Building the AI Sales Intelligence Platform That Turned a $100B Engineering Firm's Product Data Into a Quote Conversion Engine
A $100B German engineering firm's product configuration data lived in nested XML files, disconnected from SAP and Salesforce and invisible to pricing teams. Ideas2IT built the data platform and AI intelligence layer that made it queryable, forecast demand across 100,000 products, and lifted quote conversion by 19%.
Client
Leading German Engineering Firm
Industry
All Industries
Service
Artificial Intelligence
Data Engineering
Outcome
19% quote conversion increase
Stack
AWS · Python · Custom ML
01 Challenge
Product configuration data for millions of electrical components sat locked in nested XML and HTML files, unqueryable and disconnected from SAP and Salesforce. Price analysts manually priced every quote. Demand forecasting for factory planning was absent. Sales teams had no real-time pipeline visibility or conversion intelligence.
02 Solution
Ideas2IT built a unified AWS data lake and Redshift warehouse that ingested SAP, Salesforce, transaction records, and XML configuration data into one governed layer. On that foundation, a custom parametric decision tree drove price prediction and AWS DeepAR powered 6-month demand forecasts across 100,000 products. A real-time sales intelligence dashboard put the output in front of every pricing and sales stakeholder.
03 Outcome
Quote conversion increased 19%. Discounts offered dropped by 1%. Demand forecasts now run across 100,000 products at country, region, state, and city levels. Price analysts shifted from manual quote work to high-value pricing decisions supported by an always-on ML recommendation engine.
Phase 01
Turning disconnected product data into a queryable foundation
Data foundation and product configuration retrieval: unifying SAP, Salesforce, and XML config data into a single queryable layer
Product configuration metadata for electrical components was stored in complex nested XML and HTML files accumulated over decades, structurally incompatible with any analytics layer.
Ideas2IT built
- AWS Glue pipelines in Python and Spark to parse, flatten, and load that configuration data alongside transaction records, SAP master data, and Salesforce quote history into an S3 data lake and Redshift warehouse.
- AWS Appflow handled Salesforce sync, keeping quote and pipeline data current without manual export cycles.
- A discount management workflow fed configuration data and pricing updates back into SAP via AWS Lambda and Step Functions.
Monitoring and notification pipelines flagged job failures before they propagated downstream.
Deliverables
- AWS Glue ETL pipeline (Python + Spark)
- Ingests transaction DB, SAP master data, and Salesforce records
- XML/HTML product configuration parser
- Flattens nested electrical component metadata for analytics
- S3 data lake and Redshift data warehouse
- Unified storage layer across all source systems
- AWS Appflow Salesforce integration
- Real-time quote and pipeline data sync
- Lambda + Step Functions discount workflow
- Pushes pricing and discount updates back to SAP
- Glue job monitoring and notification pipeline
- Event-triggered alerting on job failure
Phase 02
Replacing manual analyst pricing with ML-driven recommendations
Price prediction and demand forecasting: parametric ML and DeepAR replacing manual analyst effort across 100,000 products
Manual quote pricing was the bottleneck the business needed to break.
Ideas2IT built
- a custom parametric decision tree algorithm that encoded product hierarchies, geographic segments, and mandatory business rules into a model price analysts could trust. The algorithm produced price recommendations with prediction insights surfaced directly in the analyst workflow.
- Alongside it, AWS DeepAR ran demand forecasting for 100,000 products across four geographic levels, country, region, state, and city, projecting 6-month inventory requirements for factory planning.
- A Conversion Estimator layer ran probability scoring on open quotes, weighing geography, counterparties, and discount level to tell the pricing team which deals were worth pursuing at what price.
Deliverables:
- Custom parametric decision tree model
- Price prediction with product and geo hierarchy encoding
- AWS DeepAR demand forecast engine
- 6-month horizon across 100K products at 4 geo levels
- Conversion Estimator
- Quote conversion probability by geography, parties, and discount
- Optimal Discount Engine
- ML-driven discount recommendations for pricing analysts
- Market Basket Analysis model
- Cross-sell and upsell product recommendations
- NLP-based product component model
- Lambda/ECS pipeline for product catalogue intelligence
Phase 03
Putting AI-driven pricing intelligence in front of every stakeholder
Sales intelligence platform: real-time dashboards and self-serve analytics for pricing teams, engineers, and partners
The intelligence built in Phase 02 had no value without a delivery layer every business user could operate without engineering support.
Ideas2IT built a real-time Sales Intelligence Dashboard on Amazon QuickSight with Athena as the query layer, giving Sales Managers, Price Analysts, Field Sales Engineers, and Channel Partners a live view of pipeline trends, conversion probability, and discount performance.
Flexible filtering let teams slice data by product, geography, channel, and time period without writing queries. An embedded COMPAS dashboard surfaced configuration and pricing analysis for operational decisions. Self-service analytics gave business teams direct access to the Redshift data warehouse through QuickSight without BI tool dependency.
Deliverables:
- Real-time Sales Intelligence Dashboard
- QuickSight-powered, multi-role access for pricing and sales teams
- Amazon Athena query layer
- Self-service analytics over Redshift without SQL dependency
- COMPAS embedded dashboard
- Product configuration and pricing operational view
- Multi-dimensional pipeline view
- Filters by product, geography, channel, and time period
- Discount management self-service tool
- Business-operable discount setup with SAP sync
- Role-based access controls
- Scoped views for Sales Managers, Price Analysts, Field Engineers, Partners
The Outcome
What the platform produced across pricing, forecasting, and conversion
The 19% conversion improvement and 1% reduction in discounts offered were the direct consequence of connecting product configuration data to ML pricing models for the first time. The data foundation was the prerequisite: without XML configuration metadata in a queryable layer, parametric decision trees could not run on product hierarchies and DeepAR could not forecast at SKU level. The intelligence layer produced outcomes because the data architecture made the intelligence possible.