Published by
Published on
October 7, 2026
By Prasanna Ragavan, Principal Data Scientist, Optilogic
What happens across your network — facility utilization, transportation cost, service risk — if you run a deeper promotion in Q3? What if you acquire a new brand? What if a competitor exits a region and demand suddenly shifts your way? These are questions every supply chain design team should be able to answer — but their planning systems were never built to solve.
Enterprise planning and execution systems are good at what they were designed for: running today's routes, fulfilling today's orders, managing this week's inventory. Many planning workflows operate in a 12–26 week window, and within that window, they work.
But supply chain design asks different questions: what should the network look like one or two years from now, across a range of strategic scenarios, at a level of granularity that drives real location, capacity, and sourcing decisions. Planning systems don't answer those questions. They weren't built to.
The result: many network design teams work off static demand assumptions: last year's numbers, broad regional aggregates, or forecasts from a planner who's already told them they can't go beyond 20 weeks. Strategic decisions get made without ever stress-testing demand. What teams assume and what actually happens are rarely the same thing, and the difference shows up as wrong capacity, wrong inventory, wrong sourcing decisions, spot-rate premiums, and service failures.
One major consumer goods company was running 40,000 time series forecasts that took six hours to complete, and still couldn't see beyond 12 weeks. A proof of concept with Optilogic's demand modeling engine cut that runtime to 15 minutes and extended the forecast horizon to a full year, at SKU-level granularity, giving the design team realistic demand scenarios they could actually run through network decisions.
That's exactly the problem demand modeling for supply chain design is built to solve.
Prasanna Ragavan, who has spent more than 10 years in supply chain planning, forecasting, and demand, describes the problem in three parts.
Your planner may have a 20-week forecast. Your network design project needs 52 weeks. What happens to the remaining 32 weeks? Most teams use assumptions, extrapolations, or last year's data, none of which reflect current market dynamics, growth trajectories, or the strategic scenarios you're evaluating. The issue isn't just filling those missing weeks. It's filling them with something defensible: demand scenarios believable enough to hold up when stakeholders ask why they should trust the recommendation.
Commercial planning often rolls demand up by sales region, channel, account, or category. Supply chain design needs that translated into shipping lanes, DC territories, zip clusters, and facility-service relationships. A beverage company deciding whether to open a new DC in the Southeast doesn't need to know that "the South" will grow 12%; they need to know how many cases are moving to which zip codes, and what happens to that pattern if they run a regional promotion or change a co-packer.
Sales teams are often looking at total volume, revenue, and account growth. Supply chain decisions need the next layer down: which product, which location, which lane, which facility, and which time period.
Even when horizon and granularity are managed, the deeper problem is decision preparedness. Supply chain teams should be able to walk into a stakeholder meeting with scenarios already tested, not scrambling to answer the obvious questions after the fact. Most planning platforms weren't designed to quickly model what happens if you run a deeper promotion, integrate an acquired brand, or respond to a competitor exiting a key region. Each change typically requires a lengthy reconfiguration project, if it's possible at all.
Related Reading: Optimize the Supply Chain You Have Today — Optilogic
To make demand useful for network design, teams need algorithms and workflows capable of handling business context, scenario drivers, uncertainty, and repeatable data preparation. Here's what that looks like in practice, and how Optilogic supports it.
A food and beverage company might have product categories (beverages, snacks, dairy), product families, individual SKUs, and location hierarchies from country down to individual distribution territory. The intersections create thousands, sometimes millions, of time series that need forecasting.
The challenge isn't choosing top-down versus bottom-up forecasting. It's learning where the signal is strongest across the hierarchy and reconciling it into one usable demand view.
Optilogic's hierarchy intelligence layer:
At the SKU-location level, demand can be extremely noisy; a single customer order or one-time promotion event can create patterns that don't repeat. At the product-family or regional level, those fluctuations smooth out. Knowing where to generate the forecast, and how to reconcile across levels, is often the difference between a usable output and one that sends the modeling team back to their spreadsheets. The value is a demand view that design teams can actually build scenarios around.
