A supply chain planning proof of concept (POC) tests a planning platform on your own data and workflows, for a limited scope, before you commit to a full rollout. A useful one covers one planning decision for a defined set of products and sites, runs for weeks to a few months, and is judged on criteria agreed before it starts: the quality of the recommendations, how often planners would accept them unchanged, and whether the system runs inside your planning calendar.
What a Proof of Concept Should Prove
A POC answers three questions:
- Quality: does the system produce good recommendations on your data, with your constraints?
- Acceptance: would your planners act on those recommendations without editing them?
- Operability: does it run inside your planning window, with your data feeds, at your volume?
A demo on the vendor’s sample data proves none of these. Nor does a proof of concept built on made-up data, which traditional RFP processes sometimes include, as we argue in why RFI, RFQ and RFP processes do not fit SaaS purchases.
The terms overlap, so it helps to agree on them. In this post, a proof of concept tests feasibility on historical data, offline. A pilot runs on live data, in parallel with the current process or in production for a small scope. An implementation rolls the solution out across the business; our post on supply chain planning implementation covers that stage. With SaaS platforms the line between a pilot and the first implementation phase can disappear, because the configuration built for the pilot is the one that goes live.
Scope: One Decision, a Few Sites, the Hard Constraints
Pick one decision with a clear owner and an outcome you can measure, for example store replenishment for two or three categories, the production plan for one plant, or DC replenishment for one region.
Include the hard parts. If shelf life, pack sizes, minimum order quantities, line capacity or truck fill make planning difficult in your business, they belong in the POC. So do the awkward items: slow movers, new products, promoted items and items with supply problems.
Write down what is out of scope, and keep a list of requests that come up during the POC for the next phase. Scope creep makes the result harder to read and the timeline longer.
Data: Use Your Own, and Expect Gaps
A planning POC needs:
- Item and location master data, hierarchies and the sourcing network (which location supplies which).
- Sales or shipment history, ideally covering at least two seasonal cycles where demand is seasonal.
- Inventory positions, open purchase and transfer orders, and lead times by supplier or lane.
- Constraints: pack sizes, order minimums, capacities, shelf life, delivery calendars.
- Promotion and price calendars, if they drive demand.
Data extracts are enough for a historical test. If the POC includes a parallel run, connect at least one live feed, because that is where integration problems show up. Record each gap or error in the data, how it was fixed and who owns it; that list is one of the POC’s results. Our post on supply chain planning data integration lists the feeds a planning system depends on.
Success Metrics for a Supply Chain Planning Proof of Concept
Agree the metrics, their formulas and the pass thresholds in writing before the POC starts, and measure the current process the same way so there is a baseline.
| Metric | What it shows | How to measure it |
|---|---|---|
| Recommendation quality | Whether the plan is better than the current one | Hold back a recent period, generate recommendations as if it had not happened, and replay them against actual demand: projected service, inventory, waste and cost versus what the business achieved |
| Forecast accuracy and bias | Whether demand inputs are sound | Same held-back period, at the level decisions are made (product, location, week) |
| Recommendation acceptance | Whether planners would execute the plan | Share of recommendations planners accept without edits during the parallel run, with a reason code for each edit |
| Planner time | Effort per planning cycle | Hours to produce and release the plan, before and during the POC |
| Run time and stability | Whether it fits the planning calendar | Solve time at full POC volume, and failed or late runs |
Recommendation quality comes first. Acceptance only means something when the recommendations are good and planners review them honestly; a planner who accepts everything without looking, or edits everything out of habit, tells you little. Reason codes on edits show whether the planner knew something the model did not, such as a local event, or was overriding a sound recommendation.
Forecast accuracy on its own is a weak test. A better forecast does not guarantee better orders if the system ignores pack sizes or capacity, and orders are what the business executes.
How Long Does a Proof of Concept Take, and Who Decides?
A POC on historical data can fit in a few weeks: scoping and success criteria, data extraction and validation, configuration, the backtest and a readout. A POC with a parallel run needs enough planning cycles to judge acceptance and planner time, so for weekly replenishment that means at least several weeks of live running on top. Set the end date at the start, along with the decision it leads to.
Five groups need a seat:
- An executive sponsor who owns the budget and the go or no-go decision.
- A planning process owner who defines what good looks like and signs off the success criteria.
- The planners who will use the system. Their acceptance is one of the metrics, so they need to be involved from the first week, not shown the result at the end.
- IT and data owners, for integration, access and security review.
- Finance, to agree the baseline and the method for valuing the results.
Agree the decision rule in advance: the thresholds for going ahead, extending the POC, or stopping.
Common Traps in a Planning Proof of Concept
These traps recur in planning POCs:
- Calling a scripted demo on sample data a POC.
- A scope that is too broad (every module at once) or too easy (only fast movers).
- Success criteria written after the results are in.
- No baseline, so there is nothing to compare the results with.
- Planners kept out until the readout.
- Judging the system on forecast accuracy alone.
- Running several vendors on different data sets or different scopes, so the results cannot be compared.
- Throwing the POC configuration away, so the implementation starts again from zero.
For the wider selection criteria a POC should test, see how to choose supply chain planning software.
How Solvoyo Runs a Test and Learn
Solvoyo’s equivalent, closer to a pilot than an offline POC, is a Test and Learn: a funded project with a limited but meaningful scope that uses your actual planning workflows and exchanges data with your critical data sources, within 3-4 months. It follows the same path as Solvoyo’s customer journey: discovery interviews with your subject matter experts and a signed conceptual design; automated data pipelines from ERP, WMS and POS systems, validated for completeness, consistency and grain; configuration of business rules, exception thresholds and reporting; then forecasting and optimization models running on your live data in parallel with your current process, with structured acceptance testing. Go-live starts with decision support, where planners review and approve every recommendation, typically within 12 weeks of contract. The first scope goes live as a set of MVP categories, and the next phase is planned from there.
Recommendations that have been tested against your own constraints and checked by your planners in a parallel run rarely need editing once they go live: over 95% of Solvoyo’s plans execute untouched. To scope a Test and Learn on your own data, request a demo.
Frequently Asked Questions
A supply chain planning proof of concept tests a planning platform on your own data and workflows for a limited scope, such as one planning decision for a few categories or sites, before you commit to a full rollout. It shows whether the recommendations are good, whether planners would accept them and whether the system runs inside your planning calendar.
A proof of concept on historical data can take a few weeks. One that includes a parallel run on live data needs enough planning cycles to judge the results, which for weekly planning means several more weeks. Solvoyo’s Test and Learn, which uses live workflows and data sources, runs within 3-4 months.
Start with recommendation quality: replay the recommendations against a held-back period and compare service, inventory, waste and cost with what actually happened. Then measure recommendation acceptance (the share planners would execute without edits), planner time per cycle, forecast accuracy and bias, and solve time at full volume. Agree formulas and thresholds before the POC starts.





