Mastering Volatile Demand Forecasting with LikwidERP

Forecasting Volatile Demand: The Practical Reason SMEs Should Replace Legacy ERP Now Volatile demand used to be “a bad quarter.” Now it’s the operating environment.

Forecasting Volatile Demand: The Practical Reason SMEs Should Replace Legacy ERP Now

Volatile demand used to be “a bad quarter.” Now it’s the operating environment. If you’re running on a legacy ERP, that volatility turns into a predictable pattern: slow monthly forecasting cycles, spreadsheet workarounds, brittle MRP parameters, and last-minute expediting that drives both stockouts and excess inventory.

Modern ERP replacement isn’t just an IT refresh—it’s an opportunity to redesign how forecasting, replenishment, and S&OP/IBP work when volatility is permanent. Macro signals show the shift is already underway. Gartner forecasts worldwide IT spending will reach $5.61 trillion in 2025 (+9.8% YoY), driven in part by GenAI-related upgrades—creating urgency for SMEs to modernize planning workflows inside and alongside ERP (Gartner press release).

Why legacy ERP makes volatility worse

Legacy ERP implementations often “freeze” planning into a monthly cadence because that’s what the system and governance can tolerate: batch imports, manual overrides, and static safety stocks. When demand swings weekly—or daily—those assumptions break.

• Slow reforecasting: Monthly cycles miss demand inflections, leaving planners to react with expedites.
• Brittle parameters: Fixed reorder points, safety stock, and lead times don’t adapt as variability increases.
• Disconnected decisioning: Finance, sales, and ops carry different numbers because the forecast isn’t a governed system-of-record.

Meanwhile, buyers are explicitly moving toward AI-supported planning during ERP replacement. Panorama Consulting reports organizations with significant or moderate AI deployment rose from 53.4% (2024) to 72.6% (2025)—a clear sign that forecasting and planning capabilities are becoming table stakes (Panorama Consulting 2025 ERP Report (PDF)).

What “volatility-ready forecasting” looks like in a modern ERP stack

Volatility-ready forecasting is less about finding a single perfect model and more about building a fast, governed loop: detect changes, reforecast quickly, evaluate tradeoffs, and push decisions into execution (MRP, purchasing, production, and available-to-promise).

1) Forecast when history is no longer representative

If promotions, stockouts, channel shifts, or product changes have made history “dirty,” start by separating demand signals:

• Baseline demand: What demand would have been without interventions (promos, constraints).
• Lifts and events: Promo uplifts, channel launches, and one-off deals treated as add-ons, not baked into baseline.
• Constrained vs. unconstrained demand: Stockouts hide true demand; capture lost sales and treat as censored observations where possible.

In practice, SMEs can use an automated/ensemble approach (multiple lightweight models with automated validation) instead of betting on a single technique. Research aimed at SMEs recommends automated pipelines and diverse model sets to improve accuracy in resource-constrained environments (arXiv research (Dec 2024)).

If you want a primer on methods and where they fit in manufacturing, see Understanding Demand Forecasting Methods for Manufacturers.

2) Handle outliers and one-off events without corrupting the baseline

Outliers (weather spikes, policy changes, a single large contract) should be explained, not ignored. A workable rule set:

• Tag events at the transaction level (promo, tender deal, anomaly, stockout period). Don’t “fix” history in spreadsheets.
• Use robust statistics (median/MAD, winsorization) for detection, then apply human review to classify root cause.
• Store the baseline forecast separately from event adjustments so the model doesn’t learn one-off spikes as repeatable demand.

Modern ERP plus connected planning makes those tags operational: they flow into replenishment logic and explain why plans changed—critical for trust and auditability.

How often should you reforecast: weekly vs daily without planning chaos

More frequent reforecasting isn’t automatically better—especially if it whipsaws MRP. The goal is high cadence, low noise: refresh signals often, but change decisions only when it matters.

A practical cadence framework

• Daily signal refresh: Update demand sensing inputs (orders, shipments, cancellations, web signals where relevant).
• Weekly operational reforecast: Publish an updated forecast for purchasing/production execution and exceptions.
• Monthly consensus (S&OP/IBP): Align finance targets, capacity, and policy decisions using scenarios.

Research also shows forecast update cadence influences MRP outcomes and parameter effectiveness. A simulation study on periodic forecast updates and MRP planning parameters found that underestimating demand can be marginally more costly than overestimating, reinforcing the need to tune policies around asymmetric error costs—not just “accuracy” (ScienceDirect (July 2025)).

Measure performance beyond MAPE: bias, stability, and cost-of-error

MAPE is popular, but it can mislead in volatility (especially with intermittent demand). A modern governance layer should track:

• Bias: Are we systematically over/under forecasting (by SKU-family, channel, region)?
• Service impact: Fill rate, backorders, and stockout days attributable to forecast error.
• Forecast stability: How much does the plan change week-to-week? Stability is often an operational requirement, not a nice-to-have.
• Cost-of-error: Penalty curves for under-forecast vs over-forecast (lost margin vs holding/obsolescence).

