SourceLace Docs
FAQOpen the app

Matching records across systems

Tell SourceLace how to recognise the same customer, supplier or person in two systems, such as a Salesforce account and a NetSuite customer, so that every question and agent that combines the two counts them the same way.

Status: Available to every organization. Tell support@sourcelace.com if a rule does not behave as you expect on your data.

Why matching is needed

The same company is often written differently in two systems: "ACME Corp." in Salesforce, "Acme Corporation GmbH" in NetSuite, with the VAT number written "DE811569869" in one and "DE 811 569 869" in the other. A question such as "revenue per customer with open support tickets" that joins the two on the name finds no match for these records, and the answer comes back short without saying so.

A matching rule says, once for your whole organization, how records of two objects are recognised as the same. When a question or an agent combines those two objects, SourceLace applies the rule, so everyone gets the same answer, and the answer says how many records matched and how.

What a matching rule is

A rule has:

  • a name: the thing being matched, such as Customer, Supplier or Employee;
  • a left object and a right object: one kind of record in each system, such as Account in Salesforce and Customer in NetSuite, each with its key field (the field that identifies one record, such as Id);
  • one or more ways to match, tried in order. Each way is a set of checks, all of which must hold. A check compares a field of the left object with a field of the right object.

Your organization has at most one rule for each pair of objects. A published rule applies to everyone and to every agent, whenever the two objects are combined in a question. Nobody can write their own way around it.

What matching does and doesn't do

Without a rule. When two objects have no published rule, or matching is not turned on for your organization, a question that combines them joins the records the way the assistant writes the query: usually on values that are exactly equal, such as the same customer number. "ACME Corp." and "Acme Corporation GmbH" then don't count as the same customer, and the answer shows no match counts and no Check pairs. To match records that are written differently, publish a rule.

With a rule, SourceLace:

  • compares the fields your rule names, with the comparisons below, trying its ways in order;
  • matches the same way for every person and agent in your organization;
  • says in each answer how many records matched, how, and which had no match;
  • marks less certain pairs Check, shows them side by side, and can answer again with exact matches only;
  • can show which system and field each column came from, and each row's record ids.

What matching doesn't do:

  • It doesn't learn from your data or use AI to score pairs. Two records match only when a way of your rule says so.
  • It doesn't recognise nicknames or initials, such as "Bob" and "Robert". Match people on email or phone first (see Matching people).
  • It doesn't change your systems. It never merges records, never writes matches back, and keeps no list of matched pairs.
  • It doesn't decide which system is right when matched records disagree, such as two spellings of a name. The answer shows the columns the question asked for, and Where this came from shows where each came from.
  • One rule covers two objects. To combine three systems, use two rules that share an object, such as Salesforce Account with NetSuite Customer and Salesforce Account with Zendesk Organization.
  • A record id in the Sources column is text, not a link to the record in its system.

Comparisons

Comparison What it means Example that matches
Same The values are identical DE811569869 and DE811569869
Same after cleanup Identical once case, accents, spaces and punctuation are ignored ACME Corp. and acme corp; DE 811 569 869 and DE811569869
Same company name Cleanup, and legal forms such as Inc, Ltd, LLC, GmbH, AG, SA, BV, Corp and PLC ignored Acme Corporation GmbH and ACME Corp
Same domain The same web or email domain, without www.. Free email providers such as gmail.com never count www.acme.com and jane@acme.com
Same phone The same digits, with the country code (+ and a leading 00 read alike) +49 30 1234-567 and 004930 1234567
Same country Country codes and English names read alike Germany, DE and DEU
Similar, high Nearly the same text once cleaned up (a similarity of at least 0.92 out of 1) Globex Industries and Globex Industrys
Similar, medium A little less alike (at least 0.85) Northwind Traders and North Wind Trading
Within Numbers or dates close enough: a number of days, a percent of the larger value, or an amount An invoice date within 3 days; an amount within 1 percent

The two Similar comparisons are what is often called fuzzy matching: they find values that are spelled a little differently, such as typos or extra words. Pairs they find are marked Check unless the way also has a Same check on an identifier (see Ways and bands).

One side of a check can name up to four fields joined with a space, such as firstname + lastname against a full name.

A field that your data protection rules mask or hide can't be used in a rule.

