Performance Marketing World Unlocked: 1-2nd March London
BrightBid were lucky enough to present to the 1000+ performance marketing attendees at the PMW Unlocked event held at Excel in London in March:
We talked about GPT3, which was launched on the 20th of June 2020 and ushered in the first wave of AI startups including BrightBid: which launched in September 2020.
We used GPT3 as the underlying model for some of our automated AdCopy language translation into the 58 languages Google supports but are constantly evaluating all the new Large Language Models (LLM) such as LAMDA, PaLM and GPT4 to understand which LLM is driving the best results for Paid Search.
We moved to GPT4 for Ad Copy and language translation in March 2023 as it has significant improvements over language translation vs GPT3 and GPT3.5.
However, we don’t just use the LLM straight out of the box as we’ve also built an Unsupervised Machine Learning model on top. This is to ensure that we A/B test each AdCopy version and make sure it offers an improvement for our customers in terms of Click Through Rates in PPC.
We showed this graph comparing the numbers of brain synapses to the billions and trillions of parameters in the new LLMs to show that we’ve just moved from hedgehog to squirrel. LLMs have a long way to go before they hit human levels of synapses.
We then outlined how we’re using another unsupervised machine learning model in our Bid Automation building on top of Google and Bing’s Bid Automation processes and building different parameters to test around each element such as:
- Audiences
- Keywords
- Ad copy
- Competition
- Quality Score
- Account structure
- Performance Monitoring
- A/B testing
BrightBid’s AI engine
We discussed how we’ve built our model to the following factors:
1. Effectiveness (Accuracy)
We continuously 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.
2. Safety (Deployment)
Small bid increments to mitigate unwanted compound effects (using a normalising constant) with several safety mechanisms in place to detect outliers in data (using a max bid adjustment threshold). A/ B testing to measure bid optimizer performance.
3. Model Explainability
MX is made up of a linear combination of variables and is therefore explainable to stakeholders and easy to debug for developers
We then discussed how the AI model improves performance in Paid Search by generating more than 10 times the number of optimizations across all parameters that can be achieved by manual bidding alone. In our customers’ Watchcom’s case, we were able to shift to 17 times manual bidding. In 2021 only 678 bidding changes were made manually by the internal Watchcom team but in 2022 over 11,800 were made which vastly improved results.
We talked through how this is helping our customers get better value from Paid Search along 5 major axes:
-
- Reduction in Cost per Clicks/ Increase in Return on Ad Spend: Inflation in PPC has been running high over the last few years. Our AI engine helps marketing teams reduce CPC’s and increase their Return on Ad Spend by accelerating the amount of testing achievable on your Ads account to drive more revenue for lower spend.
- Auction Exclusivity: BrightBid’s AI engine uses all the data inputs from the Big Tech platforms whilst ensuring your business goals are prioritised. We have found that Big Tech AI is powerful but it tends to bid up the same ‘high-value searchers’ so if every competitor in the auction uses the platform AI it will inevitably lead to CPC inflation. BrightBid only works with one competitor in each auction meaning our AI is put to use for your business and not the category.
- International Expansion: Our AI engine uses the latest Large Language Models, like ChatGPT to translate Paid Search Ads into multiple languages and select the right keywords based on local demand. It then tests in each market to capture local nuances. This is a cost-effective way of handling local market PPC for multi-market advertisers.
- Model Explainability: Talking to customers who’ve previously used Google’s Smart Bidding platform: we’ve found that 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 or ‘wait for the AI to work’ which can be a long time in B2B with low data points. Our aim at BrightBid is to give our customers access to more controllable AI and automation solutions for paid 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 so we’ve built our models with linear combinations of variables and it is therefore explainable by our digital specialist team and to wider customer stakeholders. Effectively it helps unbox Google’s Black Box.
- PPC Expertise: Our expert digital specialist team offers a fully managed service, working with you to train the AI to deliver your exact business goals. Our product development is guided by a steering group of highly experienced PPC Subject Matter Experts who have managed some of the largest global search budgets and dedicated digital specialists assigned to each customer. This helps many marketing teams handle paid search: which can be a complex channel with expensive resources to recruit in house.
We then discussed the potential options for using AI for paid search:
- Platform solutions such as Google’s Smart Bidding which can work for some sectors with high numbers of data points and low numbers of your competition using Smart Bidding; Model Explaination as outlined above.
- Self Built hiring in AI scientists to work inhouse to build: great solution but very expensive as this is expensive talent in today’s tech world.
- LLM layers: using some of the copywriting tools, such as Jasper.ai or copy.ai to build your own AdCopy which can be a cost-effective solution for one aspect of your search account.
- Turnkey: an all-in-one solution like BrightBid’s that has employed the expensive AI experts and data scientists to build a turnkey solution for paid search and staffed up with digital specialists who can train the model, explain results and whose job it is to continually optimizing paid search delivery for our customers.
We then discussed how BrightBid can deliver a 50%+ improvement in business value from paid search
Our customers have seen an average of 23% reduction in CPC and CPA metrics over the last 12 months to Q1 2022, instead of 17% increase in line with the market At constant advertising spend, this translates into 50+% more business than the status quo.
Customers with no previous agency optimization have seen in excess of multiples of this improvement with significant benefits occurring after one month but accounts keep seeing improvements even after one year, with a goal of 50% CPC / CPA reduction.
Large benefits have been seen from optimized keywords reaching completely new customers in the long tail of searches, which is not possible through increasing the budget alone and optimized bidding per multidimensional cube segment ensures the right price is paid for the right value at every time.
Summary
PMW Unlocked gave the team great insight into what customers at the cutting edge of PPC performance marketing are thinking and needing and we’re continually building on our AI engine to incorporate these new ideas. We recently launched our AI engine for Bing Search to help marketing teams optimize PPC in this cost-effective channel.
Your mad not to use Bing if you’re using Google Ads for b2b lead marketing, here’s why