Integrations that stop breaking overnight.
A transfer that fails silently twice a month costs more than the project that fixes it. We build and repair Salesforce integrations with retries, monitoring, a rollback path and a runbook your own team can follow. Thirteen changes to GoodLeap's live lending system; none caused an incident.
One broken integration fixed in two weeks, $5,500 · a new integration layer from $15,000 · monitoring in Run
It usually sounds like one of these.
"The nightly transfer broke again."
Found in the morning, fixed by hand, nobody knows which change caused it. At GoodLeap a 20,000-record transfer failed about every two weeks for months.
"HubSpot and Salesforce disagree about every lead."
Two systems, two owners, two lifecycle stages. Viedoc's integration was fixed in the first weeks of a Run engagement, with the account lifecycle automated.
"The custodian feed lives in a spreadsheet."
Positions and balances arrive by file and get pasted. At 49 Financial, 100% of custodian data now lands in one client view, event-driven.
Named systems, named clients.
Through MuleSoft where the client already runs it; through Platform Events, Change Data Capture and REST where it does not.
| System | What moves | Where |
|---|---|---|
| HubSpot | marketing to sales handoff, lifecycle sync | Viedoc |
| Custodians: Schwab and others | positions, balances, households | 49 Financial |
| Loan systems: ICE Encompass, Figure | origination data, status, documents | GoodLeap |
| Contact centre: Genesys Cloud CX, Mogli SMS | calls, SMS, screen pops | GoodLeap |
| Analytics: Domo, Databricks | nightly transfers, attribution | GoodLeap |
| ERP and finance: NetSuite, QuickBooks | invoices, payments, account status | scoped from a Health Check |
Also in production at our clients: Insellerate, Verse, Experience.com, HeyGen, PandaDoc, Typeform and Shopify. Named integration pages: HubSpot, NetSuite, QuickBooks, Slack, DocuSign, Stripe.
Five rules every integration we ship follows.
They are the difference between an integration that works on demo day and one that still works in March.
- 01
One owner per field
Every field has one system of record. The mapping is written down before the first line of code, and it is yours.
- 02
Idempotent writes, retries with backoff
A message delivered twice changes nothing twice. A failed call retries, then lands in a queue a person can see, with an alert.
- 03
Event-driven where the volume is
Platform Events and Change Data Capture instead of polling, so a managed package's own triggers and ours do not fight over governor limits.
- 04
Tested in a copy, released with a rollback
Full sandbox, then user acceptance, then a release window you choose and a rollback tested beforehand. Every release gets a change record.
- 05
A runbook your admin can follow
What to check on Monday morning, what the alerts mean, how to replay a failed batch. Documentation you keep after us.
Three integrations, three numbers.
Work that used to take 6–12 months now ships in weeks.
Read the case 49 FinancialWealth management · custodian & reporting−40%client onboarding time, with every custodian's data in one placeAdvisers 60% more productive with one client view. No lock-in to the package.
Read the case ViedocClinical trials · Sweden39 daysfrom first contact to signatureA big partner launched Salesforce for 52 users and left. We took it over as a Run engagement: full review and fixes in the first week.
Read the caseFor a small team, having a partner who tidies up as they go rather than adding to the pile is worth a great deal.
Questions about integrations.
Which systems have you integrated with Salesforce?
Do you use MuleSoft, or build point-to-point?
What does "stops breaking" mean in practice?
What does an integration fix cost?
Can your integration be maintained by someone else?
Make your Salesforce earn.
Thirty minutes, one hypothesis about your business, no pitch. We reply within one business day.
One reply from a person, within one business day.