NoiseFront Tally

CONVERT CUSTOMER CALL RECORDINGS INTO INSIGHTS

Over the last year I've been cooking up this new service - here's the introductory blog post! 

Millions of customer calls are recorded for quality assurance every day

  • Only a small fraction of them is reviewed, listened to and analyzed. Manual review is reserved for contested recordings due to resourcing constraints.
  • Very little insight is distilled from actual customer interactions. Post-call surveys and self-reporting by agents are unreliable and subjective. 
  • Market studies capture a distorted view of the reality under idealized conditions.

NoiseFront Tally’s AI transcribes and analyses thousands of calls every hour

  1. Retrieving call records from our integration partners and monitor for new ones.
  2. Transcribing the calls as soon as they are retrieved and verifying language and accuracy.
  3. Analyzing the transcripts using natural language on-premise GPT implementation.
  4. Delivering accurate insights through statistical methods and multi-tiered queries.

Tally builds on years of research in audio surveillance and remote sensing

  • NoiseFront infrastructure was built in support of defense industry surveillance and remote sensing requirements
  • Our unique audio processing and retrieval stack matches large call center and telecommunication needs
  • Integrated and scalable solution allows NoiseFront to offer competitive pricing without the need for multi-party contracting

Our system performs the following tasks for each call:

  1. Retrieval. 1-3s per call.

    • Supported API include CallRail, Cisco, Five9
    • Historic data backfill is a key differentiation from competing solutions
    • Customizable calls selection to maximize speed and cost efficiency
    • Manual uploads for calls on unsupported platforms. Our team would be delighted to integrate upon request.
  2. Transcription. 3-5s per call.
    • Automatic speaker determination is a valuable differentiator for multi-party receptions and call transfers
    • Conversion to text is multi-lingual with focus on US English; Spanish and other languages supported and flagged with decreased accuracy
    • Automatic and manual accuracy verification feeding into spot-checks
  3. Analysis. 3-7s per call.
    • Automatic AI analysis featuring custom-designed prompts specific to customer business
    • AI errors and hallucination tightly controlled through both self-determination of confidence scores and manual spot-checks
    • On-premise high-performance GPT implementation 
  4. Insights. Real-Time
    • Multi-tiered statistics and customized dashboards 
    • Dependent queries focusing on key customer subsets, grouping and differential analysis
    • Statistically significant slicing of data for cost and time savings, achieving preliminary results in fractions of total runtime.

Our AI has listened to over 1M minutes of calls in 2024

Our proprietary scalable infrastructure allows each dialogue to be queried using natural language with results in seconds. Examples of questions you may ask:

  • “Was the price discussed on the call?”
  • “Did our agent recommend a competing product?” 
  • “Was the customer dissatisfied with the service received?”

Check it out: 

NoiseFront Tally