Because most of the time isn’t spent modeling — it’s spent writing custom code and waiting for data. Gartner has noted that the benefits of supply chain planning technology typically appear only after implementation is complete and processes stabilize, with average implementation times of around 1.1 years — roughly 13 months. Our experience is that teams can go from spreadsheets to a first AI-optimized plan in as little as six weeks. The gap comes down to two things.
Where the time goes
A typical long-running project spends its time roughly like this:
| Phase | Common approach | Why it’s slow |
|---|---|---|
| Data preparation | Wait for master-data governance to finish first | Governance is never finished; data is always “not ready yet” |
| Requirements | Collect every rule and exception across the company | Scope keeps expanding; the boundary is never drawn |
| Model development | Custom-code every special rule | Schema change, code, test, release — weeks per rule |
| Validation | Run the new system in parallel with the old process | When results differ, someone has to find which rule is missing |
| Rollout | Repeat for each new plant and line | A separate data structure for each industry and site |
The third and fifth rows are where most of the cost lives. If rules live in code, every new rule is a development cycle. If the data structure is fixed per industry, every new business is a rebuild.
Two places to cut
First: run data governance and optimization in parallel, not in sequence.
You don’t need perfect data to start optimizing. Data quality issues are real, but they aren’t a prerequisite. Scheduling needs a fairly small set of objects — lines, products, routings, changeovers, tooling — not the whole company’s master data. Running governance and optimization side by side, fixing data as you use it, is far more realistic than waiting for a clean dataset that never arrives.
Second: put rules in configuration, not code.
This is the technical root of the timeline difference. In an ontology, adding a changeover rule or substitution means adding a relationship to the model — not changing code or database tables. Compare the same rule two ways:
- In code: request → prioritize → develop → test → release. Weeks.
- In configuration: a business user adds a conditional relationship to the model. Same day.
The more unusual rules a plant has, the bigger that difference becomes.
A six-week path
- Model only what the current schedule needs: lines, products, routings, changeovers and tooling. Don’t try to cover the whole company.
- Start from an industry template, then add your own rules: reuse the entities, relationships and rules we’ve built for industrial equipment, electronics, food & beverage and more, then make your experienced planners’ conditional knowledge explicit.
- Validate by replaying history: run the model against the last few months of manual schedules and compare. Every difference points to tacit knowledge that isn’t in the model yet. This replaces a long parallel run.
- Turn on agents in recommend-only mode: let agents monitor and propose replans while planners approve every change. Widen their authority as trust builds.
- Register new plants and lines in the ontology: connect them to the existing network instead of starting another project.
Starting with scheduling and an industry template, a first model usually takes a few weeks — and then grows with the business, rather than being a one-time modeling megaproject.
Connecting data isn’t heavy either: import Excel, connect your ERP, or link directly to operational systems. Many plants’ first version starts from a single scheduling spreadsheet.
A final caution: don’t treat speed as the only metric
Remember the first half of Gartner’s point: benefits arrive after implementation is complete and processes stabilize. So a model that goes live in six weeks should be judged on whether the constraints are complete and whether planners actually use it — not on feature coverage.
The first version only needs to cover the current scheduling problem. It will grow. But the starting point has to be small.
Want a realistic plan for your plant? Talk to us about a scoped six-week pilot.