Ways and bands

Ways are tried in order: the first way that finds a pair wins, so put the most reliable first. A typical customer rule:

  1. Same after cleanup on the VAT number (an identifier). Band: High.
  2. Same domain on website and email, and Same country. Band: High.
  3. Similar, high on the name, and Same country. Band: Check.

Every way needs at least one Same check (Same, Same after cleanup, Same company name, Same domain, Same phone or Same country). SourceLace compares only records that agree on it, so a German Acme is never compared with an American Acme, and matching stays fast. A way with only Similar or Within checks can't be saved.

Each way has a band for the pairs it finds:

  • High: the records are matched.
  • Check: the records are matched, and the answer marks them for a person to look at.

A way with a Similar check is High only when it also has a Same check on an identifier: a tax number, a D-U-N-S number, an id, a domain, an email or a phone number. Tick This is an identifier on a Same or Same after cleanup check that holds one. Otherwise SourceLace saves the way as Check and tells you why.

When one record matches several records on the other side (two NetSuite customers with the same domain, say), every such pair is marked Check with "several matches", whichever way found it.

Matching people

Names alone miss many true matches: nicknames, initials and typos ("Jon" and "John", "Bob" and "Robert") make names of the same person look different. Match people on email or phone first, and keep a name way as Check:

  1. Same after cleanup on email (an identifier). Band: High.
  2. Same phone, and Similar, high on the name (firstname + lastname on one side). Band: Check.

Examples

Each example shows what someone asks the assistant, the rule it drafts and what an answer shows. The field names are examples: the assistant drafts from your systems' real fields.

Customers across CRM and finance

Asked in a chat: "Match our Salesforce accounts with NetSuite customers. Use the VAT number when we have it, otherwise the website domain, otherwise a close company name in the same country."

Way Checks Band
1 VAT_Number__c Same after cleanup vatregnumber (an identifier) High
2 Website Same domain email, and BillingCountry Same country country High
3 Name Similar, high companyname, and BillingCountry Same country country Check

Question: "Revenue per customer this year, with open support tickets."

Customer Revenue Open tickets Band Why
Acme Corp €1.2M 4 High same VAT number
Brightpath €640K 0 High same domain brightpath.io, same country
Globex Industries €800K 1 Check similar name 0.96, same country

Under it: "Customer matching: 412 of 430 Salesforce Account records matched (High 408, Check 4); 18 had no match. 9 NetSuite Customer records had no match." with Check pairs and Show unmatched.

Contacts across support and marketing

Asked: "Match Zendesk users to HubSpot contacts by email. If the email differs, same phone and similar name."

Way Checks Band
1 email Same after cleanup email (an identifier) High
2 phone Same phone phone, and name Similar, high firstname + lastname Check

Jane.Doe@Acme.com and jane.doe@acme.com match by way 1. +1 (415) 555-0100 "Jon Smith" and 14155550100 "John Smith" match by way 2, as Check.

Employees across HR and IT

Asked: "Workday workers are ServiceNow users with the same employee ID."

Way Checks Band
1 Employee_ID Same employee_number High

Nothing is compared loosely, so there is never a Check pair. "Which leavers still have active ServiceNow accounts?" joins through the rule, and Show unmatched lists the workers with no ServiceNow user ("12 Workday workers had no match").

Suppliers across two ERPs

Asked: "Match S/4HANA suppliers to NetSuite vendors by tax number; if missing, the same company name in the same country and city; failing that, a similar name in the same country, to check."

Way Checks Band
1 TaxNumber1 Same after cleanup taxidnum (an identifier) High
2 SupplierName Same company name companyname, Country Same country country, CityName Same after cleanup city High
3 SupplierName Similar, medium companyname, Country Same country country Check

"Müller Präzisionstechnik GmbH" in Stuttgart and "MULLER PRAZISIONSTECHNIK" in Stuttgart match by way 2.

Tickets across two service desks

Asked: "Match Zendesk tickets to ServiceNow incidents: same requester email, opened within one day of each other."

Way Checks Band
1 requester.email Same after cleanup caller_id.email, and created_at Within 1 day opened_at Check

A Within check never makes a pair High, so every pair of this rule is Check by design: "28 of 31 escalated tickets matched an incident, all to check."

What a refused rule looks like

  • "Match customers by similar name." Refused when saved: "Each way needs at least one Same check, such as the same country or domain, so SourceLace only compares records that could be the same. Add one."
  • "Match by tax ID" when your data protection rules mask TaxNumber1: refused when published: "TaxNumber1 is protected for some people, so it can't be used to match. Choose another field."

