Situation
A Canada-based transportation and logistics solution provider faced significant challenges from limited visibility into recurring demand patterns across its freight network. Without a structured view of which lanes had predictable volume, procurement decisions were reactive and ad-hoc, driven by last-minute needs rather than forward-looking intelligence.
The consequences were material: higher costs from last-minute purchasing and expedited freight, missed opportunities for volume discounts, and a planning team spending most of its time on reactive fire-fighting rather than strategic work. The business needed a way to identify which lanes carried stable, recurring demand and plan asset procurement in advance on that basis.
Objective
Success meant delivering a reliable daily dispatch planning tool that could surface recurring demand patterns at the lane level. Four goals shaped the engagement:
- Identify which lanes had predictable, recurring demand and quantify baseline volume by lane and day of week.
- Build a forecasting model accurate enough to support advance asset procurement decisions at the individual and grouped lane level.
- Deliver lane-level views (both grouped and individual) aligned to the client’s own destination definitions.
- Reduce operational costs measurably, with an 8%+ reduction in procurement costs as the target benchmark.
Approach
We built the demand planning tool in two stages: a pattern recognition layer to identify recurring baseline demand, and a time-series forecasting layer to project that demand two weeks forward.
Pattern recognitionWe analyzed trip patterns by lane and day of week to estimate baseline dispatch volumes. To ensure the analysis was grounded in representative data, we compared trends with and without COVID-impacted months, ran scenario analyses excluding outliers, and focused the final synthesis on a refined time window that excluded pandemic periods, public holidays, and the bottom 10% of volume outliers. This produced a defensible recurring baseline demand figure for each lane by day of week.
Time-series forecasting with ARIMAXUsing the same refined time window, we incorporated a calendar as an external regressor to account for weekday and holiday effects. We identified and evaluated multiple time-series models (including ARIMA and TBATS), selecting best-fit models for each lane. The models forecasted baseline dispatch volumes for each lane over the next two weeks and calculated lower-bound confidence intervals based on acceptable over-forecast thresholds. Pattern recognition outputs and ARIMAX forecast outputs were merged, with pattern recognition taking priority for final baseline estimates where both were available.
Destination grouping for low-volume lanesAt the individual lane level, low dispatch activity limited reliable standalone forecasts. We resolved this by aggregating low-volume lanes into client-defined destination groups, enabling more stable baseline forecasts for routes that would otherwise have been too sparse to model individually.
Planning tool deliveryThe outputs were delivered as a daily dispatch planning tool surfacing the minimum number of legs to plan per lane for the coming two-week window, available at both grouped and individual lane levels, with clear drill-down capability from grouped to individual lanes.
Figure 1Recurring demand by lane and day of week (left): minimum dispatch volumes identified for each lane across Monday to Sunday. Distribution of lanes by number of days with recurring demand (right): 50 lanes have recurring demand on 5 days per week; 40 lanes on just 1 day.
Figure 2ARIMAX forecast for a representative lane (May 2023 - Oct 2024): historical trip volume (blue), predicted demand (orange), and confidence interval bounds (grey). The two-week-ahead forecast table (top) shows the minimum number of legs to plan per day.
Figure 3Destination grouping reference table: individual lanes mapped to client-defined destination groups by origin, destination, region, lane type, status, and cost, enabling aggregated forecasts for low-volume routes and drill-down to individual lanes for high-volume ones.
Figure 4Demand planning tool output: minimum number of legs to plan by grouped lane and individual lane for a two-week forward window (Oct 26 - Nov 6). Grouped lane totals and individual lane breakdowns are calculated independently.
Impact
The demand planning tool identified recurring demand for individual lanes covering approximately 45% of total dispatch volume, enough to enable proactive, advance procurement on nearly half the network. On those lanes, the client moved from reactive last-minute purchasing to a standing procurement arrangement, cutting operating costs by at least 8%.
The planning team’s capacity shifted materially. With stable lanes managed through the tool’s baseline forecasts, planners could direct attention to high-variability lanes requiring active oversight, rather than spending the same effort across every route. The tool made it explicit which lanes needed close management and which could run on autopilot.
The client also used the terminal-level KPI outputs to identify locations operating below expected dispatch efficiency and implemented targeted measures to improve performance at those sites. A repeatable, scalable planning process replaced the ad-hoc approach, one the client can refresh and extend as the network evolves.