The future of S&OP: How AI is revolutionising supply chain planning for paper & packaging Jari Kaukiainen, Advisor Integrated Business Planning and SCP, I Plan

PAPERTECHNOLOGYINTERNATIONAL
1. Demand Planning & Demand Reprocessing
Non-generative ML use case:
XGBoost-based demand forecasting.
By applying tree-based ensemble models to
historical order data, companies can identify
complex interactions between variables (e.g.,
promotions, seasonality, customer segments)
and improve forecast accuracy by 15–30%
versus traditional time series models.
Real-time demand adjustment
using supervised learning (e.g., XGBoost
or Random Forest). ML models can
continuously learn from incoming short-term
signals like:
O POS Data O eCommerce transactions
O Weather changes O Promotion lift
variances
O Macroeconomic shifts (e.g., CPI
spikes)
These models reprocess the original baseline
forecast and update short-term SKU-level demand across
the horizon. This is particularly useful when:
O A sales campaign underperforms or overperforms
O A competitor launches a surprise product
O An unexpected event (e.g. sports event win,
regulatory news) boosts specific category demand
As an example a corrugated packaging producer
sees a spike in demand for speci昀椀c ecommerce boxes
due to Black Friday week trending stronger than predicted.
The ML model reprocesses recent demand signals,
identifies the new baseline, and adjusts replenishment
recommendations for key box SKUs in real time.
Simulation use case:
Scenario simulation for product launches.
Companies simulate multiple demand curves for new SKUs
using historical analogs, allowing planners to stress-test capacity
and inventory positions before launch.
Simulation models can test how different demand
reallocation scenarios affect the supply chain in real time. This
includes:
O Allocating constrained raw materials or machine time
O Reprioritising orders across regions or customers
O Assessing lead-time trade-offs for urgent replenishments
A pulp and paper company experiences an unexpected
demand shift from Southern to Northern Europe. Simulations
assess:
O How to redirect stock-in-transit o Which production lines can
accommodate the new mix o The impact on service levels and
logistics costs
The simulation provides planners with options ranked by
feasibility and cost impact, allowing them to reprocess demand with
operational realism.
2. Inventory Planning
Non-generative ML use case:
•
Predictive safety stock modelling. ML models ingest
lead time variability, supplier reliability, and historical
stockouts to dynamically recommend safety stock
levels by SKU-location, reducing inventory by up to
20% while maintaining service levels.
Simulation use case:
• Multi-echelon inventory simulations. Simulates how
inventory flows and buffers behave across the network
under different demand and disruption scenarios,
supporting optimal positioning of stock.
3. Production Planning
Non-generative ML use case:
• Predictive maintenance for production scheduling.
Using ML to predict machine downtimes enables
better sequencing and resource planning, reducing
unplanned delays and enabling more reliable execution
of production plans.
Simulation use case:
• Finite capacity production simulations. These simulate
the impact of production constraints (machine, labor,
changeovers) under different plans, helping planners
choose the most feasible and cost-efficient schedule.
4. Network Planning
Non-generative ML use case:
• ML-driven cost-to-serve models. These models
analyse order patterns, shipping costs, and customer
behaviours to recommend optimal regional distribution
points and transportation modes.
Simulation use case:
• Network design simulations. Planners model the effect
of adding/removing warehouses, changing production
locations, or rerouting transport to evaluate trade-offs in
cost, lead time, and service.
5. S&OP Scenario Planning
Non-generative ML use case:
• Revenue/margin optimisation. ML models evaluate
product mix scenarios against historical sales elasticity,
pricing sensitivity, and cost structures to recommend
the optimal scenario in S&OP meetings.
Simulation use case:
• S&OP scenario war-gaming. Teams simulate different
demand and supply-side scenarios (e.g., raw material
shortage, spike in demand) across a 12–18 month
horizon to stress-test financial and operational KPIs.
76

Paperturn flipbook viewer