In many environments, statistical models, machine learning, causal factors, and scenario workflows are separated or require heavy configuration to work together.
Pulsar, Optilogic's demand engine, provides the following in a single unified workflow:
The output isn't a single best number. It's a scenario-ready demand framework that connects hierarchy, uncertainty, and business drivers to the network decisions being tested.
This is where most demand modeling projects fail, before the algorithms even run.
The data reconciliation problems are real:
Most open-source forecasting tools leave all of this to the user. Optilogic handles it through an automated, configurable workflow: aggregation, disaggregation, time-period alignment, and missing value treatment, all in a unified pipeline, without rebuilding the workflow from scratch every time.
When supply chain designers can run forecasts in minutes instead of hours, the entire workflow changes. Instead of one set of assumptions baked in weeks ago, they can test six or seven scenarios before a meeting, and walk in already prepared with evidence-backed answers for the questions stakeholders are going to ask.
These aren't theoretical benchmarks. They're outcomes from actual customer engagements, including work with a large retailer managing tens of thousands of weekly delivery routes, optimizing labor, fleet capacity, and vendor contracts simultaneously.
The business impact of a 5–7% WMAPE reduction across delivery territories:
Most forecasts give you a single number: "We expect 12,000 units next quarter." That may be sufficient for some planning workflows.
Network design needs more. It needs to know what changes under different demand scenarios: P10, P50, P90, peak seasons, promotions, product transitions, and regional shifts. A single point estimate can't answer those questions.
Optilogic's demand engine includes probabilistic forecasting that quantifies uncertainty around each forecast. Crucially, it recognizes that uncertainty is not symmetric and not constant over time.
Two very different risk profiles, same manufacturer:
Optilogic surfaces this by SKU and by location:
"The past is not directly informative of the future. Yes, it informs the future in a way, but it is not indicative of the future. Probabilistic forecasting tells you where to hold inventory — more, or less — based on future variability, not just historical patterns." — Prasanna Ragavan, Demand Modeling Lead, Optilogic
This matters for inventory design in particular. As research from Gartner consistently shows, safety stock calculations based on historical variability systematically misrepresent future risk, especially in volatile markets. Probabilistic forecasting gives supply chain designers a defensible, forward-looking basis for those decisions.
The most important shift isn't producing a better forecast. It's what happens after: demand scenarios that flow directly into network design, optimization runs, scenario comparison, and executive discussion. Demand shouldn't stop as a forecast output. It should drive the decisions that follow.
When demand scenarios connect directly to network optimization, the conversation shifts from debating a static forecast to comparing how different futures affect cost, service, capacity, and resilience.
Network design decisions don't stay static:
The companies that respond fastest aren't the ones with better data. They're the ones who can rapidly re-evaluate demand assumptions and understand how they cascade through the network. According to a July 2026 Gartner survey, 72% of supply chain leaders had to revisit final approvals for network investment decisions at least once, a pattern Gartner attributes to organizations treating network decisions as fixed rather than adaptable.
Optilogic is designed to close that loop. Optilogic’s Demand Modeling connects directly to Cosmic Frog optimization models, so a revised demand scenario doesn't just produce a new forecast; it immediately powers new network optimization runs, scenario comparisons, and executive-ready insights through the Executive Insights App.
The result is a supply chain design process that isn't locked in time. It's continuous, scenario-driven, and connected to the operational reality of where demand is actually going.
Demand modeling for supply chain design is not just a data science project. It is not just about producing a better number either. It is about realism: building demand scenarios that help teams pressure-test decisions upfront, anticipate stakeholder questions, and come prepared with evidence-backed answers before committing to a network that must hold up against more than one version of the future.
Ask yourself:
If the answer to any of these is no, the cost is real, even if it's invisible in your current reporting.