Exact matches only

Asked: "Revenue per customer, exact matches only." For the customer rule above, only ways 1 and 2 count. The answer says: "Exact matches only, at your request: 408 of 430 Salesforce Account records matched; 22 had no exact match."

Writing a rule in Business setup, Matching

Admins open Business setup → Matching, then New rule:

  1. Enter the Name, such as Customer.
  2. Under Left object, choose the system, then the object (such as Account) and its Key field (such as Id). Do the same under Right object (such as NetSuite, Customer, id). The fields are read live from each system with your own sign-in; when they can't be read, you can type the names as the system spells them.
  3. Under Ways to match, set each way's Band and add its checks: a field on each side and a Comparison. Add ways with Add way, and change their order by dragging them or with Move up and Move down.
  4. Optionally type a Note for History, then Save draft. A draft applies to no question.

The rule's table name, such as "match"."customer", comes from its name. AI apps use it to join the two objects through the rule.

Drafting a rule in a chat

Anyone in your organization can have the SourceLace assistant draft a rule, without writing one by hand. Ask in a chat, such as "Draft a matching rule for our customers in Salesforce Account and NetSuite Customer". The assistant also offers a draft when you ask a question that combines two objects with no rule yet. On Business setup → Matching, Draft from our systems opens a new chat with that request typed in: name your two systems and objects, then send it.

The assistant reads both objects' fields with your own sign-in, drafts the ways, and tests the draft on your data as you. You see it as a match card:

  • the rule's name, its two objects and its ways, in plain words;
  • if your organization already has a rule for the same two objects, its name and status;
  • what the test found: the pairs each way found, how many records on each side had no match, and a sample of pairs and of records with no match. The sample is yours alone, as in any answer.

To change the draft, tell the assistant, such as "use the D-U-N-S number first, then the website domain". It drafts a new card. Then choose:

  1. Try in this chat. Your next questions in this chat use the draft in place of your organization's rule for those two objects, and each answer says "Matched with your draft, not your organization's rule". Nobody else is affected. Stop trying goes back to your organization's rules. One draft at a time can be tried in a chat.
  2. Suggest for my organization. Sends the rule and its test counts to your admins, with an optional note on why it helps. Your sample records are never sent. You follow it under Suggestions → My suggestions: Waiting, Approved or Declined with the reason.
  3. Save to Business setup → Matching (admins, and the editors of the rule those two objects already have). Saves it as your organization's rule for the two objects, as a draft that keeps its test, so you can publish it there.

Reviewing suggested rules

Admins review suggested rules under Suggestions, and at the top of Business setup → Review. Each shows who suggested it, their note, the rule in plain words and what their test found. Choose:

  • Approve and publish. The rule becomes your organization's rule for the two objects: a new rule, or a new version of the rule they already have. It is published at once, because its test ran every way. When the test could not run every way, the button says Approve as a draft, and you test and publish it on Business setup → Matching.
  • Edit, then approve. Change the rule first in the same editor as Business setup → Matching (it keeps the same two objects). Your changed rule is saved as a draft version, because the suggester's test no longer describes it: test it and publish it there.
  • Decline. Write a short reason; the person who suggested it sees it.

A rule that uses a field protected by your data protection rules can't be approved for publishing: choose Edit, then approve and use another field. Suggestions are kept for as long as your organization keeps suggested changes, then deleted.

Testing a rule on live data

On the rule's page, open Test on live data and choose Test version N on live data. SourceLace runs the newest version on your systems as you, with your own sign-ins, access and data protection rules, and shows:

  • the pairs each way found, and how many records on each side had no match;
  • a sample of up to 20 pairs per way, the closest calls first, with the fields the rule compares;
  • a sample of up to 20 records with no match on each side;
  • the pairs marked "several matches".

The sample is yours alone and expires like any result. Only the counts are kept with the rule, such as "Tested by fi@corvanta.com on 2026-10-15: way 1 312, way 2 96, way 3 (Check) 4; unmatched 18 left, 9 right."

To test, you need to be signed in to both systems and allowed to read both objects. If an admin can't see the data, they name someone who can, such as Finance's data owner, as an editor of the rule (see Who can edit a rule).

A large test asks first, like any large question: SourceLace shows how many rows it would read, and you choose Run this plan.

