Industry

Inventory Simulation: Deterministic vs. Stochastic

October 2026 | Inventory Planning

Every inventory policy is a bet on the future, placed with working capital. APQC’s benchmarking puts the median cost of carrying inventory at 10.0% of inventory value, across a sample of 6,468 companies, so a bet on too much stock is expensive. A bet on too little is expensive too: IHL Group’s 2026 Inventory Distortion Study estimates that stock which doesn’t match demand costs global retail $1.7 trillion a year, about 6.2% of retail sales.

Simulation is how planners test the bet before placing it. But “run a simulation” hides a fork in the road. A deterministic simulation replays one forecast through a policy. A stochastic simulation plays out many randomly generated demand patterns and reports how the policy behaved across all of them. They answer different questions, and confusing the two is how a policy ends up approved because it looked perfect on paper.

Most replenishment rules in practice are variations on four policies. The notation comes from inventory theory, and the rules are compact enough to state in one line each.

Inventory position means on-hand stock plus open orders, minus backorders. The rules of thumb in MIT’s logistics course notes pair the order-up-to policies (s, S) and (R, s, S) with high-value A items and the periodic (R, S) policy with lower-value B and C items. They are rules of thumb, and the notes say as much.

If theory already says which policy is best, why simulate? Because the strongest result in this area comes with conditions attached. Herbert Scarf’s work on the optimality of (S, s) policies, presented in 1959 and published in 1960, shows that this policy form is optimal in the dynamic inventory problem. The problem it solves is a clean one: a single item, a fixed cost per order, and well-behaved holding and shortage costs. Real networks add minimum order quantities, truck-fill economics, shared capacity, and review calendars that nobody chose for inventory’s sake. Which policy wins on your SKUs is an empirical question, and simulation is the empirical test.

02 Deterministic Simulation: Replaying the Plan

A deterministic simulation takes a fixed demand forecast, one number per period, and runs the policy against it. Nothing in the run is random. Given the same inputs, it produces the same inventory path, the same orders, and the same costs every time.

That makes it the right tool for several jobs. It verifies the mechanics of a policy: whether reorder points trigger when expected, whether orders land in time given the lead time, whether the order-up-to level produces sensible order sizes. It handles demand that is known rather than random, such as a seasonal build-up or a planned promotion. It is cheap to run, so it scales across thousands of SKUs in a single pass. And its output is easy to put in front of finance, because it reads like a plan.

Its weakness is built into the method. If the forecast is treated as exactly right, a sound policy will show no stockouts, because the demand it was sized for is the demand it receives. In 2002 Sam Savage gave this problem a name in Harvard Business Review, the flaw of averages: “plans based on assumptions about average conditions usually go wrong.” A deterministic run answers a narrow question, does this policy work if the forecast is right? It cannot tell you how often the forecast will be wrong or what that costs.

And forecasts are wrong, persistently, at exactly the level where inventory is decided.

If uncertainty is a permanent feature of item-level demand, a method that assumes it away has a ceiling on how much it can tell you.

A stochastic simulation describes demand as a distribution, typically through its average and standard deviation, and then generates many random demand patterns from it. Each pattern is a different version of how the year might unfold. The policy is run against every one, and the results are collected as distributions instead of single numbers.

Three things change when you read the output.

The result is a range, not a point. A deterministic run reports one service level. A stochastic run reports the average fill rate, how bad the bad years get, and how often the policy stocks out at all. The table below shows what that looks like on a single item. It comes from a simple simulation we ran for this article, with the assumptions listed underneath, and it is meant to illustrate the method rather than benchmark any real network.

Illustrative simulation, not a Log-hub benchmark. One SKU under an (R, S) policy with weekly review (R = 1) and a 2-week lead time. Weekly demand is normally distributed with a mean of 100 and a standard deviation of 30 (floored at zero). Unmet demand is lost. 2,000 simulated years of 52 weeks. The safety stock in the middle row is z × σ × √(R + L) with z = 1.645.

Read the first row against the deterministic column. A policy sized to exactly the average demand over its protection interval looks flawless in the deterministic run and stocks out in 17 of 52 weeks on average once demand is allowed to vary. The middle row carries a 95% service-level buffer and still produces at least one stockout week in 83% of simulated years, because a service level measured per replenishment cycle compounds over 52 weeks of cycles. The bottom row shows the price of chasing the last points: roughly 67% more average inventory than the middle row for under one additional point of fill rate.

