Lima reads the records for a customer, product or supplier where they live, standardises them, matches them in pairs, and assembles one golden record field by field.
The same five steps for every domain; the keys and the rules differ.
The value exists in one or more sources and survivorship picks the winner: legal name from the registry, address from the verified feed.
The value follows from other values and a rule: lifecycle 'active' from open orders in the last 90 days; tier 1 from spend rank; legal form from the name.
The value is assembled from parts held in different places: the site list from SAP plants plus portal addresses; the contact list from CRM, portal users and order creators.
A deduced value is recalculated when its inputs change and carries the rule that made it; a pieced-together value keeps the source of each part. Both are as auditable as a constructed one.
Both lines are set per domain and per rule, on your rows: the match run shows the pairs at each score before the lines are saved. A strong-key rule usually merges from a high score with a narrow review band; a name-and-address rule keeps a wide review band. The steward can move a line and see the queue change.
| field | candidates | winner | rule |
|---|---|---|---|
| legal name | SAP · CRM · registry | Acme Components GmbH | registry rank 1 |
| registered address | SAP · portal · Deutsche Post check | Werkstraße 12, 71034 Böblingen | verified within 90 days |
| VAT ID | SAP · portal · VIES | DE812345678 | verification passed |
| contact e-mail | CRM (Feb 2024) · portal (Jul 2026) | j.keller@acme-components.de | freshest verified; decay 9 months |
| payment terms | SAP · contract | Net 45 | contract rank 1 |
| bank account | SAP · portal | •••• 4471 | two-person verification |
Rules are per field or field group: an address survives as a unit, a name survives with its legal form, a bank account survives only once verified. A steward's edit outranks every feed until a newer verified value arrives, and the audit trail shows the whole sequence.
The catalogue lists the keys for all thirty domains; these are the most common.
| Domain | Strong keys | Standardised attributes | Deciding evidence |
|---|---|---|---|
| Customer | VAT ID · DUNS · registration number | legal name + country · street + postcode | shared invoices; e-mail domain |
| Supplier | VAT ID · DUNS · LEI · verified IBAN | legal name + country · address | a different verified IBAN keeps records apart |
| Material | GTIN · manufacturer + part number · drawing number | description tokens · physical attributes | same attributes with different words match |
| Site | geocode · plant code in the crosswalk | name + country | same geocode, different role: one site |
| Employee | worker id · work e-mail · directory id | name + hire date | HR is the system of record |
| Asset | serial + manufacturer · asset tag · fixed-asset number | model + site | the fixed-asset register reconciles |
Every domain page names its keys and how they decide.
Match rules, survivorship strategies, thresholds, unmerge and cross-references are standard in every master-data hub, and buyers should expect them everywhere. The difference here is where the work happens: the hubs load a copy of the sources into their own store and match there; Lima matches on the rows in place and keeps the cross-reference as a join, so a source-side correction shows at once and the transactions of each record are on the proposal.
A second difference is what the matcher can see. A hub that holds master data only compares names and addresses. Lima's bed holds the transactions too, so shared invoices, common delivery addresses and order history count as evidence, and two suppliers with one name and two verified bank accounts stay apart.
They stay in their systems, unchanged, and stay addressable by their own keys underneath the golden record. Write-back, where enabled, sends the golden values to them through their released interfaces.
On your rows. The match run shows the pairs at each score, the steward sets the lines per rule, and the queue shows the effect. Lines are versioned like everything else.
Yes, when the evidence says so: a different verified IBAN, a different registration number, different transactions. The rule that keeps them apart is recorded on both.
It proposes on the pairs a rule leaves open, with its reasoning; the thresholds and the steward decide. See AI-assisted.
Bring two systems that hold the same customers, suppliers or materials. In one working day we connect them, run the match on live rows, review the queue with your steward and publish the first golden view with its API.