Hiring a Salesforce Integration Consultant? Avoid These 8 Costly Mistakes

Your systems are connected, but your team is still chasing missing updates, untangling duplicate records, and checking which numbers to trust.

A Salesforce integration consultant should help you make the workflow dependable, not just demonstrate that two applications can exchange data.

Before you approve a new build or an overdue repair, agree on the evidence you will need to sign it off. These eight lessons from ten firsthand problem reports show what to ask, what to test, and how to recognize a recovered business process.

The reports come from public community forums. The people who wrote them were doing what good administrators and developers do: describing symptoms in public and asking for help. Integrations fail for understandable reasons, including ambiguous requirements, settings that change underneath them, and production conditions that a test environment never showed. Treat each example as a blameless retrospective. Where a thread never confirmed the root cause, this article says so.

If you are still deciding what kind of help to buy, start with what Salesforce integration services should include. This article picks up where that one ends: the failure patterns, the safeguards, and the evidence you should see before you accept the work.

1. Confusing business identities

In a HubSpot Community thread about multiple Salesforce Leads for a single contact, an organization explained that one person can legitimately generate several Salesforce Leads: a new form, a different service, a separate sales team. HubSpot keeps one contact per email address, so distinct Leads for the same person updated the same HubSpot contact. The team needed each inquiry counted separately for monthly attribution reporting and routed to the right service team. Overwriting erased that history.

Business consequence: repeat inquiries disappear from reports, campaigns look less effective than they are, and a second service request can be routed as if it were the first.

Safeguard: before anyone maps fields, define what each record represents: a person, an organization, an inquiry, or a transaction. Deduplication should merge copies of the same thing, not distinct business events that happen to share an email address. That may mean modeling inquiries as their own records rather than as fields on a contact.

Evidence to request: a test showing two inquiries from the same person, for different services, arriving a day apart, and where each one appears in both systems and in the monthly report.

What remains uncertain: the thread does not record which design the organization finally chose.

Acceptance question: "If one customer sends us two different requests, will we still see two requests in every system and report that matters?"

2. Unclear data ownership

A HubSpot Community discussion about company information being overwritten described a contact whose HubSpot company had been created from their email domain. In Salesforce, a contact with the same email address belonged to a different account. When the records matched, the Salesforce account became the contact's primary company in HubSpot and its information replaced what had been there. The suggested fix was to disconnect company and account sync, which the reporter said would also stop data they did want to share.

Business consequence: account managers lose context, territory and ownership reporting shifts without anyone choosing it, and teams start keeping side spreadsheets.

Safeguard: decide ownership for relationships as well as values. A field-mapping sheet that says "Company Name: two-way" is not enough. Write down which system is authoritative for each important field and association, what happens when the two disagree, and which exceptions a person reviews.

Evidence to request: a field and relationship ownership table, plus a test in which both systems hold conflicting values and the outcome matches the agreed rule.

What remains uncertain: the thread does not show the full connector settings, so we cannot say whether a different configuration would have prevented it.

Acceptance question: "When Salesforce and the other system disagree about this customer, which one wins, and can you show me that test?"

3. Missing repeat-safety controls

In a Reddit discussion of duplicate transaction records in Salesforce, a Salesforce user described a donation-processing workflow that created duplicate records even though lookups and duplicate checks were in place. The author suspected that closely timed processing was involved.

That suspicion fits a well-known pattern. Many systems deliver notifications at least once and retry when they do not receive a timely acknowledgment. Intuit's QuickBooks Online webhook best practices describe repeated retries, possible out-of-sequence events, and processing queued changes in order so the same change is not handled more than once. A "check whether it exists, then create it" step can also let two near-simultaneous arrivals both pass the check.

Business consequence: duplicate gifts, orders, or invoices distort revenue reports, receipts, and follow-up. Duplicate Salesforce records do not necessarily mean a customer or donor was charged twice, but someone has to prove that.

Safeguard: give each transaction a stable external identifier and make the destination enforce it, so processing the same event twice produces one record. Design retries deliberately; our guide to what to retry, fix, or escalate covers that in more detail.

Evidence to request: a repeat test (the same event sent twice) and a concurrency test (two copies arriving together), each ending with exactly one record.

What remains uncertain: the thread did not confirm a root cause.

Acceptance question: "What happens if the same payment notification arrives twice, ten seconds apart, or at the same moment?"

4. Assuming no errors means the sync is complete

In a HubSpot Community discussion about missing records, an administrator using the native two-way integration described companies that never reached Salesforce, names arriving blank until a manual resync, and deal updates that did not appear, yet found no related sync errors. A reply suggested places to look, such as sync filters, required fields, and integration user permissions. The thread does not record a confirmed fix.

Business consequence: a clean error dashboard gives false confidence. Sales and finance work from incomplete data while everyone believes the integration is healthy.

Safeguard: an error list can only show records the connector tried and failed to process. It cannot show records that were never selected. Reconciliation can: compare the count of records that should have synced with what actually arrived, and spot-check key values.

Evidence to request: a reconciliation report covering a defined period, a named owner who reviews it, and a sample of mismatches that were investigated.

What remains uncertain: the cause of these particular discrepancies.

Acceptance question: "If a record silently fails to sync, how and when will we find out?"