The price of service is nonlinear. That last result is not a quirk of this example. Under the standard safety stock formula, z × σ × √(R + L), the buffer scales directly with the service-level factor z, the matching point on the normal distribution: 1.28 for a 90% service level, 1.65 for 95%, and 2.33 for 99%. Holding demand variability and the protection interval constant, moving from 90% to 95% raises safety stock by about 28%, and moving from 95% to 99% raises it by another 41%. A deterministic run can’t show you this curve. A stochastic run lets you compare policies at several service levels and see where the cost starts to outrun the benefit.

The number of runs matters. If the quantity you are estimating is a probability near 95%, such as the chance a policy gets through a year without stocking out, the 95% margin of error from ten simulation runs is about ±13.5 percentage points. At 100 runs it is about ±4.3 points, and at 1,000 runs about ±1.4 points. Ten runs is a useful sanity check on whether a policy is in the right range. Reading tail risk off ten runs is not.

What Goes In Determines What Comes Out

A stochastic simulation is only as good as its demand parameters, and three habits make them better.

Calibrate at the stocking point, not the end customer. Variability tends to amplify as orders move upstream, which Hau Lee, V. Padmanabhan, and Seungjin Whang described in their work on the bullwhip effect. A standard deviation measured at the shelf understates the variability a distribution center actually sees.

Treat irregular items separately. Mean and standard deviation describe a smooth, bell-shaped demand pattern well. For intermittent items, where many periods have no demand at all, they can mislead, which is why the M5 results above single those series out as the hardest case. How often demand occurs is as important an input as how large it is.

Use scenarios for the uncertainties that aren’t statistical. Randomness around a stable average is one kind of risk. A supplier whose lead time moves from 2 weeks to 4 is another, and it is better tested as a what-if than as a random draw. Recent tariff swings are a live example of lead times and demand assumptions shifting at the same time; our analysis of why manufacturers kept stockpiling covers the market side of that story.

04 Deterministic or Stochastic? How to Choose

The two methods are complements. Each is blind to something the other sees.

In practice a workable sequence has three steps.

Segment first. Stable, predictable items with reliable forecasts don’t need the same treatment as volatile or intermittent ones. An ABC-XYZ split, which crosses value with demand variability, is the usual way to decide where the effort goes.

Match the method to the segment. Use deterministic runs to validate mechanics and confirm that a policy produces the order pattern the plan expects. Use stochastic runs to size buffers and set service targets, especially on the volatile and irregular segments, where a fixed forecast tells you the least.

Stress-test the winner. Take the policy that looks best and change the inputs: a longer lead time, a higher standard deviation, a different review period. A policy that stays acceptable across those changes is worth more than one that is optimal for a single set of assumptions.

05 Five Ways Simulation Results Mislead

06 Conclusion

Supply Chain Visualization: Network Maps vs. How You Actually Manage the Business

September 2026 | Supply Chain Visualization

Supply chain is one of the only business functions where the underlying reality is inherently spatial. Warehouses, plants, suppliers, customers, and the lanes connecting them all exist on a map, not in an abstract data model. That should make supply chain the easiest function in the business to visualize well.

In practice, most network visualizations still fail at the one thing that matters most to the people using them: they show the network as a flat set of locations and lanes, when the people making decisions about it manage it as a structure — by region, by country, by business unit — that the map doesn’t reflect.

This isn’t a design nitpick. It’s the difference between a map an executive can act on in a regional review and one that gets exported to Excel first so someone can rebuild the rollup by hand.

Visualization’s effect on decision-making is often summarized as a single, punchy number — a specific percentage lift in how persuasive a message becomes once it’s shown as a chart or a map. The more useful picture comes from a much larger body of academic research, and it’s more specific than any single percentage.

That’s a more useful finding than a single inflated percentage, because it points to the actual risk: a dashboard or map doesn’t automatically make a decision better just by existing. A cluttered network map with every lane and location rendered at once can slow a reviewer down and produce false confidence in a pattern that isn’t really there. The bar isn’t “add a visualization.” It’s “add the right level of visualization for the decision being made.”

That bar matters more than usual right now. Sage’s February 2026 supply chain preparedness survey, fielded across more than 200 operators in January 2026, found that half of supply chain teams don’t have strong confidence in their ability to respond to disruption heading into the year — and the report ties that confidence gap directly to visibility and system maturity, not to the disruptions themselves. Teams aren’t short on data. They’re short on being able to see it in a form that supports a fast, correct call.

02 A Network Map Is Only as Useful as the Structure Behind It

Supply chains are almost never managed as one flat network. They’re managed by region — DACH, Benelux, EMEA, APAC, whatever structure a given organization uses — with regional leaders accountable for cost and service in their piece of it, and a global view that only matters when it rolls those pieces up correctly. A network map showing thousands of individual shipment lanes with no structure behind it is accurate, but it isn’t usable in that context. It’s the wrong altitude for the question being asked.

