• 80% of monthly forecasts were outside range by >10%. This created cashflow and stock control issues.

• Ave 75hrs labour per month and 12d cycle time to forecast.

• Four relevant systems used for planning and forecasting. Typically each system’s and team leader’s output conflicted requiring further human checks and judgement.

• This caused delays, frustration, conflict, inefficiencies and waste. Also reduced sales due to stock shortages.

• Largely adopt our Consulting approach to fully understand the problem, root causes, goals and relevant data.

• Clean, structure and optimise data.

• Confirm our hypothesis to solve problem with AI approach.

• Rapid (<3 days) PoC to stakeholders. Use PoC feedback to iterate. Agile approach minimises risk, ensures tool evolves to meet the genuine user needs.

Develop PoC into robust, secure, scalable enterprise-ready system. Emphasis on integration, ensuring our solution seamlessly connected with existing forecasting finance and stock systems.

Predictive & probabilistic forecasting:

• Gradient Boosted Trees.

• Models for time series (Temporal Fusion Transformer, N-BEATS).

• Probabilistic models (Prophet, DeepAR) for uncertainty ranges.

Gen AI augmentation:

• Fine-tuned LLM to validate data, identify anomalies and ensure data quality.

• Daily, weekly, monthly agentic generation of sales reports & projections.

• Conversational interface for historical & projected sales data.

• Gemini, Mistral + LangChain to validate data anomalies and generate data quality summaries.

Stock & cashflow:

• Integer programming for inventory optimisation.

• Simmodels to balance cashflow, reorder points, and service levels.

• Integrate results back to ERP for automated reorder recommendations.

Decision intel & explainability:

• Summarise model outputs, highlight discrepancies, and suggest decisions in natural language.

• Power BI dashboards for end-user visibility.

• Accurate single source sales forecast (units and $) for use across multi functions, including sales, marketing, finance, CEO, procurement and ops.

• Forecasts now 85% accurate (from 20%).

• Labour average is 10hrs from 75hrs per month.

• Cycle times <1d, from 12d.

• Improved staff engagement, stock control, targeted marketing campaigns when forecasts look problematic.

• App developed to ensure real time data streams.

• Fully transparent and auditable by functional leaders to inputs, process, outputs, assumptions and trends.

• All data safely stored in client’s private AI infra tenant. This is arguably the company’s most sensitive data, so it is protected accordingly.