AI-Powered Call Intelligence System for Multi-Location Marketing Analysis

netsertive · Media & Entertainment · 2025 · source Agentic: no · promotional bias: high · source-backed · extracted with prompt 0.5.4 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: Multi-location call intelligence

Netsertive needed to turn high-volume customer call transcripts into timely marketing and service insights for multi-location brands, replacing slow manual review with automated analysis.

Constraints:

• customers can have high call volumes
• insights need to work inside an existing MLX platform
• customers may need context-specific prompts
• analysis must support location, regional, and national reporting
• cost and response time matter for real-time processing

Architecture

Netsertive routes call transcripts through the Lead API into Bedrock/Nova Micro, returns structured JSON analysis, stores results in Aurora, and reuses them for both real-time MLX views and scheduled aggregate reports.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  lead-api["Lead API"]
  bedrock-analysis-layer["Bedrock analysis layer"]
  standard-prompt["Standard prompt"]
  customer-specific-context["Customer-specific context"]
  structured-json-response["Structured JSON response"]
  aurora-results-store["Aurora results store"]
  real-time-mlx-views["Real-time MLX views"]
  aggregate-reporting-jobs["Aggregate reporting jobs"]
  lead-api --> bedrock-analysis-layer
  standard-prompt --> bedrock-analysis-layer
  customer-specific-context --> bedrock-analysis-layer
  bedrock-analysis-layer --> structured-json-response
  structured-json-response --> aurora-results-store
  aurora-results-store --> real-time-mlx-views
  aurora-results-store --> aggregate-reporting-jobs
  real-time-mlx-views --> lead-api
  aggregate-reporting-jobs --> aurora-results-store

Components

Data flow

Incoming transcripts and metadata go through the Lead API to Nova Micro, which returns structured JSON with sentiment, summaries, terms, themes, and coaching suggestions stored for real-time and aggregate reporting.

Why this architecture

Nova Micro and prompt customization were chosen because high-volume call intelligence needs fast, low-cost, context-specific analysis.

Models used: Amazon Nova Micro

Scale

• Manual call review previously took hours or days for high-volume customers; Netsertive evaluated the Bedrock/Nova approach in about one week.
  “only takes minutes to create actionable insights, compared to their previous manual call review processes, which took hours or even days for customers with high call volumes.”article, implied

  “Netsertive chose Amazon Bedrock and Amazon Nova Micro for their solution after a swift evaluation period of approximately 1 week of testing different tools and models.”article, implied
• Nova Micro was selected for generation speed above 200 tokens per second and low cost.
  “With its generation speed of over 200 tokens per second and highly performant language understanding skills, this text-only model proved ideal for Netsertive.”article, stated
• The model/tool evaluation period lasted about one week.
  “Netsertive chose Amazon Bedrock and Amazon Nova Micro for their solution after a swift evaluation period of approximately 1 week of testing different tools and models.”article, stated
• Manual call review that took hours or days can become minutes for high-volume customers.
  “only takes minutes to create actionable insights, compared to their previous manual call review processes, which took hours or even days for customers with high call volumes.”article, stated
• Aggregate analysis runs weekly and monthly for trend-level reporting.
  “The aggregate analysis process automatically initiates on both weekly and monthly schedules. During each run, the system gathers call data that falls within the specified time period.”article, stated

Alternatives considered (roads not taken)

One generic prompt for every franchise customer forever

Rejected because: The architecture keeps a standard base prompt but allows customer-specific context when franchise needs differ.

Only immediate per-call analysis

Rejected because: Stored Aurora results support weekly and monthly trend reporting alongside real-time views.

Real-time-only call intelligence

Rejected because: Netsertive stores analysis results so the same data supports immediate views and weekly/monthly aggregate reporting.

Customer-specific prompts in the first baseline for everyone

Rejected because: The architecture starts with a standard prompt and reserves customer-specific context as an added layer.

Patterns

structured-output-contract (model-ops)

The model returns a structured JSON object with predefined analysis fields so downstream reporting can store and aggregate the results.

cost-aware-routing (operations)

Netsertive selected Nova Micro for fast, low-cost call analysis rather than a larger model.

prompt-versioning (model-ops)

The architecture centers on a standardized base prompt but leaves room for customer-specific prompt customization.

batch-offline-processing (operations)

The same call-analysis foundation is reused for scheduled weekly and monthly aggregate reports.

Tradeoffs

Which model to use for production call analysis.

Chose Amazon Nova Micro over larger or slower model options tested during evaluation.

Rationale: The post says Nova Micro met Netsertive's need for fast, low-cost, consistent call insights.

How to support both immediate and trend-level insights.

Chose real-time call processing plus scheduled aggregate analysis over only ad hoc transcript review.

Rationale: Real-time processing supports immediate insights, while scheduled reports track performance over time.

Gotchas (what bit them)

Manual call review does not scale for high-volume franchise customers.

Root cause: Reviewing many calls by hand delays insights that could improve service and conversion.

Resolution: Process call transcripts automatically and store indexed analysis results for immediate and future use.

A single generic prompt may not fit every franchise customer.

Root cause: Multi-location brands can have different context, terminology, and reporting needs.

Resolution: Keep a standardized base prompt but design for customer-specific prompt context.

Evaluation

Netsertive evaluated tools and models for about a week, then created and tested prompts before integrating Amazon Bedrock into MLX and launching the feature in beta.

Metrics: approximately one week of model/tool evaluation · over 200 tokens per second generation speed · development completed within approximately 30 days before beta · analysis time reduced from hours or days to minutes

Outcomes

• Call Insights AI produces actionable call insights in minutes.
• Netsertive can extract sentiment, summaries, keywords, themes, and coaching suggestions from call transcripts.
• The feature became part of the MLX roadmap based on customer feedback and internal marketing expertise.

AWS co-authored customer story; the article describes beta launch and platform integration but does not provide independent accuracy scores.