If a field of the rule is protected for you, the test runs with the ways that don't need it, and doesn't count for publishing: someone who can read every field of the rule tests it.

Publishing

Publish version N makes that version the one every question and agent applies. It is offered only after a test of that very version in which every way could run. A rule that uses a field protected by your data protection rules can't be published.

Changing a published rule saves a new version, while the published version keeps applying. Test the new version, then publish it. Unpublish stops the rule applying (it becomes a draft again); its history stays.

Once a rule is published, an answer that combines the two objects says how many records matched (High and Check) and how many had no match (see Reading an answer).

Reading an answer

Under the table of an answer that combined the two objects, a line says how your rule matched them, such as:

Customer matching: 371 of 412 Salesforce Account records matched (High 352, Check 19); 41 had no match. 27 NetSuite Customer records had no match.

  • High pairs were found by a way you trust, such as the same VAT number. Check pairs were found by a less certain way, such as a similar name in the same country, or one record matched several on the other side. In the table, rows whose band is Check are labelled Check.
  • Check pairs shows those pairs side by side, field by field, with the way that found each one and why, such as "similar Name 0.93, same country", the closest calls first. When a row of the table is one pair, its Check label opens that pair. For example, "Northwind Traders" in Germany next to "North Wind Trading" in Germany.
  • Show unmatched lists the records on one side that no way of the rule matched, such as a customer that exists only in Salesforce.
  • Exact matches only asks the same question again, leaving out every pair found by a similar name or a close value. The line then starts "Exact matches only, at your request" and counts records with no exact match.
  • A line that starts "Matched with your draft" used a draft you are trying in this chat, not your organization's rule.

The pairs and records are read from the answer while it is kept, with your own access: values your data protection rules protect stay protected. Nothing you open is saved, so the same Check pairs come back with the next answer until the rule is changed.

Where this came from

Every column of an answer has a small button that says which field in which system its values came from. For answers that used a matching rule, a column made by the rule says so, for example band: "From the Customer matching rule: band". Press Show where columns came from above a table to see every column's origin under its name, and Show record ids (or Ask again with record ids) for each row's id in each system, such as Account 001A0 and Customer C-1001.

Full steps, examples and the choice of what tables start with are in Where this came from. Record ids show when the answer lists records with their ids. A total comes from many records, so it has none: ask for the records themselves, such as "list the customers with their ids". When SourceLace can't tell where a column came from, the table says so instead of guessing.

Agents and matching

Agents use your organization's published rules, the same way as everyone, and never a draft. When an agent's run reads pairs a rule marked Check, every record change it proposes after that waits in Approvals for a person, even when the agent may otherwise make changes by itself. The approval says why: "Waits for a person: this run read pairs that Customer matching marked Check." An agent that asks for exact matches only is not held.

Who can edit a rule

Admins can create, change, test, publish and delete every rule. On a rule's Who can edit tab, an admin names other people, by their SourceLace email, who may change, test and publish that rule. Editors who are not admins see Matching in the menu, with only the rules they edit. Only admins add or remove editors and delete rules.

History

Every save is a version, with who saved it, when, the note, and what its tests found. On the History tab:

  • Compare with the newest shows what changed between a version and the newest one, line by line.
  • Restore saves an earlier version again as the newest version. Nothing is deleted: the versions in between stay. Test and publish it to apply it.

Statuses

Status What it means
Draft Saved, and applies to no question.
Published Applies to every question and agent that combines the two objects.
Paused A system no longer has a field the rule uses (it was renamed or removed), or a system the rule uses was removed. Questions use exact matches only, from the ways that still work, or are refused. Fix the rule, test it and publish it again.

What SourceLace keeps

Nothing about the records it matched: no list of pairs, no keys, no labels. A draft you try in a chat is kept with that chat, like its answers, and deleted with it. A suggested rule holds the rule and its test counts, never the records in your sample. Each question matches again, on the data as it is at that moment. A rule holds only field names and comparisons, encrypted with your organization's own key, and every change to a rule (saving, testing, publishing, pausing, deleting, naming editors) is in your audit log, with counts, never record values.

