53% boost in add clicks after implementing AI-based marketing tools

Innowise has developed an analytics platform based on AI and machine learning that helps to match user queries with the most relevant ads.

Customer

Industry

Sales & Marketing

Region

Client since

2022

Our client is a performance-focused online marketing agency offering advertising campaigns, content creation, and SEO services that aim to increase qualified leads and transactions for their clients.

Detailed information about the client cannot be disclosed under the provisions of the NDA.

Challenge: Overcoming the lack of keywords-coverage with AI in digital marketing

As digital advertising continues to evolve, users may be overwhelmed with the abundance of options. Despite this, online marketing agencies still struggle to reach their target audience with relevant product recommendations at the right time based on user queries.

Our client faced a fundamental problem with an under-optimized advertisement recommendation system that failed to offer search engine ads that matched users’ needs. During the advertising efforts, the agency faced a number of significant challenges: about 30-40% of needed search engine user requests were not covered by relevant advertisements. Additionally, a large number of existing ads were irrelevant due to poor matching with user queries.

The root cause of the relevance issue was the lack of coverage of relevant keywords and assets by the existing advertising system, which affected user clicks and ad campaign performance. The client’s existing platform provided inadequate analytics, making it difficult to correct relevance problems and identify the causes of poorly covered requests.. The number of unmatched or irrelevant matched requests was too large for detailed data consideration and identification of local low-performing ad causes.

To address these issues, our customer approached Innowise for advanced analytics and summary generation for clustered subgroups of user queries, which would drive smarter insights. The client approached Innowise with the idea of advanced analytics and summary generation for clustered subgroups of user queries to drive smarter and better insights.

In summary, the scope of work included:

analyzing and clustering users’ queries;
identifying target groups of users and their features to improve ads recommendation results;
specifying the most relevant ads for the queries clusters;
identifying the gaps in existing advertisements by analyzing the queries clusters that were poorly covered with existing ads.

Solution: AI-based marketing tools for optimized ad campaigns

Our team successfully completed the project and developed an advertising campaign analytics platform with a keyword ranking analyzer using recently appeared SOTA natural language processing models. The entire neural network was deployed in AWS cloud.

The platform is integrated with Google and allows working with statistical data on user queries, identifying uncovered requests or those with ineffective advertisement, breaking them into subgroups, and generating summaries for certain categories of large amounts of data to adjust the displayed ads.

Our team has developed the solution to replace the previous system that provided only basic statistics and lacked the ability to quickly analyze advertising results and adjust matching based on revealed insights.

Clustering and summarizing user requests with keyword ranking analyzer tool

Based on the customer’s requirements, we collected Google analytics data on user queries with undisplayed ads. Our specialists configured a system to analyze these queries and cluster them using semantic embeddings from BERT family models and different clustering techniques such as hdbscan, dbscan, T-SNE, KMeans. The web application also allowed for the collection of aggregate statistics across a pool of user requests. Depending on the selected granularity level, we also made it possible to gather aggregation statistics for a pool of user requests and produce summaries for each distinct group. We used BERT, basic statistical tools, and topic modeling to display a cloud of tags with the most popular terms in a particular group of queries. Users were also able to obtain a GPT model generated summary based on specified clusters.

Smart analysis and clustering of user requests with irrelevant advertisements

The platform we developed allows for the display of user interactions with specific ads, enabling the identification of irrelevant ads matched with inappropriate queries by analyzing interaction data. Through the use of extensive statistics, tags, and summaries of specific searches with low-performing results, it is now possible to determine the reason for differences with user interests and displayed ads. This platform feature is an essential tool for identifying and filling the gaps in existing advertisements for target groups of users and their features.

Matching uncovered user queries with the most relevant ads

By using AI and ML tools, the platform offers potential ad matches for groups of queries that previously had no relevant ads. We achieved this by generating textual representations of query clusters and creating ads by specifying the most relevant ones for each cluster using similarity scores from transformer models. Additionally, we customized these ads for specific groups of users by performing prompt engineering on GPT family models to create more relevant and engaging ads tailored to their specific interests. Using the data displayed on existing queries on the dashboard, the system determines and generates relevant ad options for certain query segments. This approach allowed us to determine which current ads can be linked to user requests that were previously unmet and reveal latent demands for future ad generation or correlate such requests with ready-made advertising that suits them the most.

Technologies

Platforms

AWS

Front-end

React, Redux, HTML5, CSS3, Formik, Yup, Material UI

Back-end

Python 3.x, Flask (microservices), Flask-restful, Celery, RabbitMQ

AWS S3 PostgreSQL, AWS Sagemaker (Pipelines, Feature Store), AWS Glue PySpark, Spark AWS Airflow

DS, ML & MLOps

AWS Sagemaker (Studio, Experiments, Notebooks, AutoML, Model Monitoring), Scikit-learn, Matplotlib, BERT, Pandas, Numpy

Process

Upon receiving a client request, our team identified the main potential use cases for obtaining advanced and visual analytics by clustering information from Google Analytics. We then obtained a large amount of data on user queries and interactions with displayed ads.

Our first step was to cluster the information into smaller subgroups based on the keywords entered by users in the search string. We used generative models like GPT to create textual representations for each clustered data group. The resulting summaries were displayed on the platform to provide detailed information about uncovered queries or queries with poorly performing ads, allowing for a better understanding of the reasons for irrelevance and subsequent adjustments to the ads.

The next step was suggesting matching the most relevant ads to match with uncovered queries as closely as possible to improve performance. We searched for ads from a list of written ads that covered as many queries as possible to fill the gaps and to create suggestive summaries for potential matches.

As for project management, we adhered to Agile methodology with daily meetings to discuss completed and planned tasks and bi-weekly calls with the CEO. Our team communicated via Slack and assigned tasks and monitored performance through Jira and Confluence.

Currently, the project is still ongoing; at this stage, we continue to support the platform and implement new features.

Team

Project Manager

Data Engineers

Back-End Developers

Front-End Developers

Data Science Engineers

QA Engineers

Results: Boost in user clicks on ads with AI-based keyword ranking analyzer

We have built an AI-powered that provides our customer with more relevant and target-oriented advertisements by recognizing a group the user belongs to and using this information to drive smarter and better insights for advertising personalization. The web application analyzed runned ads campaigns and found gaps in the advertising, which prevented our client from reaching all the necessary users.

Moreover, the solution can now automatically generate new advertisements, optimizing the company’s copywriting processes.

Overall, the platform has resulted in a 53% increase in user clicks on ads. We also created recommendations for copywriters based on the most dense and largest clusters, enabling them to create ads that cover up to 92% of necessary user requests. We continue to explore the power of AI in digital marketing while developing additional AI based marketing tools to continue platform improvement.

Project duration

March 2022 - Ongoing

25%

time saved on generating new ads

53%

boost in add clicks

Related cases

Contact us!

Book a call or fill out the form below and we’ll get back to you once we’ve processed your request.

What happens next?

Having received and processed your request, we will get back to you shortly to detail your project needs and sign an NDA to ensure the confidentiality of information.

After examining requirements, our analysts and developers devise a project proposal with the scope of works, team size, time, and cost estimates.

We arrange a meeting with you to discuss the offer and come to an agreement.

We sign a contract and start working on your project as quickly as possible.