Pilot Projects

A Martenweave pilot is a small, evidence-based way to test whether a controlled model layer improves a real delivery problem. It is not a promise to replace SAP MDG, a data catalogue, or the programme’s existing delivery tools.

A useful pilot scope

Choose one bounded area where model uncertainty already creates rework:

• a Business Partner, Customer, Supplier, Material, or product model slice;
• a migration mapping workbook and representative extracts;
• a high-risk interface or local-country variation;
• a recurring AMS incident pattern; or
• a readiness decision that needs better evidence.

The scope should be small enough to validate in weeks, yet real enough to contain ambiguous definitions, source/target mappings, rules, ownership, and change history.

What the pilot produces

1. A validated canonical model baseline.
2. A documented evidence set and deterministic validation result.
3. Dataset or model gaps with clear supporting evidence.
4. Trace and impact views for selected high-risk objects.
5. A reviewable backlog of proposals and decisions, not silent model edits.
6. A handover pack that distinguishes authoritative files from derived reports and indexes.

What we need to start

• a named delivery question and accountable reviewer;
• representative, approved-to-share inputs (for example, a mapping extract and sample dataset);
• an agreed boundary for confidential information; and
• a short review cadence for findings and proposed changes.

Beginning with a mapping workbook

If an approved source-to-target workbook is the starting point, the local Core can initialize an empty pilot repository with bootstrap-assessment. It profiles the .xlsx evidence, creates a deterministic draft PatchProposal, and writes a bootstrap report for review. It does not convert inference into canonical truth automatically. Optional CSV/XLSX extracts can be profiled alongside the workbook without becoming model truth.

Safe by design

Pilot inputs remain inputs. Martenweave does not write back to SAP, and AI-assisted work produces proposals for review rather than direct canonical changes. The pilot can run entirely locally.

A clear pilot exit check

Before a pilot model is presented as ready for a wider team, run its deterministic readiness profile:

martenweave readiness --repo ./pilot-model --profile pilot --json

Treat a failing gate as evidence for the next review—not as a reason to reduce the threshold or apply a shortcut. Model changes stay proposal-first: validators verify, the required human approvers decide, and the same readiness command is rerun after the approved change.

Inspect a fictional end-to-end pilot

The Northstar Mobility Group synthetic pilot is a reproducible example, not a customer story. It joins seven SAP transformation domains, deterministic CSV/XLSX evidence, intentional blockers, lineage and impact, and a connected local Workbench using the same canonical repository. It shows what a failing readiness decision looks like without implying SAP write-back or automatic model mutation.

Discuss a pilot

Read the engagement process to see the working sequence. When you are ready, send Metalhatcats the model slice, evidence available, and decision you need to make.

Ready to start? Order a pilot on LinkedIn →