Introduction
Generative AI is revolutionising marketing strategy by enabling marketers to create highly personalised marketing content tailored to individual preferences. For marketing campaign outcomes, this means higher engagement, higher conversion rates and ultimately driving business growth
This case study explores how Kidron worked with a marketing agency and utilised generative AI to drive marketing strategies and achieve visible results for their clients, measured by conversions.
Here’s the approach that we took:
Scope of the project:
Our client wanted to optimise a seasonal marketing campaign to increase engagement and sales on their website and social media channels. The client has sales & engagement data from the past 10 years and we were tasked with analysing this data, identify any gaps in their marketing approach and look for ways to optimise their strategies and increase sales by at least 20%.
Collect, analyse and preprocess data:
We collected and analysed 10 years worth of data, including headlines, product descriptions and conversion rates. From theses analyses, we extracted key metrics such as:
Most frequent keywords used in the client’s headline and product description content
Correlations between specific keywords and conversion rates
Correlations between specific images/videos and conversion rates
After establishing the key drivers behind historical conversion rates, we now preprocess this data, using natural language and image recognition techniques with deep learning to serve as input for our generative AI models.
Model Building & Testing
We use pre-trained models specially developed to generate images and text for our use case. With our pre-processed data, we further train the pre-trained generative AI model. We fine tune the model so it can associate the correlation between text, images and conversion rates. Once trained on the client data we can then prompt the AI model to generate text and image content as well as the predicted conversion rates from each combination.
Now we have our Gen AI model which is capable of generating optimised marketing content for the seasonal campaign. The next step is to determine the absolute best content to deploy in the market. For this stage we generate various predicted high performing content from the model and deploy these against human content from the client which will be used as a baseline to measure performance. The content is optimised for the different marketing channels and deployed for a set test duration which allows us to reach adequate sample sizes for an A/B test analysis to determine which content is significantly better than the client content and then run the best content for the rest of the campaign duration.
Model Monitoring
On periodic intervals, we compare the conversion rates from the headlines and images created by our new model to historical content to detect any depreciation in performance. If performance depreciation is beyond set thresholds, we can generate new content to boost sales.
Comments