Turning ERP and CRM data into fast, meaningful search
A recurring request we get isn't "we need semantic search" — it's "we need to find a customer, product or document, and the system holding that data wasn't built to be searched." The data usually lives in an ERP or CRM that's good at being a system of record and bad at answering a loosely phrased question quickly. Being API-oriented end to end is what lets us close that gap without asking the customer to change how they run their business system.
The shape of the problem, almost every time
ERP and CRM systems are built around structured records and exact lookups — a customer ID, an order number, a SKU. That works fine for the workflows those systems were designed for. It breaks down the moment someone needs to find "the supplier we talked to about pumps last spring" or "the document mentioning the warranty clause," where the match isn't an exact field value, and the system's own search, if it has one, is usually keyword-only against a handful of indexed columns.
Why the connector is the actual work
The search ranking itself — semantic embeddings, hybrid scoring, profiles — is largely the same regardless of where the data originates. What changes every time is getting the data out: each ERP or CRM exposes its own API, its own pagination quirks, its own way of representing relationships between customers, orders and products. Building that connector once per source system, and keeping it in sync as records change, is consistently the bulk of the engineering effort in these projects — not the search layer sitting on top of it.
What "fast and robust" means in practice
Fast means query latency that holds up under real usage, not just a demo with a handful of records — the same production targets as any other Relatiq Search index. Robust means the index stays current as the source system changes: new customers, updated product data and edited documents need to show up in search without someone re-running a manual export. Since the connector talks to the source system's API directly, updates flow through the same real-time indexing path any other Relatiq Search customer uses.
The pattern generalizes
Once a connector exists for a given ERP or CRM, the same approach extends to adjacent problems in the same system — product search becomes document search becomes customer lookup, because the underlying work (API connector in, embeddings and ranking on top) doesn't change shape. That's the part of this that scales: the first integration with a new source system is the expensive one, and everything after it is mostly configuration.