OPEN RESEARCH PROTOCOL · VERSION 0.4

A fixed method. A falsifiable test.

This proposed MVP tests whether standardised, read-only software can identify material planning exceptions for independent industrial electronics distributors without recurring manual analysis. The engines are not yet validated.

1. Research question

Can the same fixed software rules identify meaningful, economically material demand, inventory and supply exceptions not effectively surfaced by existing systems? Can the process run without bespoke analysis or continuing manual work?

This is company-led product research, not a university study, audit or consulting engagement. No added value is a valid result.

2. Participants and scope

Priority: independent European distributors of industrial electronics (PLCs, HMIs, drives, industrial PCs), preferably 10–100 employees and generally below 250, holding material owned inventory across at least two regional stock locations, with structured ERP or equivalent exports and limited dedicated planning resources. €1m owned stock at cost is an indicative preference, not a requirement proven by research. Comparable products and model transitions are market-selection criteria, not extra product features.

Start with 50–100 shared high-value SKUs, 2–3 locations and preferably 12–24 months of demand and receipt history. No bespoke cleanup. New, used and refurbished items are distinct SKUs unless the participant provides a validated equivalence; no assumed substitutions.

Four standard CSV extracts: product master (SKU, part number, condition, unit cost/currency, unit); demand history (SKU, location, period, recorded demand or shipments, measure); stock snapshot (SKU, location, date, owned available quantity, reserved quantity and restricted stock where recorded); purchase/receipt history (unique PO line and receipt IDs, SKU, destination, order/receipt quantities, open quantity, order date, promised and actual receipt dates). Optional: dated forecasts, target cover, buffer parameters, availability history. Dates, quantities and units must align; missing data is never assumed zero.

3. Fixed diagnostic modules

First validate schemas, keys, dates, units, duplicate records and stock ownership. Exclude invalid records; record omissions. No automatic cross-currency sums, no supplier stock treated as owned stock, and no duplicate allocation across channels.

  1. Demand reliability: compare recent and reference periods using fixed time windows; flag material changes and long periods with no recorded demand. If dated forecasts exist, calculate bias and absolute error against actuals at matching granularity. WAPE is undefined when actual total is zero. Shipment-only data may be censored by shortages; report that limitation.
  2. Inventory capital: calculate owned available stock divided by average period demand for comparable SKU/location pairs. Zero demand produces a separate no-demand flag, not infinite coverage. Flag differences and aged/slow-moving stock only where data supports it. Flag open purchase lines when another location has candidate stock beyond its recorded target (if supplied); otherwise flag coexistence of coverage differences and open POs without asserting surplus. No claim of transferability, avoidability or optimal safety stock.
  3. Supply reliability: compare original promised receipt dates with recorded receipt dates and calculate lead times from order dates, using unique receipt records and explicitly handling partial receipts. Flag overdue open lines and material shifts in historical lead times where sufficient observations exist. No supplier investigation or prediction of disruptions.

Fixed initial thresholds are hypotheses, versioned before each trial: coverage ratio ≥3 for positive-demand pairs; recent versus previous comparable 3-month demand change ≥30% with adequate history; 6 months without recorded demand with complete history; overdue PO ≥7 days; median lead-time shift ≥20% with at least 5 comparable completed receipts in each period. These are diagnostic flags, not prescriptive decisions. Forecast or parameter checks are optional; no external lifecycle database is maintained.

4. Compare against existing tools

Before sharing results, freeze the input snapshot, rules, thresholds and existing alerts. Record whether each flag is new and useful, already known, false, or unverified. If existing alert records are missing, novelty is unknown. Participant confirmation is separate from software output.

Measure each run: usable-record share, skipped checks, confirmed new findings, false alerts, and Delisoon human time. Keep manually discovered findings separate. Target normal runs without human analysis, initial mapping ≤2 hours and recurring support ≤15 minutes per customer per month; these are unvalidated design targets.

5. Economic relevance

Rank within each exposure type using affected quantities × recorded unit cost. Report stock carrying value separately from open PO value, using original currencies unless a supplied consistent conversion is available. Deduplicate within each type by stock snapshot key or PO line ID; do not sum overlapping findings or equate exposed value with avoidable cost.

Example: 60 units × €800 cost = €48,000 affected inventory value. This is not a €48,000 saving. The participant independently determines feasibility and any actual working capital release, avoidable recurring costs or margin effects after taking action. Do not count a proposed transfer as a benefit without evidence.

After the free pilot, separately test actual paid demand. No promise of savings, optimal stock levels or service improvement.

6. Confidentiality and data handling

The website application form collects contact details and optional broad business profile information only. It is not a secure portal for business datasets. Before any research data are transferred, Delisoon and the participant will agree on the minimum fields needed, an appropriate transfer method, confidentiality arrangements, access, retention and deletion.

We aim to use non-personal, SKU-level information. Do not include consumer names, addresses, personal messages, payment details, login credentials or other unnecessary personal data. We will not disclose identifiable business results or name research participants in public materials without separate approval.

For website applications, see our Privacy Notice. Any separate dataset processing will be governed by arrangements agreed before transfer.

7. Limitations and independence

  • These are planned software checks, not evidence of a working or validated production engine.
  • No purchasing execution, optimisation, manufacturer research, bespoke modelling or recurring manual analysis.
  • Missing data disables or qualifies a check; no material findings is a valid outcome.
  • Participation is free with no purchase obligation. A paid service may later be offered.
Apply to the free pilot ↗