Skip to content
Utilizing AI to Boost Search Results and Take Customer Experience to New Heights

Hi-Tech Solutions

Improving Search Results
and Boosting CX with AI for
an International Tech Giant

Improving Search Results
and Boosting CX with AI for
an International Tech Giant

Success Story

An international technology leader partnered with Scalefocus to improve the search result sets of their main portal, including search result sets of their CPQ system. The goal was to shorten product allocation times for existing and new customers, as well as for their business partners, increase the accuracy of autocomplete suggestions, and provide smart recommendations and product configurations.

Our team used a hybrid model approach employing both collaborative and content-based filtering to achieve a robust suggestion quality. We enabled the system to create complex product predictions that improved customer experience through highly personalized results.

Improved search performance

Decreased
quote time

Smoother customer
journeys

The Client

Our client is a multinational information technology leader that provides software and hardware solutions and services to consumers, small and medium-sized businesses, and large enterprises. In addition, the company offers managed services, such as complete IT support solutions for other organizations.

Over the years, our client has made their name synonymous with innovation and impeccable quality and their diverse product portfolio aims to help businesses transform their IT and accelerate digital initiatives.

The Challenge

Our client wanted to boost the product search experience for different partners and stakeholders — these include end-customers, resellers and distributors, as well as business representatives. The goal was to significantly improve the relevance of results which would mean faster product allocation, decreased quote time, better customer experience and less frequent system overloads. For existing customers, the result optimization would be based on quote history. For new ones, the system would analyze the most quoted SKUs in the same country.

Since many tech products in an existing customer’s search history become obsolete relatively quickly, there also had to be a feature that offers a product similar to an out-of-date one. For example, if according to their quote history a customer looked at a certain device a year ago, the system would assume similarity of need and suggest the up-to-date version of the same item through feeding data to a data model with ML. Both existing and new partners need to explicitly state which products should remain within their quote, and only then proceed to finalize it. Once this has been completed, the selected products are listed in the quote as line items which concludes the CPQ process.

The Solution

We formed a team of experienced GO, Data and DevOps engineers whose main task was to create complex and useful product configurations for customers as opposed to providing a basic autocomplete feature. By employing AI models and ML algorithms, we wanted to avoid common and largely generic suggestions and instead allow for an extent of accurate and measured creativity. Thus, we would proactively anticipate a customer’s needs and expand on their search history, taking customer experience to the next level.

We also employed LLMs (Large Language Models), fine-tuning them based on client-specific data and incorporating an additional layer to summarize the exact parameters we needed from the whole data set. Once the models were trained, they began using a cached MM model to speed up the search and the prediction results. To prevent any quality issues in the process, our experts evaluated the outputs produced by the algorithm thoroughly and established an accuracy baseline. To avoid generating results outside of the data set and keep them domain-specific, we defined its limitation in the very beginning.

Our experts also developed a recommendation system responsible for auto-attaching additional products to the ones already in the results and integrated it with the products’ summary section, which resembles an e-commerce shopping cart. To achieve that, we utilized two specific ML algorithms — XGBoost and KNN, as well as Data engineering techniques. Essentially, we had to implement a smart autocomplete function that doesn’t simply predict the next word in a search but based on a starting point provided by the customer offers a smart configuration of adjacent products. Once the recommended products are in the summary section, the customer can decide to confirm them as line items or delete them as they are optional.

The Benefits

Our team of seasoned developers went beyond perfecting next-word predictions by having a deep understanding of how they work and then going a step further. There are already plenty of autocomplete engines, but we introduced AI, ML and LLMs to the equation to make more accurate, creative and complex predictions that are not seen in the client’s historical data.

We first focused on improving the search experience for the client’s existing partners through their quote history (most searched items plus a keyword) within a pattern called Google rate filtering. For new partners introduced through the client’s CPQ, we used backend APIs to predict the boosted product through our ML engine. We also developed a recommendation system that learns from users’ behavior, historical usage patterns and item attributes making intelligent predictions and suggestions to provide both simplicity and sophistication in the product search.

By employing algorithms and implementing feature engineering, we were
able to achieve recommendations tailored to all stakeholders’ unique needs,
as well as:
  • Enhanced Precision (percentage of relevant recommendations to total recommendations) from 53% to 89%
  • Enhanced Recall (percentage of recommendations retrieved from all relevant recommendations) from 54% to 90%
  • Improved search performance by speeding up the search process resulting in 95% accuracy
  • Streamlined architecture that is scalable and flexible
  • Stable performance and controlled costs
  • Automated background processes, monitoring and analytics
  • Improved overall product search experience for web, mobile and LMS

Technologies

GO
Python
Tensorflow
Elasticsearch
GraphQL
ReactJS
Redis Cache

AWS

OpenSearch
ECS
Lambda
SQS
CodeBuild
CodeDeploy
Cloudwatch
S3
CloudFront

Our Work

We have a global client base that includes Fortune 500 companies, innovative startups and industry leaders in Information Technology, E-Commerce, Insurance, Healthcare, Finance and Energy & Utilities.

Explore all

Ready to scale and meet the technology challenges of tomorrow?

Contact us