When something goes wrong

What you see What it means What to do
"Each way needs at least one Same check, such as the same country or domain, so SourceLace only compares records that could be the same. Add one." A way has only Similar or Within checks. Add a Same check to that way, such as Same country.
"Way N is Check, not High: a way with a Similar check is High only when it also has a Same check on an identifier..." SourceLace lowered the way's band. Keep it as Check, or add a Same check on an identifier and tick This is an identifier.
"Customer matching already applies to these two objects: one rule per pair of objects. Change that rule." A rule for the same two objects exists, in either order. Open that rule and change it.
"Another rule already has the table name ..." Two rules would have the same table name. Choose another name.
"... is protected for some people, so it can't be used to match. Choose another field." Your data protection rules mask or hide that field. Use another field, such as an id or a domain.
"Test version N on live data first: a rule is published only after a successful test." The newest version has not been tested. Open Test on live data and test it.
"The last test could not run every way for the person who ran it..." A field was protected for the person who tested. Ask an editor who can read every field to test it.
"... changed this rule since you opened it. Reload it, then make your change again." Someone saved the rule while you were editing. Reload the page and make your change again.
"You have not connected ... yet." You are not signed in to one of the two systems. Connect it under My connections, then test again.
"Customer matching: way 3 ... would compare 48,000,000 pairs, more than the 5,000,000 allowed..." A way's Same checks are too broad for your data. Add a narrower Same check to that way, such as the city.
"Customer matching is paused: NetSuite Customer has no field vatregnumber. Exact matches only, from way 2." The system no longer has a field the rule uses. Change the check to the field's new name, test the rule and publish it.
"Customer matching applies to Salesforce Account and NetSuite Customer. Join through "match"."customer"..." A question joined the two objects of a published rule directly. The assistant reads this and asks again through the rule. Nothing; if you write SQL yourself in an AI app, join through the rule's table as the message shows.
"There is no matching rule called ... Match tables you can use: ..." A question named a rule that doesn't exist or isn't published, or that you can't use because you can't read one of its objects. Use one of the rules the message lists, or ask an admin to publish the rule.
"Customer matching can't run for you: ..., and it is the left key of the rule." A key field of the rule is protected for you, or no longer exists. Ask an admin or an editor of the rule to fix it, then ask again.
"Customer matching is paused for you: ... No exact way of the rule can run without it, so these two objects cannot be combined now." A field every exact way needs is protected for you or gone. Ask an admin or an editor of the rule to choose another field.
"Your groups in SourceLace do not give you read access to Customer in netsuite:prod..." You have no access to one of the rule's two objects, which the rule needs to read. Ask a SourceLace admin for read access to it, as for any question.
"That is not a matching rule ..." A draft or a saved rule was missing its name, an object, a key or its ways. Ask the assistant to draft it again, or fill in the missing part on Business setup → Matching.
"Entity matching is switched off for your organization." Matching has been switched off for your organization. Ask your SourceLace contact to turn it on.
"You already suggested a matching rule for these two objects..." Your earlier suggestion for the same objects is still waiting. Wait for it, or withdraw it under Suggestions → My suggestions and suggest the new draft.
"This draft was already suggested..." You already sent this card's draft. Follow it under Suggestions → My suggestions.
"Only your SourceLace admins, and the people they named to edit this pair's rule, can save it..." You chose Save to Business setup → Matching without being an admin or the rule's editor. Choose Suggest for my organization instead.
"A reply is still being written in this chat..." You chose a card's button while the assistant was answering. Wait for the answer to finish, then choose it again.
"This draft is no longer in the chat. Ask for it again." The chat or its card was deleted. Ask the assistant to draft the rule again.
"Result ... did not use the matching table ..." The answer you opened pairs for did not use that rule. Open Check pairs or Show unmatched under the answer that used it.
"Result ... was not found" or "has expired" when opening pairs Answers are kept for a limited time, and their pairs with them. Ask the question again.
An agent's change waits with "this run read pairs that ... matching marked Check" The run used Check pairs, so a person decides. Look at the change in Approvals, and approve or reject it. To let such runs go by themselves, have the agent ask for exact matches only.
"A changed rule keeps the suggestion's two objects..." While editing a suggested rule, you changed one of its objects. Keep the two objects, or decline it and write a new rule on Business setup → Matching.
"Your organization has 100 matching rules, the most it can have. Delete one first." Your organization reached its limit of rules. Delete a rule you no longer need, or ask your SourceLace contact.
"Entity matching is switched off for your organization." Matching has been switched off for your organization. Ask your SourceLace contact.