5. Incomplete access testing

A Salesforce Stack Exchange question about an integration user missing Event records described a Salesforce-to-Snowflake integration that retrieved about 150 of just over 200 Event records. Requests for the missing records returned "not found" errors, even though the integration user had been given read access to the relevant objects. A commenter asked about record sharing; the reporter noted private sharing defaults on several related objects and that the integration user had no role. No answer confirmed the cause.

Business consequence: a data warehouse, billing system, or dashboard reports confidently on part of the data.

Safeguard: object permissions say what kinds of records a user may read; sharing settings decide which individual records they can see. Test completeness as the real integration identity, not as an administrator, against an expected count.

Evidence to request: the same count run as an administrator and as the integration user, with every difference either fixed or documented as intentional.

What remains uncertain: the root cause was not confirmed in the thread.

Acceptance question: "Does the integration user see every record it should, and how was that proven?"

6. Insufficient capacity planning and end-to-end testing

Three reports point at the same gap between a working demonstration and a working process.

Salesforce's documentation on API request limits and allocations explains that the daily allocation applies to all API calls made to the org, not to each integration separately, and describes several ways to monitor usage. For webhook-driven work, Intuit's best practices recommend a Change Data Capture call to catch events that may have been missed.

Business consequence: an integration that passed its demo fails on a busy day, or when it meets automation it never met in testing. The first error message may point in the wrong direction.

Safeguard: estimate API usage at peak alongside every other integration in the org, test with the destination's real automation switched on, and plan a recovery path for missed events.

Evidence to request: an API usage estimate, an end-to-end test that includes post-save automation, and a documented catch-up or replay procedure.

What remains uncertain: the QuickBooks thread does not establish where updates were lost, and the Zapier thread does not name the exact limit.

Acceptance question: "Has this been tested with our real automation, at realistic volume, and how do we catch up after an outage?"

7. Authentication and configuration changes

A Salesforce Stack Exchange question about an OAuth approval error described several tools, including a document-signing integration and a forms integration, failing to authorize at the same time. The most highly voted answer linked the timing to a Salesforce change in how connected apps could be used, and described installing each connected app and granting the necessary approval permission.

Nothing in the integrations themselves had changed. The environment around them had.

Business consequence: several business processes stop at once, often with an error message that only an administrator can interpret.

Safeguard: keep an inventory of connected apps, integration users, and their permissions. Decide who watches Salesforce release and security notices, and who is alerted when authorization fails.

Evidence to request: the integration inventory, a reauthorization runbook, and a monitored alert for authentication failures.

What remains uncertain: the explanation comes from a community answer; we did not examine the reporter's org.

Acceptance question: "If a security setting or the integration user's permissions change, who finds out first: our customers, our team, or an alert?"

8. Inadequate ownership and handover

In a Reddit discussion from a manager who inherited Salesforce responsibilities, the writer described an unfinished Salesforce–NetSuite integration, data still sitting in separate systems, and an ongoing integration-platform contract.

Business consequence: an unfinished or poorly understood integration becomes a recurring cost and an operational burden for someone who never chose it.

Safeguard: define ownership before go-live: who monitors it, who handles failures, who approves changes, and who manages the vendor contracts and credentials.

Evidence to request: a handover package with a data flow diagram, the ownership table from lesson 2, an exceptions runbook, an inventory of credentials and connected apps (with no secrets written down), test results, and known limitations.

What remains uncertain: the thread does not explain why the original integration was left unfinished.

Acceptance question: "If our consultant were unavailable tomorrow, could our team operate, diagnose, and safely change this integration?"

Case study: a fixed configuration is not a recovered process

On October 14, 2026, Logic Conduit ran a controlled test using HubSpot's native Salesforce connector and a Salesforce Developer Edition org. It covered one path, HubSpot contact to Salesforce Lead, with one restricted picklist field.

The test used one org pair, one to three records per test, and Leads only. It did not establish the minimum wait or the reason recovery was delayed. The full HubSpot and Salesforce picklist repair write-up has the details.

The lesson for managers is simple. "We fixed the setting" is a statement about configuration. "The affected records arrived, including the updates that were held back" is a statement about the business process. Close the ticket on the second.

Your acceptance checklist

Use this before you sign off on a new integration or a repair. Ask your Salesforce integration consultant to supply the evidence, not just a yes.

Check Evidence to request Accept when
Business identities Test with two distinct inquiries or transactions from the same customer Each appears separately wherever it is reported
Data ownership Field and relationship ownership table; conflict test Conflicts resolve by the agreed rule
Repeat safety Duplicate-delivery and concurrent-delivery tests Each produces exactly one result
Completeness Reconciliation of eligible source records against destination records Differences are explained and owned
Access Counts run as an administrator and as the integration user Counts match, or gaps are intentional
Capacity and end-to-end API usage estimate; test with destination automation switched on Peak usage fits; automation passes
Authentication and change Integration inventory, alerting, reauthorization runbook A named person is alerted to failures
Ownership and handover Handover package and named operational owner Your team can operate and change it
Recovery Affected records after any fix, including held updates The business outcome is verified, not just the setting

Bring one unreliable workflow, not a whole transformation program

If Salesforce is already connected to another system and your team still does not trust the result, start with one workflow. Tell us which systems are involved, which workflow matters most, and the business consequence you need to prevent, whether that is lost inquiries, duplicate transactions, or numbers nobody trusts.

We will help you work out what evidence would show that it is working. For a new connection, the Salesforce Integration Sprint focuses on one defined workflow, with testing, documentation, and operational safeguards built in.