Czech financial savings financial institution Česká spořitelnaa division of Austria Erste Groupjust lately collaborated with AI options creator Delicate knowledge discover the usage of GenAI in name facilities. Česká needed to enhance high quality management and optimize prices in its inbound name heart operations, which obtain round 2 million calls a 12 months. They selected the Databricks Information Intelligence platform to experiment with inner and exterior AI fashions to judge the effectiveness of name heart brokers.
Exploring a top quality management system for customer support
The Česká spořitelna name heart staff needed to check a GenAI-powered high quality assurance system that ensured brokers adhered to written pointers throughout buyer interactions. A essential problem for Ceska was guaranteeing fixed communication with brokers for routine buyer inquiries. When prospects name about account balances, brokers ought to direct them to on-line banking options, a key enterprise requirement that drives digital adoption and operational effectivity. The assist staff wanted a scalable technique to confirm agent compliance and preserve communication requirements throughout 1000’s of buyer interactions. To attain this, the staff started utilizing Whispera speech-to-text mannequin from OpenAI, to precisely transcribe conversations. The problem was to provide human-readable textual content that precisely represented the spoken phrases utilized by name heart brokers with out distorting their which means. The transcripts needed to make logical sense and precisely mirror the intention of the dialog for later evaluation.
After transcription, the staff explored the mixing of inner GPT fashions and open supply fashions corresponding to Mixtral to judge their effectiveness. The GenAI fashions have been examined in a simulated QA perform, the place they have been tasked with answering particular questions corresponding to “Did the agent redirect the shopper to on-line banking?” The purpose of this train was to judge how properly these fashions might mimic human understanding and choice making by verifying compliance with established pointers. By evaluating the efficiency of each the in-house GPT mannequin and open supply fashions, the staff got down to discover the simplest answer to enhance customer support by way of AI-powered automated high quality assurance.
Advantages of the Databricks Information Intelligence platform for GenAI
The DataSentics staff evaluated a number of choices for this answer and finally selected to implement the Databricks knowledge intelligence platform and Mosaic AI instruments in Česká spořitelna for a number of causes:
- Information administration and governance advantages: Unit Catalog makes knowledge simply accessible to totally different fashions whereas maintaining delicate knowledge beneath restricted entry.
- Complete knowledge processing capabilities: The Databricks platform helps the complete name heart knowledge preprocessing workflow, from transcription to high quality management. This permits us to provide intermediate outcomes that may be leveraged for different fashions and initiatives, corresponding to advertising, threat evaluation, regulatory compliance, and fraud detection.
- Mannequin Coaching and Help: Databricks offers sturdy assist and experience for GenAI, together with mannequin structure and coaching capabilities. This made it a really perfect platform to shortly check and deploy open supply fashions, permitting us to experiment and iterate effectively.
- Ease of cluster creation: With Databricks, it is simple to create clusters and deploy open supply fashions. This streamlines the experimentation course of and permits us to focus extra on mannequin efficiency and fewer on infrastructure administration.
Outlook and outcomes
All through the venture, we experimented with varied segmentation methods and gathered a number of precious insights:
- The standard of the enter knowledge is essential: The standard of the audio recordings different from consumer to consumer, with some talking quietly or at a distance, which might then have an effect on the accuracy of the transcription. Whisper or comparable methods will help resolve the issue.
- The class definition is crucial: We realized that if classes can’t be simply outlined for people, it’s equally troublesome for LLMs to know them. This strengthened the necessity for clear and exact class definitions to coach fashions successfully.
- Open supply fashions ship outcomes: Open supply fashions demonstrated that they might compete successfully with proprietary fashions like ChatGPT. This discovering is essential for corporations in search of to optimize prices whereas reaching high-quality outcomes.
What’s subsequent?
With GenAI instruments powered by Databricks Mosaic AI, Česká spořitelna staff can now acquire entry to solutions present in quite a lot of paperwork by way of the “good search” perform. For instance, the buying staff might have to seek the advice of a whole bunch of pages of course of documentation on how one can management and approve funds to totally different international locations. Earlier than leveraging Databricks, it might take staff hours to seek out the precise info they want. Now, RAG-powered search offers staff with solutions in seconds, together with citations and hyperlinks to the supply doc.
Trying forward, there are numerous alternatives to discover extra GenAI workloads in Česká spořitelna. Our purpose is to create a robust integration between Databricks and name heart recordings from Česká spořitelna’s inner database. This can unlock new use circumstances corresponding to abandonment detection, sentiment evaluation and gross sales sign detection as Databricks is the go-to platform for streaming knowledge. These each day stories will permit Česká spořitelna to react to adjustments in actual time and on the identical time obtain value reductions with higher high quality assurance in its name facilities.
This weblog publish was co-written by Petra Starmanova (Česká spořitelna), tereza mokrenova (Sentics Information), Dalibor Karasek (Information Sentics) and Joannis Paul Schweres (Information Bricks).