That stability point matters: a 2025 study on forecast stability notes planners often value consistent forecasts over small accuracy gains because execution systems depend on it (arXiv research (2025)). This is exactly where legacy ERP struggles: it can store a forecast number, but not the governance around it.

Safety stock and reorder points when volatility rises and lead times are uncertain

When both demand variability and lead time uncertainty increase, “set and forget” reorder points become dangerous. Instead:

• Re-estimate variability frequently: Use rolling windows for demand and lead time variability rather than annual parameter reviews.
• Segment service levels: High-margin or high-penalty SKUs deserve higher service targets; slow movers need obsolescence protection.
• Model asymmetric costs: If under-forecasting is more expensive (as the MRP simulation suggests), tune safety stock accordingly.

Modern ERP execution (inventory, purchasing, manufacturing) must be tightly linked to these policies. If you’re also evaluating how AI can help with inventory decisions, see Transform Your Business with AI Inventory Management.

Can SMEs use AI/ML forecasting without a data science team?

Yes—if you treat AI as a productized workflow rather than a one-off model. BCG emphasizes that AI alone isn’t enough: planning leaders still cite volatility and disruption as top challenges, and success requires process and operating-model change (BCG Supply Chain Planning 2026 (PDF)).

A minimal data foundation for SMEs:

• Clean item, customer, and location masters (with consistent units of measure).
• Order/shipment history with timestamps (ideally daily) and channel identifiers.
• Promotion and event flags (even basic tags beat none).
• Inventory positions and stockout indicators to detect constrained demand.
• Lead times (requested vs confirmed vs actual) to model variability.

With that, SMEs can run automated backtesting, select from an ensemble of models, and publish forecasts with confidence intervals—without hiring a full data science team.

Where forecasting should live: inside ERP, add-on planning, or a connected data platform

Most SMEs get the best outcome with a connected approach:

• ERP as the system of record: items, BOMs, routings, inventory, sales orders, purchase orders, financials.
• Planning layer for forecasting/scenarios: generates baseline + event adjustments, runs scenarios, measures bias/stability/cost-of-error.
• Integration back to ERP: approved forecasts and policy parameters (safety stock, reorder points) flow into MRP and purchasing.

This approach also aligns with the market direction: ERP spend remains strong (TechTarget cites expected ERP software spend of $147.7B in 2025), reinforcing a healthy ecosystem for integrations and modernization programs.

Prevent forecast overrides from making accuracy worse

Overrides are often necessary (e.g., a known tender deal), but uncontrolled overrides can degrade accuracy and trust. Put guardrails in place:

• Role-based permissions: who can override which SKUs/time buckets.
• Reason codes: promo, big deal, supply constraint, new channel, etc.
• Override tracking: measure whether overrides improved outcomes (bias/service/cost) versus the baseline model.
• Exception thresholds: require approval when changes exceed a stability band.

60–90 day quick wins during an ERP replacement project

You don’t need to wait until go-live to reduce volatility pain. During replacement, prioritize wins that improve both forecasting and execution:

• Master data cleanup: item/location hierarchies, units, lead times, minimum order quantities.
• Stockout visibility: start capturing lost sales signals so forecasts aren’t trained on constrained history.
• Weekly forecast + exception review: create a simple cadence and measure bias and service impact.
• Scenario playbooks: “promo spike,” “supplier delay,” “demand drop” with pre-agreed actions.

For broader ERP selection and modernization guidance, see The Ultimate Buyer’s Guide to Choosing Manufacturing ERP.

Align sales, finance, and operations on one forecast (even when targets move)

Volatility doesn’t eliminate the need for one set of numbers—it increases it. The trick is to align on a baseline forecast plus scenarios with clear decision rights:

• Baseline: most likely view used for execution.
• Upside/downside: pre-modeled scenarios tied to triggers (pipeline conversion, promo approval, macro signal shifts).
• Governance: who approves scenario switches, how it affects inventory/capacity/cash.

This is where connected planning matters. RELEX’s State of Supply Chain 2025 report found consumer demand volatility emerged as a leading challenge, reinforcing why scenario planning and frequent replanning must connect directly to ERP execution (RELEX State of Supply Chain 2025 (PDF)).

Replace legacy ERP to make reforecasting and scenario decisions practical

If volatility is permanent, your planning system can’t be monthly, manual, and brittle. LikwidERP helps SMEs modernize with a free, open-source, self-hostable ERP that unifies inventory, sales, manufacturing, and finance—making it easier to connect forecasting and scenario planning to real execution. Explore features, review pricing, or reach out via contact to discuss replacing legacy ERP with a volatility-ready planning foundation.