This creates a second, quieter problem: different people in the same review need different altitudes on the same data. A CFO wants a cost curve. A COO wants a coverage view. A regional manager wants the lane-level detail behind one specific corridor that’s underperforming. A single static map, built for one of those audiences, works poorly for the other two — not because the underlying data is wrong, but because it was rendered at the wrong level of aggregation for the question in the room.

There’s a third thing good supply chain visualization does that a spreadsheet or a paragraph can’t: it lets someone verify a claim instead of just trusting it. Demand mapping — plotting customer locations and volumes — is its own formal step in network design, alongside center-of-gravity analysis, precisely because it’s the fastest way to sanity-check a recommendation. An isochrone makes this concrete: instead of telling a stakeholder “this facility covers your key accounts,” a reachability map shows the actual area a location can serve in 15, 30, or 60 minutes over real roads, and lets them check it against delivery windows they already know. That’s a claim that’s hard to make convincingly in prose and immediately checkable on a map.

Given that, here’s what actually matters when evaluating visualization capability in a supply chain platform — not as a feature checklist, but as a set of questions tied to the problems above.

Is it genuinely interactive, not just a static export?

A map that only prints is a picture. A map where every relation line or region can be clicked into the detail behind it is a tool someone can actually use to run down an anomaly during a live review, not just admire afterward.

Does it represent the network with enough spatial realism to be trusted?

Two-dimensional dot-and-line diagrams flatten geography in ways that can mislead — a lane that looks short on a flat projection may cross a mountain range or an ocean. 3D rendering and heat-map density views of customer or demand concentration both restore context that a flattened map loses, and heat maps in particular are useful for spotting geographic clustering that a table of coordinates simply won’t surface.

Can it let someone verify a claim, not just view one?

Reachability analysis — isochrones — is the clearest example: it turns “trust us, this facility covers that region” into something a stakeholder can check against a map showing real drive times over real roads.

Can teams compare scenarios side by side, not just look at one state of the network?

Network decisions are rarely about the current state alone — they’re about comparing a proposed change against what exists today, or comparing two candidate solutions against each other. Visualization tooling that supports side-by-side scenario comparison turns a single map into an actual decision-support tool instead of a status report.

Can someone ask a question of the full dataset, not a sample of it?

As networks scale into tens or hundreds of thousands of shipment records, analysis tools that only work on a sampled subset start missing the pattern that matters. The more useful bar is being able to query and chart the complete dataset — including datasets well into six figures of rows — without pre-aggregating it down first just to make it tractable.

None of this is exotic. It’s a reasonable bar for any platform that calls itself a supply chain visualization tool in 2026. Most fall short on at least one of these, usually the first one — the ability to match the map’s structure to how the organization is actually managed.

04 Custom Map Hierarchies: Matching the Map to How You Actually Manage the Network

None of this is exotic. It’s a reasonable bar for any platform that calls itself a supply chain visualization tool in 2026. Most fall short on at least one of these, usually the first one — the ability to match the map’s structure to how the organization is actually managed.

Instead of working from one fixed default hierarchy, organizations can now define their own: assigning countries to custom regions and arranging up to four levels — continent, region, country, location — in whatever order actually reflects how the business is run. A logistics network that’s managed as DACH, Benelux, and the rest of EMEA can be mapped exactly that way, instead of being forced into a generic regional split that doesn’t match anyone’s actual reporting lines.

New Drill Up and Drill Down controls in the map bar move between those levels on the same map: the location view shows individual sites and connections, the country view groups those into country-to-country flows, and the region view rolls everything up into region-to-region relationships. Every relation line stays interactive at any level — clicking one opens the detail behind that specific connection, whether it’s a single shipment lane or an entire region-to-region flow. The new 3D Light Map Mode supports custom hierarchies as well, so the added spatial realism doesn’t disappear once a map is rolled up to a higher level.

For a regional review or a board presentation, that means a network view that’s actually structured the way the business is structured — clear region-to-region and country-to-country flows instead of an undifferentiated tangle of individual lanes, with the ability to identify which corridors carry the business and where the network leans on a single region more than it should. For the analyst building that view, it means defining the region structure once in the Admin Center and reusing it across every analysis, instead of exporting data and manually re-mapping countries to regions in a spreadsheet every time someone asks for a different cut.

05 Conclusion

Dynamic Safety Stock: How to Segment, Calculate, and Manage It

September 2026 | Inventory Planning

“Segment your inventory and make safety stock dynamic instead of stockpiling everything” has become standard advice in manufacturing circles over the past year — we made that case ourselves, using what happened after the tariff whiplash of 2025–2026 as the evidence.

