Tech Support Powered by AI

1

A support assistant that reduced response times by 87% using only verified documentation.

Smart home system manufacturer

CLIENT

INDUSTRY

IoT & Home Automation

8 Weeks

DURATION

The Challenge

1) Documentation Gaps

Documentation was fragmented across multiple product generations and sources. Support teams couldn't find accurate information quickly.

2) Hallucination Risk

AI models generate incorrect answers when lacking information. In technical support, wrong instructions could damage devices or create safety hazards.

The Solution

We implemented a RAG system that grounds every response in verified documentation.

Documentation Ingestion

Vectorised all documentation and manuals into a searchable semantic database.

Semantic Search

Vector similarity search retrieves relevant documentation for each query.

Constrained Generation

AI answers only from provided context, with clear handling when information is unavailable.

  • Embedded documentation using sentence transformers for semantic matching

  • Query → Semantic Search → Context Injection → LLM Generation

  • Response validation, citation tracking, and fallbacks for low confidence

Operational Impact

24/7 availability with instant responses

92% Customer Satisfaction

Business Impact

234K Euro annual cost savings due to reduced support staff overhead

Supply-Demand Tool by AI

2

A predictive supply-demand tool that aligns procurement with real-world sales, inventory, and lead times.

Electronics manufacturer

CLIENT

INDUSTRY

EV Mobility

DURATION

3 Weeks

The Challenge

1) Variant-Driven Complexity

The client produces multiple EV charger models, each selling at different proportions. Forecasting demand wasn’t just about total volume—it required understanding how variant mix affects component needs.

2) Fragmented Inventory Visibility

Inventory was spread across several categories: finished goods in stock, products ordered but not yet delivered, and components already ordered but still in transit. This made it difficult to assess true availability and plan ahead.

3) Lead Time and Procurement Uncertainty

Each component came with its own supplier lead time, and delays could ripple through the production schedule. The key question was: which components should be ordered, when, and in what quantities to meet future demand without overstocking or stalling production?

The Solution

We built a custom supply-demand planning engine that integrates real-time sales data, inventory status, and supplier lead times to generate precise procurement recommendations.

Constraints

Product variant sales proportions

Existing stock of finished goods and components

Ordered but undelivered products and parts

Component-specific lead times and supplier variability

Minimum order quantities and batch sizes

AI Tools

Probabilistic demand modeling to simulate variant-level sales

Constraint-based optimization for procurement timing and quantities

Rule-based logic for deterministic checks (e.g., safety stock thresholds)

Scenario simulation to test different sales and supply conditions

Integration

A modular interface for inputting assumptions and constraints

Visual outputs showing component needs over time

Editable forecasts and override options for planners

Exportable recommendations for procurement systems

  • We used historical sales data to estimate the probability distribution of product variant sales. This allowed us to simulate expected demand at the component level using Monte Carlo-style sampling.

  • The system reconciles four inventory states:

    1. Finished goods in stock

    2. Finished goods on order

    3. Components in stock

    4. Components on order

  • Each component has a lead time distribution (not just a fixed value). We modeled this using a triangular or normal distribution depending on supplier reliability, allowing for more robust planning under uncertainty.

  • We implemented a lightweight constraint solver that respects:

    • Minimum order quantities

    • Batch sizes

    • Lead time offsets

    • Production cadence

    • Safety stock buffers

  • For each planning cycle, the system runs a forward simulation to determine when each component will be needed, then back-calculates the optimal order date based on lead time and buffer.

  • Users can simulate different sales mixes, delivery delays, or stockout scenarios. This helps procurement teams stress-test their plans and make proactive decisions.

Operational Impact

90% less time in Excel

100% accuracy of the calculation

Business Impact

750K Euro annual cost savings due to inventory optimisation

24K Euro annual cost savings due to time saved in Excel