System integrations
When the off-the-shelf connector does not exist or does not go far enough, we write the integration: custom middleware, two-way sync, and API work between systems that were never designed to talk to each other.
When do you need a custom integration?
Usually you find out the same way. Someone exports a CSV every Friday, or a system has a connector in Zapier that covers three fields when you need eleven, or your ERP is from 2009 and has a SOAP endpoint nobody at the vendor remembers documenting.
That weekly export is not a process. It is a symptom of a missing integration, and it is costing you both the hour and the six days of stale data in between.
What we build
- Two-way sync between CRM, ERP, accounting and operational systems, with conflict rules for when the same record changes on both sides.
- Custom API clients for platforms with no native connector in your automation tool — including handling their pagination, rate limits and authentication refresh properly.
- Middleware services that sit between systems, normalise the data, and give you one place to look when something is wrong.
- Legacy bridges: scheduled SFTP transfers, direct database reads, fixed-width file parsing, and email-attachment ingestion for systems that offer nothing better.
- Webhook infrastructure with signature verification, replay handling and a dead-letter queue, so a momentary outage does not silently lose records.
- Data migrations with field mapping, validation, a dry run against a copy, and a documented rollback before anything touches production.
- Internal tools and portals when your team needs a screen to work from that no off-the-shelf product provides.
Why sync is harder than it sounds
Anyone can copy a record from A to B. The engineering is in the questions that follow. What happens when both sides change the same field between syncs? What happens when a record is deleted in one system? How do you avoid an infinite loop where each write triggers the other side to write back? What happens to the run that was halfway through when the API returned a 503?
Those answers are what you are paying for, and they are why an integration that "just moves data" costs more than a workflow that fires on a trigger. We define each of them explicitly during scoping so you can see the decisions being made rather than discovering them later.
Every integration includes
- Idempotency, so a retry cannot create a duplicate
- Rate-limit handling with exponential backoff
- Conflict resolution rules you have signed off
- Dead-letter handling for records that cannot be processed
- Structured logging you can search
- Health checks and alerting on sync lag, not just on hard failure
- A documented manual recovery procedure
What drives the price
Integration projects run $9,000 to $35,000.
- API quality on both ends. Two modern REST APIs is the cheap case. One SOAP endpoint with undocumented behaviour is not.
- One-way or two-way. Bidirectional sync is roughly double, because of conflict handling.
- Field count and transformation logic. Twelve fields mapped straight across is simple. Business rules that derive fields are not.
- Volume and latency. Nightly batch is cheaper to build than near-real-time.
- Data quality on arrival. Duplicates and inconsistent formats often need a cleanup pass first.
A concrete example: the Friday CSV
Before
An operations manager exports completed jobs from the field service system every Friday afternoon, opens the file in Excel, fixes the date format and the two columns that always come through wrong, removes the cancelled jobs, and imports it into the accounting system so invoices can be raised the following Monday.
Cost of that: about 70 minutes a week of a senior person's time, invoices raised up to seven days after the work was completed, and an error rate that spikes whenever they are on holiday and someone covers.
After
A scheduled integration pulls completed jobs hourly through the field service API, applies the same transformations in code, filters cancellations, and creates draft invoices in the accounting system. Records that fail validation land in a review queue with the reason attached rather than being silently skipped.
Result: invoices raised the same day work completes, the 70 minutes disappears, and the transformation rules are written down and version-controlled instead of living in one person's Excel habits. Cash collection moves forward by roughly a week, which is usually worth more than the labour saved.
Illustrative of the pattern and the level of detail we scope to. Named-client case studies are published with permission.
Common questions
Our vendor says their system has no API. Is that the end of it?
Usually not. In rough order of preference we look for: an undocumented but working API (more common than vendors admit), a database connection, a scheduled export to SFTP, an email-based report we can parse, and finally browser automation. Each step down that list is less robust and we price and flag the fragility honestly. Occasionally the right answer is that the integration should wait until you replace the system, and we will tell you that.
Where does the integration run?
Depending on the job: inside your automation platform if the logic fits comfortably there, or as a small dedicated service on infrastructure you own — typically a modest cloud instance or a container. Either way it lives in your accounts. We do not host integrations on our own infrastructure, because that would make you dependent on us in exactly the way we are trying to avoid.
What happens if a vendor changes their API?
Something breaks, eventually — every integration outlives at least one API change. Ours are built to fail loudly and alert rather than fail quietly and corrupt data. On a support retainer we get the alert and fix it as part of the monthly fee. Without one, your team gets the alert and the runbook.
Can you do a one-off data migration without the ongoing sync?
Yes, that is a common standalone project. It follows the same discipline: field mapping signed off, a dry run against a copy of the target, a validation report you review before we go, and a documented rollback. Migrations typically run $9,000 to $20,000 depending on record count and how much cleanup the source data needs.
Do you sign NDAs and work under our security requirements?
Yes to NDAs. On security, we work with least-privilege service accounts created on your side, credentials in your vault, and access that you can revoke at any point without breaking anything you own. If you have a formal vendor security review, send it over and we will complete it.
Which two systems refuse to talk?
Tell us what they are and what needs to move between them. Thirty minutes, no charge, and a straight answer on feasibility.
Related: Workflow automation · Automation audit · Automation support