The advice is right. It’s also where most articles stop. Knowing you should segment and go dynamic doesn’t tell you how to actually score a SKU, which formula to apply to which segment, or who owns the decision to change course. This is the operational version — the part that comes after you’ve agreed with the premise.

Not every item deserves an elevated buffer, and treating them as if they do is exactly the blanket-stockpiling mistake this whole approach is meant to fix. Before calculating anything, score each SKU on the factors that actually justify a bigger buffer:

Step 2: Match the Formula to the Segment — Not One Formula for Everything

Once SKUs are scored, the formula choice follows naturally, and using the same formula across every segment is a common way this breaks down in practice.

Items with lead-time variability but steady demand:

Z × σL × D̄ (service-level factor × lead-time standard deviation × average demand) — sometimes called Greasley’s Formula. A simpler variant, the Heizer-Render Formula (Z × σLT, dropping the demand term entirely), is worth using instead when demand is genuinely steady enough to ignore and you’d rather not carry an extra variable you don’t have good data for — it’s a looser fit, but it’s a defensible one for a supplier-driven-delay segment specifically.

Items with demand variability but stable lead times:

Z × σd × √L (service-level factor × demand standard deviation × the square root of lead time).

High-risk items — long lead time, single-sourced, tariff-exposed:

the combined formula, Z × √(L̄ × σd² + D̄² × σL²), which accounts for both sources of variability at once. It needs the most historical data, but it’s the right tool for the segment that actually carries the risk.

In every version, Z is a service-level factor you choose, not one the math hands you — common values run from 1.28 (a 90% service level) to 2.33 (a 99% service level), and a higher target means a bigger, more expensive buffer. That choice belongs with whoever owns the cost/service tradeoff for that segment, not buried in a spreadsheet default.

A quarterly review cycle applied to a fast-moving, tariff-exposed segment will always be behind events; the same cadence applied to a stable, low-risk segment is wasted effort. Set the recompute trigger to the thing that actually changes the number:

That’s the honest version of the story: the constraint isn’t ambition. It’s the unglamorous work of getting the underlying data and structure in shape for something to act on it.

✓ Recompute when a SKU’s supplier lead-time variability shifts materially — not on a fixed schedule.

✓ Recompute when a tariff, exclusion, or trade rule affecting that SKU’s inputs changes.

✓ Flag any SKU whose relevant policy exclusion is set to expire soon, or whose current supplier lead time already exceeds its inventory coverage, for an out-of-cycle review regardless of when the last one happened.

This is the practical difference between “dynamic” as a buzzword and dynamic as an actual process: the trigger is a real event, not a date on a calendar.

Step 4: Assign Ownership of the Reversal Signal

Step 5: Localize, Don’t Standardize

Managing safety stock is doable by hand for a limited SKU set. The math changes shape once you’re scoring and recalculating across thousands of SKUs, multiple suppliers, and warehouses spread across regions that don’t move in sync — segmentation scores that need to update on their own, formulas applied per SKU instead of copy-pasted down a spreadsheet column, lane-level lead-time data feeding the calculation rather than one global assumption. At that scale, it’s not a bigger spreadsheet anymore; it’s a network modeling problem.

That’s the kind of complexity we work through directly with manufacturing clients, through dedicated engagements for safety stock planning and multi-echelon inventory planning across a full network. If that’s the stage you’re at, here’s what that actually looks like in practice.

Conclusion

“Segment and go dynamic” is the right instinct, and the market conditions that made it necessary aren’t going away soon. But the instinct only pays off once it’s turned into an actual method: a scoring system that decides who gets a bigger buffer, a formula that matches each segment’s real risk, a cadence tied to events instead of the calendar, and a named owner for the moment the buffer needs to shrink again.

Agentic AI in Supply Chain: Expectation vs. Reality

September 2026 | AI in Supply Chain

Ask a room of supply chain leaders whether agentic AI will change how their function works, and most will say yes without much hesitation. Ask how many of their organizations have actually put an AI agent in charge of an end-to-end process, and the room goes quiet.

That gap — between what leaders expect and what’s operational today — is more interesting than either the hype or the backlash against it, and it’s backed by more data than either side usually cites.

01 What “Agentic” Actually Means (and Why the Distinction Matters)

Gartner draws a specific line worth holding onto before going further: agentic AI describes systems that plan, act, and adapt toward a goal, not tools that summarize data or flag an anomaly for a person to act on. A dashboard that flags a late shipment is insight. An agent that reroutes the shipment, notifies the affected parties, and logs why it made that call is agentic. Most of what’s been sold as “AI” in supply chain software over the past few years has been closer to the first category — which is exactly the distinction leaders are reacting to when they say agentic AI will change the function. Not more dashboards. Software that starts acting on their behalf.

