Using Artificial Intelligence and Machine Learning in Digital Marketing: BrightBid’s AI & ML models explained for Google Bid Optimisation
Did you know that there are63,000 search queries a second? And of the 97% of the global population who search for information on the internet, 61% use Google every day. Google is a vital marketing tool for any company trying to get in front of ‘high intent’ potential customers when they’re looking for the products or services it sells. It is therefore not surprising that Google is the largest advertising platform in the world generating over $178 billion from search (and youtube) in 2021.
However, it is a time consuming and specialist task for many marketing teams to get to grips with Google paid search (PPC) as it can be complex with thousands, sometimes millions of different keywords, audiences and ad creatives to track. It is also a highly competitive space with many advertisers competing for the same searchers and inflating the price. Google operates an auction model for clicks to set the price (Cost per Click). Demand from advertisers has been higher than the available supply for Google searchers over the last 2 years, which has led to very high levels of CPC inflation at 38% 2021 on 2020 and 15% 2022 on 2021 (Source).
How is Google using AI in Marketing?
Google PPC is exactly the type of task that machines do brilliantly and Google has been using AI within its Search product as an alternative option for advertisers manually optimising their paid search campaigns. The challenge for many marketing teams is that this can be the blackest black box they work with: no breakdown by price, by ad format, media channel or even what creative elements work.
This matters less when Google’s AI is delivering brilliantly but when results are worse than expectations, the advertiser finds they’ve given up control of their marketing investment. Google’s response is often ‘spend more money’ to improve performance which results in an even worse inflationary spiral.
Our aim at BrightBid is to give our customers access to more controllable AI and automation solutions for Google search to enable them to take advantage of this technology to improve their advertising performance without giving up control or having to invest to develop their own AI solution in house.
Our technology team has automated the thought processes and workflow of a PPC expert to ensure that our solution works for all types of customers. Finding the best solution that’s most applicable to our customers in performance terms was the hardest part. It required, and still requires, constant learning and testing and continual improvement as the marketing and ad space quickly changes.
Turning raw data into useful insights for clients or customers: practical deployment
Our AI engine enables customers to test thousands of different combinations of ad copy, audiences, devices, locations, countries and products to find which mix performs best for the goals they have set. We delivered over 50m optimisations in Google Search on our customers’ behalf in 2022: far exceeding the ability of a PPC expert to manually optimise.
We’ve built our own bidding platform on top of Google’s own bid-automation platform so that customers can use all the data triggers collected by Google and then reap the benefits of our AI on top to guide the performance of their spend based on their exact goals.
The parameters within our AI Engine for Google Search
We’ve built a number of our own AI and ML models for the complex process of handling Google Search. Diving deeper into one element – our bid optimisation model – is an unsupervised machine learning model built, and then tailored for each customer using their own past historical data gathered from Google and their own datasets, to set more precise and accurate pricing of the different keywords that exist in a given Google ad group, campaign and account. For this purpose, we look at different variables such as device, time of day, day of the week, audience, age range, income range and gender to price the bids.
The purpose of bid optimisation is to automatically identify high-quality traffic so as to increase the bids on impressions that lead to more conversions (or clicks) and reduce money spent (costs) on impressions that do not lead to conversions for our customers.
For our bid optimisation model, we optimise for 3 goals.
- Effectiveness (Accuracy)
- Continuously we leverage the distribution of historical data of each customer to optimise the bids: increasing bids for sub-segments (e.g., MOBILE, TABLET and DESKTOP within DEVICE super-segment) that historically have generated more conversions/ clicks and reducing bids for those sub-segments that do not
- Safety (Deployment)
- We start with small bid increments to mitigate unwanted compound effects (using a normalising constant)
- We put several safety mechanisms in place to detect outliers in data (using a max bid adjustment threshold) or discover unsatisfactory data quality, and adjust accordingly
- We do test runs to measure the performance of the bid optimiser, using A/B testing
- Model Explainability
- MX is made up of a linear combination of variables and is therefore explainable to stakeholders and easy to debug for developers
Using machine learning and algorithms to leverage and monetise data assets: For our customers
We work with over 450 customers across the UK, Sweden and Norway and have delivered some fantastic results on their behalf. By analysing the Google search performance of a cohort of our customers over a 12-month period ending in Q1 2022, we found that they experienced an average 23% reduction in CPC and CPA (cost per customer acquisition) metrics, compared to a 17% increase in line with the market due to Google CPC inflation. Factoring in the price inflation, we calculated that our customers received 52% more value from Google search using BrightBid at the same advertising spend.