Data & Model
How the estimate is produced
Current model
- Version
- v0.1
- Type
- Rules-based
- Basis
- Cost-engineering formulas
- Processes
- 3
- Materials
- 13
- Updated
- 14 Jul 2026
What drives the estimate
- 1Net / pour weight27%
- 2Manufacturing process21%
- 3Material grade15%
- 4Annual volume11%
- 5Cut length / perimeter9%
- 6Feature count · holes, bends, darts8%
- 7Wall / sheet thickness5%
- 8Tolerance class3%
- 9Coating / surface treatment1%
Connect your cost catalogue
Drop a catalogue file here: .csv, .xlsx, or connect a database
One row per historical part, with the columns below.
part numberprocessmaterial gradethicknesscoatingweightsurface areacut lengthfeature countannual volumeregionactual costdateWith ~2,000 historical parts a trained model typically reaches ±8 to 12% mean absolute error against quoted price.
Connect a system of record
Cost data is never in one place. Geometry and revisions live in PLM, rates and paid prices in ERP, quotes in sourcing. Mūlya reads from each rather than asking you to assemble a spreadsheet.
Siemens · PLM
Part master, revisions, CAD documents, BOM structure, change records
Feeds Part Library · Estimate History · Compare Revisions
PTC · PLM
Part master, revisions, CAD documents
Feeds Part Library · Estimate History
SAP · ERP
Purchase order history, actual paid prices, material master, vendor master
Feeds Rate Master · estimate-vs-actual · Data & Model training set
SAP · ERP
RFQ responses, awarded prices, supplier capability
Feeds Supplier pack responses · quoted price benchmark
Snowflake · Warehouse
Curated historical cost catalogue, cleaned and joined
Feeds Data & Model · the trained model's training set
Self-hosted · Warehouse
Any tabular cost history you already maintain
Feeds Data & Model
File upload · File
One-off catalogue extract, no integration work
Feeds Data & Model
Mūlya · API
Estimates out: into your own dashboards or a PLM property
Feeds Outbound: writes the estimate back onto the part in PLM
Connections are read-only by design. Mūlya never writes to your systems of record, except the roadmap REST API, which writes the estimate back onto the part in PLM once you ask it to.
Roadmap to a trained model
- Catalogue ingestion & cleaningawaiting data
- 2Feature engineeringawaiting data
- 3Model training & validationawaiting data
- 4Accuracy reporting vs. actual quotesawaiting data
