Integrations, interrogated
Published
Event Tech Notes Integrations, interrogated
Every vendor integrates with everything, according to every vendor. The word covers three different mechanisms with three different failure modes, and the contract rarely says which one you bought.
"Integration" is the most elastic word in event tech. On a slide it means an arrow between two logos. In production it means one of three quite different mechanisms, and knowing which one you have is the difference between a badge desk that knows about a ten-minute-old registration and one that finds out tomorrow.
The three grades
Native integration. The two vendors built and maintain a direct connection: API to API, event-driven or near-real-time, with error handling both sides claim as their responsibility. This is the grade the sales conversation implies. It is the least common.
Connector-grade. The link runs through a third-party automation platform (Zapier and its peers) or a middleware layer. Genuinely useful, quick to stand up, and structurally different: it typically fires on simple triggers, handles one record at a time, retries naively, and is subject to the automation platform's own rate limits and outages. Crucially, it is usually configured by you or a consultant, which decides the maintenance question below.
CSV-grade. A person or a scheduled job exports a file from one system and imports it into another. Despised, universal, and sometimes the honest choice. Its failure modes are at least visible: a file did not arrive, a column moved. Its real cost is latency and the silent divergence of the two systems between transfers.
The interrogation starts with one question per arrow on the vendor's diagram: which grade is this, exactly? The follow-ups below sort the real answers from the hopeful ones.
Sync direction
Every integration has a direction, and "bidirectional" is a claim to test, not accept. If registration pushes to the CRM but the CRM never pushes back, a correction made by your sales team lives only in the CRM, and the systems drift apart with each edit. Ask, per record type: which system is the source of truth, which direction do changes flow, and what happens on a conflict, when the same record is edited in both places between syncs? If the vendor has no conflict answer, the practical answer is "last write wins, silently," which means data loss on a schedule nobody controls.
Latency is direction's twin. "Synced" can mean milliseconds or midnight. For check-in, the difference is a person standing at your desk who registered at breakfast.
Field mapping, and who owns it when it breaks
Two systems never share a schema. Somebody maps "Company" to "Organization," decides what happens to your registration form's conditional questions when the target has no such concept, and picks which of the two "email" fields wins. That mapping is the actual integration; the plumbing is generic.
Mappings break for undramatic reasons. Someone renames a question on the registration form. A vendor ships an API version change. A picklist gains a value the mapping never heard of. Records then drop or land malformed, usually without an error anyone sees, and the failure is discovered weeks later as "the CRM numbers look wrong."
So the questions that matter are ownership questions. Who built the mapping, and who is contractually obliged to fix it when either side changes? What monitoring exists: does anyone get an alert when syncs start failing, or is detection left to a human noticing absence? Is the mapping documented anywhere outside the head of the consultant who configured it? With connector-grade integrations the honest answer to "who owns it" is usually "you do," which is acceptable only if you know it and staff it.
Before you sign
Name the integrations you depend on in the contract, with their grade, direction, latency, and maintenance owner spelled out; the RFP questions page has phrasing that forces those specifics. Then test the loop before the event: create a registration, watch it arrive, edit it on the far side, and see what happens. The test costs twenty minutes. The alternative is running it live, on show day, with the discovery arriving as a queue.