category is ideal for low data volumes because it’s split into just three values (mobile, desktop and tablet). 

Between these three, more often than not, you’ll see a meaningful pattern worth acting on. 

Similarly, Gender – so far evading the PC treatment – has just three values (‘unknown’ being the third)  whereas Age Range, with seven, requires much more data to distribute into a reliable pattern.

Day of the Week – also with seven values – is more demanding still… 

Any individual day could be skewed by external factors (due to one-off or annual events, or some unusual activity on the site or campaign side)… Say you’re unlucky, and that happens on two consecutive Wednesdays. It will then take months to gather enough data to iron out the wrinkle in your data patterns. 

If you’ve been working with PPC for a while you will have seen this in practice. How often have you made day-of-week adjustments and then found they bear no relation to performance trends the next time you check the report a few weeks later? 


The Solution


One practical way you can use this principle is in granularity of the audiences you apply.

If your campaigns aren’t teeming with data, don’t use remarketing audiences split into: 15-day site visitors / 30- day / 60-day / 2xproduct page viewers / cart abandoners etc. 

Limit it to fewer, more coarse-grained groups. e.g. ‘all users’ as step one / then add ‘product viewers’… ‘cart abandoners’ and so on when you data stretches to it.


I usually add In-Market audiences on ‘Observation’ to my campaigns, as they often reveal some useful, unintuitive patterns. 

But – if you add them all – and you’re running a £1000/month campaign with CPCs around £2-£3… well, you get the picture (i.e. you won’t get the picture). 

Instead, add fewer, broader categories.  

e.g. rather than adding ‘Banking Services’, ‘Credit & Lending’ etc, just try ‘Financial Services’.

What Else?

When you have fragmented data, is it safe to lump low-volume items together?

To put this another way, can you treat

Paris: spend £20,000 / transactions: 100

The same as 

A group of 50 lower-volume cities with aggregate spend: £20,000 / aggregate transactions:100 ?

Could you apply a bulk bid adjustment to that group of cities with the same confidence you would to Paris?

This is a truly consequential question for us PPCers, but the answer (to me) isn’t obvious – so I ran it by an econometrician friend this week…

Q: In the aggregated group of cities each with a low volume of data, each individual city is on shaky ground, but the group as a whole will consist of some cities showing uncharacteristically high performance and some showing the opposite… does that variance even itself out?

A: The reliability here depends on the number of values (cities) as well as the volume of data. So if you’re dealing with a high number of locations, then yes it’s reasonable to aggregate [good; I’ve been working on that assumption for 15 years!] 

Just as ‘Paris’ could be broken down into sub regions, our small cities could be grouped into an aggregate that has a certain, defined performance level.

So, look for opportunities to aggregate where your data is fragmented but again, approach this with caution… and don’t try to paint the Mona Lisa when you’ve only got 49 pixels to play with.

More Useful Stuff…

Join my Free PPC Facebook Group

Get 1-2-1 Help with Your Campaigns

Want Me to Manage Your Account?

Share this post

Lastest Posts

Google Ads Updates - Q1 2024 Roundup

No slowdown in the pace of change with Google Ads updates in Q1 2024… Here’s…

Aggregation vs Articulation: Competing Virtues of PPC Management

Three ‘A’s Two concepts have been central to the recent evolution of Google Ads: aggregation…