By Prasanna Ragavan, Principal Data Scientist, Optilogic
What happens across your network — facility utilization, transportation cost, service risk — if you run a deeper promotion in Q3? What if you acquire a new brand? What if a competitor exits a region and demand suddenly shifts your way? These are questions every supply chain design team should be able to answer — but their planning systems were never built to solve.
Enterprise planning and execution systems are good at what they were designed for: running today's routes, fulfilling today's orders, managing this week's inventory. Many planning workflows operate in a 12–26 week window, and within that window, they work.
But supply chain design asks different questions: what should the network look like one or two years from now, across a range of strategic scenarios, at a level of granularity that drives real location, capacity, and sourcing decisions. Planning systems don't answer those questions. They weren't built to.
The result: many network design teams work off static demand assumptions: last year's numbers, broad regional aggregates, or forecasts from a planner who's already told them they can't go beyond 20 weeks. Strategic decisions get made without ever stress-testing demand. What teams assume and what actually happens are rarely the same thing, and the difference shows up as wrong capacity, wrong inventory, wrong sourcing decisions, spot-rate premiums, and service failures.
One major consumer goods company was running 40,000 time series forecasts that took six hours to complete, and still couldn't see beyond 12 weeks. A proof of concept with Optilogic's demand modeling engine cut that runtime to 15 minutes and extended the forecast horizon to a full year, at SKU-level granularity, giving the design team realistic demand scenarios they could actually run through network decisions.
That's exactly the problem demand modeling for supply chain design is built to solve.
Prasanna Ragavan, who has spent more than 10 years in supply chain planning, forecasting, and demand, describes the problem in three parts.
Your planner may have a 20-week forecast. Your network design project needs 52 weeks. What happens to the remaining 32 weeks? Most teams use assumptions, extrapolations, or last year's data, none of which reflect current market dynamics, growth trajectories, or the strategic scenarios you're evaluating. The issue isn't just filling those missing weeks. It's filling them with something defensible: demand scenarios believable enough to hold up when stakeholders ask why they should trust the recommendation.
Commercial planning often rolls demand up by sales region, channel, account, or category. Supply chain design needs that translated into shipping lanes, DC territories, zip clusters, and facility-service relationships. A beverage company deciding whether to open a new DC in the Southeast doesn't need to know that "the South" will grow 12%; they need to know how many cases are moving to which zip codes, and what happens to that pattern if they run a regional promotion or change a co-packer.
Sales teams are often looking at total volume, revenue, and account growth. Supply chain decisions need the next layer down: which product, which location, which lane, which facility, and which time period.
Even when horizon and granularity are managed, the deeper problem is decision preparedness. Supply chain teams should be able to walk into a stakeholder meeting with scenarios already tested, not scrambling to answer the obvious questions after the fact. Most planning platforms weren't designed to quickly model what happens if you run a deeper promotion, integrate an acquired brand, or respond to a competitor exiting a key region. Each change typically requires a lengthy reconfiguration project, if it's possible at all.
Related Reading: Optimize the Supply Chain You Have Today — Optilogic
To make demand useful for network design, teams need algorithms and workflows capable of handling business context, scenario drivers, uncertainty, and repeatable data preparation. Here's what that looks like in practice, and how Optilogic supports it.
A food and beverage company might have product categories (beverages, snacks, dairy), product families, individual SKUs, and location hierarchies from country down to individual distribution territory. The intersections create thousands, sometimes millions, of time series that need forecasting.
The challenge isn't choosing top-down versus bottom-up forecasting. It's learning where the signal is strongest across the hierarchy and reconciling it into one usable demand view.
Optilogic's hierarchy intelligence layer:
At the SKU-location level, demand can be extremely noisy; a single customer order or one-time promotion event can create patterns that don't repeat. At the product-family or regional level, those fluctuations smooth out. Knowing where to generate the forecast, and how to reconcile across levels, is often the difference between a usable output and one that sends the modeling team back to their spreadsheets. The value is a demand view that design teams can actually build scenarios around.
