The real cost of getting demand planning wrong

Bad demand planning costs global retail roughly $1.7 trillion a year, or about 6.2% of total retail sales, split almost evenly between products that sell out and products that sit on shelves too long, according to IHL Group's 2026 Inventory Distortion Study, which has tracked this number for two decades (IHL Group). That 6.2% isn't an abstraction. Apply it to a $20M multi-channel retailer and you're looking at roughly $1.24 million a year bleeding out through stockouts on the fast movers and overstock write-downs on everything else. Blue Yonder's own tracking of the IHL data shows this number climbing year over year, from $1.77 trillion globally in prior reports to the current figure (Blue Yonder), which tells you the problem isn't getting easier as companies add channels and SKUs. It's getting worse, faster than most finance teams are pricing it into their forecasts.

The split matters too. Roughly $1.2 trillion of that waste comes from running out of things customers want, and the rest from overbuying things they don't, in a given period. If your demand planning conversation only talks about stockouts, you're solving half the problem. The overstock half is quieter but just as expensive, because it shows up as markdowns, dead inventory, and working capital tied up in a warehouse instead of funding growth.

Why spreadsheets and entry-level tools stop working

Spreadsheets and tools like Netstock or Inventory Planner stop working at a fairly predictable point: when a company crosses roughly 2,000 to 3,000 active SKUs sold across more than one channel, with more than one warehouse or 3PL in the mix. Below that line, a well-built spreadsheet model with reorder points and a few seasonality adjustments genuinely works fine. Above it, the math that made the spreadsheet tractable (one sales channel, one lead time, one seasonality curve) stops matching reality.

The specific triggers we see most often, working with e-commerce and distribution clients in this revenue band:

• SKU count past 2,000-5,000, where manual reorder-point calculations can't keep pace with velocity changes across the catalog.

• Three or more sales channels (owned DTC, Amazon, wholesale, retail) each with different lead times, return rates, and demand patterns feeding the same inventory pool.

• Multiple warehouses or 3PLs, which turns a single-location reorder problem into an allocation problem: not just how much to buy, but where it should sit.

• Promotional or seasonal swings above roughly 30% of revenue, which entry-level tools built for steady-state consumer demand simply weren't designed to model.

• Frequent new product introductions with no sales history, which breaks any forecasting method that leans purely on historical trailing data.

Any one of these alone is usually manageable. Two or three together, which is the normal state for a growing $5M-$50M retailer or distributor, is what pushes teams into building increasingly fragile spreadsheet workarounds, usually maintained by one person who becomes a single point of failure. We covered a closely related version of this breakdown point in our piece on product information management past 200,000 SKUs: the catalog data problem and the demand data problem hit the same companies at almost the same time, because both are functions of SKU count and channel complexity.

Why jumping to Anaplan, Kinaxis, or o9 is usually the wrong next move

The instinct once spreadsheets fail is to go shopping for "real" software, and the search results point straight at Anaplan, Kinaxis Maestro, o9's Digital Brain, RELEX, and Blue Yonder. These are excellent products. They are also built, priced, and implemented for a different company than yours. Gartner's own market framing for supply chain planning platforms groups the serious buyers of these tools in the $1B-$10B revenue range (Gartner Peer Insights), with implementation timelines running 6 to 12 months and requiring a dedicated planning team to configure, run, and maintain the model once it's live.

That planning team is the part vendors don't put in the pitch deck. A demand planner's loaded cost typically runs $75,000-$110,000 a year, and enterprise implementations usually need two to four of them just to operate the platform day to day, before you count the software license itself, which for companies this size often starts in the low six figures annually and climbs fast with usage and modules. Run the arithmetic: a $300K-$1M implementation, a 6-12 month runway before it's actually forecasting anything useful, and $150K-$400K a year in dedicated headcount to keep it running. For a $20M company trying to claw back $1.24M in distortion costs, that math can still work eventually, but the implementation risk and time-to-value gap is exactly where these projects stall or get shelved before go-live.

Is Anaplan or Kinaxis overkill for a company under $50M in revenue?

For most companies under $50M in revenue, yes, enterprise demand planning suites are overkill, and the signal isn't revenue alone, it's complexity relative to planning headcount. Use this as a rough sizing check rather than a hard rule:

• Enterprise tools make sense when you already have (or are about to hire) a dedicated planning team of three or more, operate across dozens of warehouses or international markets, and have the implementation budget and internal IT capacity to run a multi-quarter rollout.

• Entry-level SMB tools make sense when you're single-channel or close to it, under roughly 2,000 SKUs, and your main need is automated reorder points rather than true demand sensing across channels.

• Neither fits when you're the company in the middle: multi-channel, multi-warehouse, 5,000-50,000 SKUs, promotional complexity, but no appetite to hire a four-person planning department or fund a seven-figure platform rollout. That's the gap this whole SERP ignores, and it's where a right-sized, purpose-built system usually beats both options.

If you recognize your company in that third bullet, the honest next question isn't "which vendor" but "build, buy, or augment", which is the same framework we laid out for a related supply-chain function in our piece on procurement automation breaking down in this same revenue band.

What a right-sized, custom demand forecasting system actually requires

A custom forecasting system for a company this size needs four things, in order, and skipping the first one is why most of these projects underdeliver. Start with data readiness, not modeling. Before any forecasting model is worth building, you need clean, unified sales and inventory history across every channel, ideally 18-24 months of it, with returns, promotions, and stockout periods flagged (a stockout period looks like "zero demand" to a naive model, and that's wrong). Most companies in this band don't have this in one place yet. It's scattered across Shopify, a marketplace API, an ERP, and a few spreadsheets nobody trusts fully.

