Why Supply Chain Planning Still Costs $2M in the AI Era

Introduction

If you've sat through a planning vendor's pitch lately, you've heard it. AI has made their engineers dramatically more productive. They build faster, configure faster, deploy faster. Then the quote arrives, and it looks exactly like the one you'd have gotten before anyone said the word AI: seven figures, a services line bigger than the license, and a timeline of many months to years.

If the vendor's cost to deliver really fell, that drop should show up somewhere in your price. It doesn't. That leaves two explanations. Either the savings are real and the vendor is keeping them, or the productivity claims are bigger than the productivity. For most of the market, it's a bit of both.

If the savings are real, they aren't reaching your quote

A large planning implementation is priced off a staffing plan: so many consultants, so many weeks, so many workshops, billed by the hour or by the phase. We've worked on the vendor side of these projects, and that plan gets built from what the last project took. Nobody rebuilds it downward because the tools got better.

So when a delivery team really does get faster, the vendor has two choices. Lower the quote, or keep the price and bank the difference. Nothing in an enterprise sales cycle pushes toward the first. A handful of large vendors compete on features, references, and analyst rankings, not on implementation cost, and the buyer almost never sees the staffing math behind the number. The productivity dividend lands where it always lands, in the margin.

Implementation is priced off staffing plans built from past projects. When the team gets faster and the price doesn't move, the difference is margin, and nothing in the sales process forces the vendor to hand it back.

If the price is honest, the AI claims aren't

The other possibility is that the vendor's costs barely moved, and the AI story is mostly marketing. That's plausible too, because AI speeds up the part of the job that was never expensive. It writes code, screens, tests, and data plumbing quickly and well.

The expensive part of a configurable APS rollout is weeks of workshops spent forcing a process plant into a template built for something else. Deciding how to represent a changeover that costs an hour going light to dark and a full washout going dark to light. Faking a tank level with an object designed for a machine queue. Documenting the rules that will stay in a planner's spreadsheet anyway, because the template can't hold them. Today's AI doesn't sit in those workshops. So a vendor that claims AI transformed its delivery, then quotes the same months of configuration, is telling you two things that can't both be true.

AI speeds up writing code, but a configurable rollout's cost lives in fitting your plant into a template. A vendor that claims a transformation and quotes the same configuration effort is overstating one or the other.

It's usually both, and you pay for both

In practice the two explanations stack. The engineering teams probably did get faster at the parts AI helps with, and that gain stayed in the margin. The headline claims are probably bigger than the delivery reality, because the slowest, most expensive part of the project never changed.

You pay for both. A price that didn't fall, built on a delivery model whose real bottleneck today's AI hasn't touched yet. Ask any vendor, us included, how much of the quote is license and how much is services, and what those services hours are actually spent on. If the answer is mostly configuration, the AI slide in the pitch deck won't lower your price anytime soon.

The savings that did happen stayed with the vendor, and the cost that matters most is one today's AI can't cut yet. Ask what the services hours are spent on, and the AI pitch stops explaining the price.

Where WonForge puts the AI savings

WonForge is a custom-built APS, and custom used to mean expensive, because every client meant building the screens, data pipelines, and infrastructure from scratch. That's exactly the work today's AI coding tools do well. So we build it once, as one platform, and each client gets its own deployment on managed AWS, with hosting, security, and the solver license inside one monthly subscription.

What stays custom is the optimization model: your changeover rules, tank levels, and process splits, in the same model as purchasing, production, transfers, inventory, and shipments. That's the work a configurable rollout spends months approximating in workshops. We spend it building the real thing.

Cheaper doesn't mean instant. There's still a fixed-scope pilot on your own data and a production implementation that takes months. It's just priced like the tools actually got better.

Custom used to cost more because the software was rebuilt for every client. WonForge builds it once, with AI doing much of the work, so what you pay for is the model of your plant.

Frequently Asked Questions

If WonForge costs less, what's being cut?

Rebuilt software and configuration work, not the model. The screens, data pipelines, and infrastructure are built once and reused across clients, so you aren't paying to rebuild them. And because the optimization model is written for your plant directly, there's no template to translate your plant into and no customizations to maintain through upgrades. The model itself is built during the pilot against your real constraints, and it has to solve your actual problem before anything goes to production. The math is the one part that never gets trimmed.

Does WonForge use AI to make the planning decisions?

No. AI coding tools help build the software around the model, like the screens, data pipelines, and tests. The plan comes from a mixed-integer optimization model solved by a commercial solver. It's deterministic, so the same data produces the same plan, and every decision traces back to a constraint or a cost in the model. A forecast can feed in as an input, but the decisions about what to make, where, and when come from the optimization.

Is a custom model harder to keep up to date than a configurable product?

Usually it's simpler, because there's no template layer between your plant and the math. When something changes, say a new grade with its own transition costs or a new tank, the change goes straight into the model's constraints and data, made by people who know how the model is built. Platform uptime and core stability are covered by the subscription. Logic changes are enhancement work billed by the hour, so you pay for the changes you actually make and nothing else.

Conclusion

Planning vendors keep telling you AI made them faster. Either it did and they kept the difference, or it didn't and the pitch got ahead of the delivery, and most quotes charge you for a bit of both. WonForge passes on what the tools save and skips the template that made these projects expensive in the first place. If you want to see what planning your plant should actually cost, Check Your Fit is a short, no-commitment call to find out.

If AI makes software development so efficient, why does supply chain planning still cost $2 million?