Senior Search Director
Recently, I got the chance to speak on one my favourite topics: SEO and Forecasting. For those of you who missed it, here’s a recap…
At BrightonSEO, I presented a talk titled ‘Stat Packages: A secret recipe to get more SEO implemented’. Firstly, a big thank you to those who attended. It was an amazing experience; in Brighton, with iCrossing, with the broader SEO community. Even the weather held out!
I wanted to do this talk because I have valuable first-hand experience in seeing the challenges SEOs face in getting changes implemented. This led me to think: what could help us?
The answer? Better informed business cases and clear value predictions.
This is where Facebook’s Prophet and Google’s Causal Impact come in.
These two packages have been around for roughly 5 and 7 years respectively and were built by the two big techs’ data scientists.
Prophet was built to showcase the growth potential of advertising on Facebook’s platforms.
Causal Impact’s mission was to help clearly show the value driven by a change to a campaign.
They both populate using time series data – perfect for all channels. But for today, I’m shining a spotlight on Organic Search!
It’s very good at mapping seasonality - It does this by running additive regressions day vs day, week vs week, month vs month
It can work with missing data/can exclude events to prevent skew
It’s output is laid out as upper/middle/lower conditions: Upper – performance improves as much as possible; Middle – some growth and Lower – decline by not making improvements
This, therefore, really helps show the value SEO implementation can bring over the coming year, helping make sure the channel does not get overlooked.
It’s built to allow advertisers to see what changes contributed to upticks in performance - It does this by running Bayesian linear regressions to determine probability
It’s very good at isolating value
It visualises this as culminative growth: + additional clicks, + % growth, the model found the change significant and by week total improvement since launch
In combination, these two are an unstoppable force in demonstrating the impact of implementing change, evidencing this as a number, which shows a very clear return on investment.
These forecasts are by no means always certain. But they’re the best we have. We just need to keep our data in a tidy fashion, giving the models what they need to get a more granular understanding of the time series data.
To do this:
Consider moving data storage to BigQuery
Have everything labelled from all keywords to the pages they click through to (think: intent behind searches, query theme, product, user journey stage)
See the output of the forecast as indicative (not certain)
Put simply, we get a forecasting methodology that moves us along from click curve models.
Click curve models estimate position improvement from SEO changes. But, as time’s gone on, they’ve been found to be less and less accurate. In comparison, Prophet and Causal Impact’s forecasting (looking at actual business metrics such as clicks and impressions) works wonders.
Second to this, these packages were made by Google and Facebook. This helps in showing that they were made by some of the best statisticians, to predictably work with time series data.
Finally, they prove the value that a set of SEO implementations has driven. This helps quantify the uplift driven as a result of that dev release, and unmuddies some other factors at play.
Forecasting has come a long way, and is now better enabled using these packages
Prophet is amazing at showing how much growth is possible for the channel
Casual Impact can get granular, showing the results implementations have driven
I hope by sharing these packages, and therein more people using them, will help us all show what opportunity there is in taking SEO recommendations seriously. It helps us show we’re true to our word, especially when combined with a very clear review interval after something’s gone live.
In terms of how we work, I’d like to see SEO teams and devs work much closer to ultimately get more of our hard work implemented and drive the results we know we can achieve.
Got a question about how to use stats for forecasting, SEO and data science more broadly, or how to use these for your business cases? Get in touch. I’d be delighted to talk with you and your teams.
We believe that moving too slowly in digital is the biggest risk your business faces. If you are ready to move faster in digital, we are here to help.