02 The Expectation Is Already There

Gartner surveyed 509 supply chain leaders globally between July and October 2025, plus 15 leaders from Gartner’s Supply Chain Top 25 companies, and found the expectation is already widespread: 55% expect agentic AI to reduce entry-level hiring needs, 51% believe it will drive overall workforce reductions, and 86% agree that adopting it requires building entirely new talent pipelines. Asked what will most redefine supply chain strategy over the next two years, leaders didn’t point to tariffs, freight capacity, or reshoring — they pointed to changes in ways of working driven by AI and agentic AI.

03 But Most Organizations Aren’t Actually Ready

The expectation runs well ahead of the infrastructure to support it. PwC surveyed 767 operations and supply chain executives at US companies with $100 million or more in revenue in early 2026 and found only 27% have fully embedded an AI strategy across business units, and just 37% are comfortable assigning AI agents to execute full end-to-end operational processes — the rest still want a person in the loop somewhere before an agent’s decision becomes real. The same survey found 94% of respondents expect their organization to shift toward more networked, horizontal operating structures, against only 41% that actually operate that way today. That’s not a small gap, and PwC ties it directly to data: 87% say poor data quality has held back the value they’ve gotten from AI investments so far.

That’s the honest version of the story: the constraint isn’t ambition. It’s the unglamorous work of getting the underlying data and structure in shape for something to act on it.

04 The Trust Curve Is Moving — Just Slower Than the Hype

None of this means leaders are dismissing AI — they’re getting more comfortable with it, just carefully. A 2026 State of the Supply Chain report covering more than 500 leaders across retail, wholesale, and manufacturing found 67% report more confidence in AI than a year earlier, against only 3% who report less. But that confidence hasn’t translated into blind trust: only 10% would let AI make a critical decision without human review, and 54% prefer a hybrid model where AI surfaces a recommendation and a person decides. The same report names three areas where agentic AI is realistically earning that trust today: purchase optimization, always-on integrated business planning, and autonomous root-cause analysis. Notice what those three have in common — they’re bounded problems with a clear right answer to check against, not open-ended authority over the business.

05 Where the Friction Actually Lives: Fragmented Tools, Not Fragmented Ideas

If data quality is the blocker PwC’s respondents name, it’s worth being specific about what that looks like day to day. In Log-hub’s own Supply Chain Network Design Maturity Assessment, 70% of respondents ranked lack of supply chain visibility as their top issue, and 65% cited data issues — missing or inconsistent information — as a major challenge, with most still running on Excel, partial system integration, or ad-hoc tools rather than one connected process. That’s our own research, worth flagging as such rather than presenting as neutral outside evidence — but it lines up with the pattern above: the constraint isn’t a shortage of ideas about what AI should do, it’s that the data those ideas need to act on lives in five different places.

That fragmentation has a cost that’s been studied outside of supply chain too. Research on how knowledge workers switch between software tools mid-task documents real, repeated costs from every switch: converting data between formats, losing focus, and time spent refocusing once the new tool is open. None of that shows up as a line item anywhere. It just quietly taxes every analysis that touches more than one system — which, in most supply chain teams, is nearly all of them.

06 What’s Actually Working Today

Put the last two sections together and a pattern emerges: the agentic AI that’s actually earning trust right now isn’t autonomous in the dramatic sense — it’s narrow, bounded, and built to reduce exactly the kind of manual reconciliation described above, with a person still checking the output.

That pattern is also showing up in how AI assistants themselves are being deployed. In August 2026, Anthropic rolled out enterprise-managed authorization for Claude’s MCP connectors, letting IT admins authorize a connection to a company’s existing tools once, rather than requiring every employee to set it up individually. Early adopters cited include Ramp, deployed across roughly 2,000 employees, along with Webflow and HubSpot. The point isn’t the specific vendor — it’s the broader shift it represents: instead of building a custom agent stack from scratch, companies are increasingly connecting the AI assistants their teams already use directly to the specific tools those teams already run their work in.

The same shift is happening inside Log-hub’s own tools. A supply chain professional can now connect Claude or ChatGPT to Supply Chain Apps using just an API key, and ask it to pull in outside data, reconcile a new file against a prior analysis, or put together the dashboard for a stakeholder update — without leaving the environment already used for the rest of the work. It’s more accurate to call this an AI assistant with direct tool access than a fully autonomous agent in the Gartner sense described earlier — the plan-and-act loop still runs through a person asking for it, not the system deciding on its own. But it’s aimed at the same underlying problem this whole piece has been about: less time spent moving data between disconnected systems, more time spent on the decision itself.