Second comes model selection matched to the actual demand pattern, not the fanciest available method. A slow-moving B2B distributor SKU with lumpy orders needs a different forecasting approach than a fast-moving DTC SKU with steady weekly velocity and seasonal spikes. A single model type applied to the whole catalog is one of the most common reasons in-house forecasting attempts underperform even expensive software.

Third is integration with the systems that already run the business, meaning your ERP or OMS for inventory positions and your purchasing workflow for actually acting on the forecast. A forecast that lives in a dashboard nobody checks isn't a forecasting system, it's a report. And fourth is ownership: who maintains the model as SKUs, channels, and seasonality shift, and who owns the underlying data pipeline and IP once it's built. This is the part we push hardest with clients, because it's the difference between a system that compounds in value and one that quietly decays the day the vendor contract ends.

We've worked with e-commerce operators managing well over a million SKUs on the catalog side (our case study on automating SKU-level data work at scale for iCanvas is public), and the pattern holds across catalog and demand problems alike: the diagnosis of what data actually exists, and in what shape, determines whether the build succeeds far more than the choice of forecasting algorithm does.

Build vs. buy vs. augment: a decision framework for demand planning

Run through these four questions in order, because each one eliminates an option before you spend money on it.

• Is your data centralized and clean enough to forecast from today? If no, no software purchase fixes that first. Fix the data pipeline before evaluating any vendor, enterprise or otherwise.

• Does your complexity (SKU count, channel count, promotional load) exceed what an entry-level tool can model? If no, buy the entry-level tool and revisit this in 12-18 months. Don't over-engineer a problem spreadsheets still handle.

• Do you have, or plan to hire, a dedicated 3+ person planning team and a multi-quarter implementation window? If yes, an enterprise suite is a legitimate option and probably the right one.

• If no to the above but your complexity has outgrown entry-level tools, can a right-sized custom system, built around your actual data and channel mix, get you forecasting accuracy gains without the enterprise headcount commitment? This is where most $5M-$50M multi-channel companies land, and it's the scenario this whole topic exists to name.

This maps directly onto the broader framework we use across every build-vs-buy decision at Genta AI Solutions, laid out in full in our build vs. buy AI decision guide. Demand planning isn't a special case. It follows the same logic as procurement, claims intake, or catalog management: diagnose the actual bottleneck before committing to a six-figure platform or a headcount build-out.

What to ask before you buy (or build)

Bring this list into any vendor conversation or internal build discussion. It's the same list we walk clients through during Genta's diagnosis phase, before anyone talks about a solution.

• What's our actual SKU count, channel count, and warehouse count today, and what will it be in 18 months at current growth rate?

• What percentage of revenue comes from promotions, seasonality, or new product launches with no sales history?

• What does our inventory distortion actually cost us right now: stockout lost sales plus overstock carrying and markdown costs, as a dollar figure, not a vibe?

• Do we have clean, unified sales history across every channel, or does someone need to build that pipeline first?

• If we buy an enterprise suite, who runs it day to day, and have we budgeted their salary alongside the license?

• If we build custom, who owns the model, the data pipeline, and the code when the project ends? (The answer should be you, not the vendor.)

• What related problems, like returns handling or catalog data, feed the same inventory numbers and need to be solved alongside this one?

That last question matters more than it looks. Demand planning, returns, and catalog data are the same inventory accuracy problem viewed from three angles, which is why we wrote about returns management software breaking down for multi-channel retailers as a companion piece to this one. Fixing one without the others usually means redoing the work twice.

If you're working through this decision, this is exactly what our Discovery phase maps out before anyone writes a line of forecasting code, and we're happy to compare notes.

Frequently asked questions

What's the difference between demand planning and demand forecasting software?

Demand forecasting software produces the number: how many units of each SKU you'll likely sell in a given period. Demand planning software uses that forecast to drive decisions: purchase orders, inventory allocation across warehouses, and S&OP meetings. Forecasting is one input into planning. Most companies this size actually need both, built on the same data pipeline rather than bought separately.

How much does demand planning software cost for a mid-sized retailer or distributor?

Entry-level SMB tools run a few hundred to a few thousand dollars a month. Enterprise suites like Anaplan or Kinaxis typically start in the low six figures annually plus implementation costs that can reach $300K-$1M, plus two to four dedicated planners at $75K-$110K each. A right-sized custom system usually lands well below the enterprise range, scoped to your actual complexity.

When should a growing retailer move off spreadsheets for demand planning?

Watch for SKU count crossing roughly 2,000-5,000, three or more active sales channels, multiple warehouses or 3PLs, or promotional and seasonal swings exceeding about 30% of revenue. Any one of these alone is manageable in a well-built spreadsheet. Two or three together is the signal spreadsheets are becoming a liability, not just an inconvenience.

Is Anaplan or Kinaxis overkill for a company under $50M in revenue?

Usually, yes, unless you already have or plan to hire a dedicated planning team of three or more people and can absorb a 6-12 month implementation. Gartner frames these platforms' core buyer base around $1B-$10B revenue companies. Below $50M, the implementation cost and headcount requirement typically outweigh the forecasting gains until complexity and scale genuinely justify it.

Can you build a custom demand forecasting model instead of buying enterprise software?

Yes, and for companies stuck between entry-level tools and enterprise suites, it's often the better-fitted option. A custom build starts with unifying sales and inventory data across channels, then matches forecasting methods to how each product actually sells, integrates with your existing ERP or OMS, and leaves you owning the model and data pipeline outright instead of renting a seat indefinitely.