In many environments, statistical models, machine learning, causal factors, and scenario workflows are separated or require heavy configuration to work together.
Pulsar, Optilogic's demand engine, provides the following in a single unified workflow:
The output isn't a single best number. It's a scenario-ready demand framework that connects hierarchy, uncertainty, and business drivers to the network decisions being tested.
This is where most demand modeling projects fail, before the algorithms even run.
The data reconciliation problems are real:
Most open-source forecasting tools leave all of this to the user. Optilogic handles it through an automated, configurable workflow: aggregation, disaggregation, time-period alignment, and missing value treatment, all in a unified pipeline, without rebuilding the workflow from scratch every time.
When supply chain designers can run forecasts in minutes instead of hours, the entire workflow changes. Instead of one set of assumptions baked in weeks ago, they can test six or seven scenarios before a meeting, and walk in already prepared with evidence-backed answers for the questions stakeholders are going to ask.
These aren't theoretical benchmarks. They're outcomes from actual customer engagements, including work with a large retailer managing tens of thousands of weekly delivery routes, optimizing labor, fleet capacity, and vendor contracts simultaneously.
The business impact of a 5–7% WMAPE reduction across delivery territories:
Most forecasts give you a single number: "We expect 12,000 units next quarter." That may be sufficient for some planning workflows.
Network design needs more. It needs to know what changes under different demand scenarios: P10, P50, P90, peak seasons, promotions, product transitions, and regional shifts. A single point estimate can't answer those questions.
Optilogic's demand engine includes probabilistic forecasting that quantifies uncertainty around each forecast. Crucially, it recognizes that uncertainty is not symmetric and not constant over time.
Two very different risk profiles, same manufacturer:
Optilogic surfaces this by SKU and by location:
"The past is not directly informative of the future. Yes, it informs the future in a way, but it is not indicative of the future. Probabilistic forecasting tells you where to hold inventory — more, or less — based on future variability, not just historical patterns." — Prasanna Ragavan, Demand Modeling Lead, Optilogic
This matters for inventory design in particular. As research from Gartner consistently shows, safety stock calculations based on historical variability systematically misrepresent future risk, especially in volatile markets. Probabilistic forecasting gives supply chain designers a defensible, forward-looking basis for those decisions.
The most important shift isn't producing a better forecast. It's what happens after: demand scenarios that flow directly into network design, optimization runs, scenario comparison, and executive discussion. Demand shouldn't stop as a forecast output. It should drive the decisions that follow.
When demand scenarios connect directly to network optimization, the conversation shifts from debating a static forecast to comparing how different futures affect cost, service, capacity, and resilience.
Network design decisions don't stay static:
The companies that respond fastest aren't the ones with better data. They're the ones who can rapidly re-evaluate demand assumptions and understand how they cascade through the network. According to a July 2026 Gartner survey, 72% of supply chain leaders had to revisit final approvals for network investment decisions at least once, a pattern Gartner attributes to organizations treating network decisions as fixed rather than adaptable.
Optilogic is designed to close that loop. Optilogic’s Demand Modeling connects directly to Cosmic Frog optimization models, so a revised demand scenario doesn't just produce a new forecast; it immediately powers new network optimization runs, scenario comparisons, and executive-ready insights through the Executive Insights App.
The result is a supply chain design process that isn't locked in time. It's continuous, scenario-driven, and connected to the operational reality of where demand is actually going.
Demand modeling for supply chain design is not just a data science project. It is not just about producing a better number either. It is about realism: building demand scenarios that help teams pressure-test decisions upfront, anticipate stakeholder questions, and come prepared with evidence-backed answers before committing to a network that must hold up against more than one version of the future.
Ask yourself:
If the answer to any of these is no, the cost is real, even if it's invisible in your current reporting.
Fill out the form to unlock the full content
.jpg)