07 Conclusion

The honest read on agentic AI in supply chain right now is that the expectation is real and the readiness isn’t — not because leaders lack conviction, but because most of the work standing between them and it is unglamorous: fixing data quality, connecting fragmented systems, and starting with bounded problems a person can still check. The organizations closing that gap aren’t the ones waiting for a fully autonomous supply chain. They’re the ones removing friction one connected workflow at a time.

Carrier Selection at Scale: The Hidden Cost of Manual Freight Routing

September 2026 | Supply Chain Myths

An LSP rarely works with just one carrier. A mid-sized LSP might hold contracts with half a dozen — each with its own rate structure, its own surcharges, its own rules about which lanes, cargo types, and weight classes it actually covers. Multiply that by hundreds or thousands of shipments a month across multiple clients, and the question “which carrier should handle this shipment?” stops being a simple lookup and becomes a genuinely hard optimization problem.

Most LSPs still answer it manually — a planner who knows the carriers well, working from memory and a handful of spreadsheets. That works, up to a point. This post covers how freight pricing actually works for an LSP, what’s really inside a carrier rate matrix, and why manual carrier selection has a cost that rarely shows up on any single invoice, but adds up fast across a portfolio.

01 How LSPs Actually Price Freight

An LSP’s business model is simple in concept: charge the client one rate, pay the carrier a lower rate, keep the difference as margin. The complexity is in the middle. Every carrier an LSP works with has negotiated its own rate structure — its own base rates by lane, its own surcharges, its own discount thresholds. The LSP’s margin on any given shipment depends entirely on which carrier actually moves it and how accurately that carrier’s true cost was calculated.

This is why freight pricing isn’t a single number an LSP sets once. It’s a live calculation that has to run for every shipment, because the cheapest carrier for one lane, cargo type, and weight class is often not the cheapest for another.

~15%

The average overall margin freight brokers report, with top-performing brokers landing in the 12–18% range — meaning a few percentage points of margin lost to a suboptimal carrier choice is a meaningfully large share of the total.

02 What’s Really Inside a Carrier Rate Matrix

A carrier rate matrix looks simple from the outside, a rate for moving cargo from A to B. In practice, several layers sit underneath that single number:

Surcharges

Fuel surcharges, hazardous goods fees, accessorial charges, handling costs — these sit on top of the base rate and can meaningfully change what a shipment actually costs to move.

Coverage rules

Not every carrier covers every lane, cargo type, or weight class. A carrier’s rate is only relevant for the shipments it’s actually eligible to handle.

An LSP evaluating carriers on base rate alone is comparing incomplete numbers. The real comparison only works once zones, surcharges, discounts, and coverage rules are all factored in — for every carrier, for every shipment.

03 The Hidden Cost of Manual Carrier Selection

This is where margin leakage — a recognized problem in freight and logistics pricing, not a hypothetical one — shows up. The mechanism is arithmetic, not a theory: an LSP’s margin on a shipment is the client rate minus whatever the carrier actually costs, so any shipment routed to a carrier that wasn’t the cheapest eligible option produces a smaller margin than it could have, by definition. Freight cost is specifically named as one of the recognized drivers of margin leakage in pricing and revenue management literature — alongside things like discount exceptions and rebates, which some industry analyses point to as an even larger factor than carrier choice itself.

What’s genuinely hard to put a single industry-wide number on is how often a suboptimal carrier gets picked in practice — that depends entirely on how disciplined a given LSP’s comparison process already is, and there isn’t a rigorous, independently verified study measuring that specific rate across the industry. What’s not in question is the mechanism: at small volumes, an experienced planner can catch most mismatches by memory and instinct. At scale — multiple clients, multiple carriers, hundreds of lanes — consistently re-running a multi-carrier comparison with current surcharges and discount tiers, for every shipment, by hand, gets harder to sustain. Whether that gap actually costs a specific LSP money depends on how good their process already is — this is a real, mechanical risk, not a guaranteed loss.

04 Carrier Choice Isn’t the Whole Problem

There’s a second, less obvious version of this problem: even when someone does sit down to compare carriers, the inputs themselves are often built by hand, and hand-built inputs are more exposed to error. Two examples make this concrete.

First, zone assignment. If a carrier prices by zone rather than by exact lane — a common structure in LTL freight pricing — someone has to map each shipment’s origin and destination to the right zone before any rate lookup is even possible. Done manually across hundreds of postal codes and cities, that mapping is more exposed to mistakes than an automated lookup would be — and every mismatched zone means a wrong rate, independent of whether the “right” carrier was even considered.

Second, shipment-specific surcharges. A hazardous goods fee, an accessorial charge, a client-specific handling cost — these don’t apply uniformly across all shipments, only to the ones that meet certain conditions. If applying them is a manual, case-by-case judgment call, some get missed and some get applied when they shouldn’t. Either way, the “cost” being compared across carriers wasn’t accurate to begin with.

So the hidden cost of manual routing isn’t only about picking the wrong carrier. It’s also about the cost estimate being wrong before the carrier comparison even starts.

05 A Real-World Scenario: One RFQ, Three Different Rules

Picture an LSP that manages freight for several clients and works with a handful of carriers — DHL, Schenker, and FedEx, say, each with its own contracted matrix. A new opportunity comes in: three potential accounts, each sending an RFQ around the same time. By the time the LSP finishes carefully working through the first response, the other two have already signed with a competitor. This part of the scenario is illustrative rather than a documented case, but the underlying dynamic is well recognized in freight sales commentary: pricing that takes too long to put together loses deals regardless of how competitive the eventual number would have been.

Now assume the same LSP is fast enough to respond to all three. Each account, it turns out, needs to be priced differently:

• The retailer wants all of its shipments handled by a single carrier for consistency, rather than splitting the account across whichever carrier happens to be cheapest lane by lane. That still leaves a cost question — the LSP needs to work out which single carrier is cheapest for the account as a whole, not just accept any one of them.
• The large manufacturer doesn’t care which carrier is used at all — only that each shipment goes to whichever carrier is cheapest for that specific lane and weight class. Because of its volume, it also qualifies for a negotiated discount that has to be reflected in the numbers.
• The smaller manufacturer needs shipments grouped per store rather than treated individually, and several of its shipment types carry surcharges that apply only under specific conditions.

Three accounts, three different sets of rules, all layered on top of the same underlying carrier rate matrices. Working this out by hand — for one RFQ — is already time-consuming. Doing it correctly and consistently for every RFQ, every renewal, and every shipment going forward is where manual processes run out of road.

06 The Business Case for Systematic Carrier Selection

Treating carrier selection as a systematic calculation rather than a judgment call pays off in a few concrete ways:

Margin protection

Every shipment routed to the actual cheapest eligible carrier, instead of a “good enough” one, keeps margin that would otherwise leak away unnoticed.

Operational efficiency at scale

Comparing carriers by hand doesn’t scale past a certain shipment volume. A systematic process does, without a proportional increase in headcount or hours.

Stronger carrier negotiations

An LSP that can show exactly how competitive each carrier is, lane by lane, walks into a contract renewal with data instead of impressions.

More accurate client quotes

Knowing the true landed cost — surcharges and discounts included — before quoting a client reduces the risk of pricing too low and eating the difference in margin later.

Auditable routing decisions

A routing decision based on transparent rules is easy to explain to a client, a manager, or an auditor. “Because that’s who we always use” is not.

None of this is a niche concern — the global third-party logistics market is projected to reach roughly $1.46 trillion in 2026, and in the US alone, business logistics costs reached $2.58 trillion in 2024, or 8.8% of GDP, per CSCMP’s Annual State of Logistics Report. At that scale, a percentage point of margin protected per shipment is not a rounding error.

07 What to Look for in Freight Pricing Software

Whether or not you use Log-hub’s tools specifically, here’s a reasonable checklist for evaluating any carrier selection or freight pricing software:

✓ Does it compare all carrier matrices at once, or one at a time? A tool that only evaluates carriers one by one doesn’t actually solve the “which carrier is cheapest for this shipment” problem — it just moves the manual comparison into a different spreadsheet.

✓ Can it respect client-specific routing rules? Some clients need everything through one carrier; others want the cheapest option every time. A tool needs to handle both without forcing one client’s preference onto another.

✓ Does it model real surcharges and discounts, not just base rates? A comparison based on headline rates alone isn’t a real comparison.

✓ Does it enforce carrier eligibility rules? Lane, cargo type, and weight class coverage need to be respected, so no shipment is compared against a carrier that was never actually eligible to move it.

✓ Does it handle zone-based rating automatically? If your carriers price by zone, manually mapping shipments to zones is exactly the kind of manual step that introduces the errors described above.

08 Conclusion

Freight pricing for an LSP isn’t a single rate — it’s a margin that has to be defended shipment by shipment, across carriers that each price, discount, and surcharge differently. Manual carrier selection isn’t wrong, exactly; it’s just a process that stops holding up once volume, client count, and carrier count all grow at the same time. Whether the fix is a sharper spreadsheet process or dedicated software, the underlying discipline is the same: know the true cost of every carrier for every shipment, respect the routing rules that actually matter to each client, and don’t let “good enough” quietly become the default.

Why Automation Will Peak in Mega-Sites and Stop Short Everywhere Else

August 2026 | Supply Chain Myths

Automation in the supply chain is often discussed as a destination: lights-out warehouses, autonomous flows, humans removed from the loop.

In reality, automation behaves more like a migration. It moves aggressively into a few places — and quietly stalls everywhere else. Not because the technology fails. But because variability beats machines far more often than vendors admit.

It’s a pattern we flagged in our roundup of 11 supply chain myths leaders need to unlearn in 2026 — here’s the full picture behind it.

01 Where Supply Chain Automation Works Best

Some environments are almost unfairly well-suited for automation. They share a few characteristics: high volumes, standardized flows, predictable demand, repetitive tasks, and a stable SKU mix. Processes don’t change often. Exceptions are rare and well-defined.

Think of mega distribution centers, e-commerce fulfillment hubs, parcel networks, or high-throughput cross-docks. These are environments where the same items move through the same paths, day after day.

Robots thrive here because they are built for rules and repetition. Amazon alone has passed a million deployed robots across its fulfillment network, and today robots touch roughly three-quarters of everything it ships. When the flow is stable, automation scales beautifully — and the ROI is obvious.

02 Why Automation Investment Concentrates in Mega-Sites

In these “perfect” nodes, automation has three advantages that are hard to replicate elsewhere.

First, utilization stays high. Machines are busy most of the time, which justifies the capital investment.

Second, the process landscape is stable. Layouts don’t change weekly, SKU profiles are predictable, and exceptions are limited.

Third, performance gains are easy to measure: more throughput, lower unit cost, faster processing.

This combination makes automation both operationally viable and financially defendable — and it’s why Gartner expects half of all new warehouses built in developed markets by 2030 to be designed with humans as optional, not essential. That’s the mega-site model, scaling into the future.

03 The Messy Middle: Where Warehouse Automation Struggles

Outside of those ideal nodes lies most of the supply chain. Regional distribution centers. Multi-client 3PL warehouses. Industrial spare-parts networks. Operations handling pallets, cases, and eaches — all in the same shift.

These environments are defined by change. SKU mixes evolve constantly. Orders are unpredictable. Customers have special rules. Processes are overridden daily to “just make it work.”

This is not a failure of execution. It’s the nature of real supply chains. And it is deeply hostile to full automation.

The scale of this middle is easy to underestimate: even with automation investment accelerating everywhere, four out of five warehouses today still run without it.

04 Why Exceptions Break the Automation ROI Model

Robots are excellent at following defined paths, repeating known actions, and handling known item types. They struggle the moment reality deviates: damaged goods, wrong labels, mixed pallets, last-minute priority changes, new SKUs with missing data, or the countless human workarounds that keep operations running.

Every exception triggers intervention. Someone has to step in, reset the process, redesign logic, or manually move inventory. Each exception adds cost, delay, and fragility.

The more exceptions an operation has, the faster automation ROI erodes. That’s why, in practice, exceptions eat robots for breakfast.

05 Why Full Automation Won’t Scale Across the Supply Chain Network

There are structural reasons companies stop short. When volumes fluctuate, robots sit idle. When variability is high, engineering effort explodes. And in multi-client 3PL environments, standardization simply isn’t realistic — what works for one client breaks another.

Labor pressure explains why interest is everywhere: turnover runs near 36%, and most leaders expect to adopt robotics within five years. But interest isn’t fit. The gap between that intent and the reality on the ground looks like this:

At some point, the economics stop working — not because automation is bad, but because the environment is wrong.

06 Selective Automation: What Actually Works in Supply Chain Operations

The future is not end-to-end robotics. It is selective automation: small islands where repetition exists, assisted picking where humans remain central, automated transport in fixed loops, and manual handling of exceptions.

In other words: automation where it fits, humans where it doesn’t.

This hybrid reality is not a compromise. It’s an optimized design choice.

07 What This Means for Supply Chain Leaders

The winners won’t be the companies chasing “lights-out” operations everywhere. They’ll be the ones who: automate for throughput, not for hype; design processes assuming humans will remain essential; and use analytics to decide where automation pays — and where it doesn’t.

This is where decision-driven analysis matters more than flashy demos. We’ve mapped 25 proven cost savings and operational boosts for manufacturers — several of them squarely about knowing where automation pays off and where it doesn’t — so capital gets applied surgically, not ideologically.

08 The Real Conclusion

Automation is not a revolution that sweeps the entire supply chain. It’s a concentration phenomenon. It will dominate a small number of perfect nodes, deliver enormous value there — and stop politely at the door of the messy middle.

Not because technology failed. But because supply chains are human systems first. And designing for that reality is what separates serious operators from slideware strategies.