Insights · CRM connectors
"Integrated with our CRM" describes at least four different products at four different prices, and the gap between the cheapest and the most useful of them is where most disappointing projects live. This guide sets out the four depths of integration, what screen-pop genuinely requires, how call logging differs by CRM, the questions that decide residency and tenancy, and how to make the build-versus-buy call without regretting it in year two.
Before comparing vendors, work out which of these you are actually being sold.
1. Link. The contact centre opens a CRM URL when a call arrives. No shared state; the CRM doesn't know a call is in progress and the phone system doesn't know what the agent did. Cheap, fast, and it does solve "the agent had to search". It solves nothing else.
2. Embed. A softphone is placed inside the CRM as an iframe or a panel, but the two don't talk. The agent stops alt-tabbing, which is a real productivity gain, and still logs calls by hand. Many "integrations" in marketplaces are this.
3. CTI. Bidirectional. Call events drive the CRM — screen-pop on ringing, matched to the right record — and CRM actions drive the phone: click-to-dial, transfer, hold, disposition. Calls are written back automatically as the CRM's native activity object, with duration, outcome, notes and a recording link. This is what most organisations mean when they say integration, and it's the level at which the numbers in a CRM report start being trustworthy.
4. Full workspace. The agent lives entirely in the CRM: presence and availability set from the CRM UI, omnichannel routing of voice, chat, email and tasks through one engine, multi-session handling with per-session state, and productivity tooling on top. This is where Microsoft's Channel Integration Framework v2 sits, and it's a materially bigger implementation than CTI — sessions, templates and application tabs all have to be provisioned in the CRM.
Depth 3 is the right target for most contact centres. Depth 4 is right when agents handle several concurrent conversations across channels and the CRM is the system of record for all of them. Paying for 4 and using it as 3 is a common and expensive mistake.
Screen-pop demos beautifully and breaks in production for reasons that are always the same six things.
0400 123 456, +61400123456, 0061 400 123 456 and (07) 3000 0000. The ANI arrives in one canonical form. Matching has to normalise both sides to E.164 before comparing, and it has to handle the Australian leading-zero rule rather than assuming North American formatting.+ in an E.164 number is a reserved character in several CRM query languages and must be encoded before the lookup is issued. Get this wrong and every mobile number silently fails to match while landlines work — the classic "it works for some callers" bug.Ask any vendor to demonstrate all six on your data. It takes twenty minutes and it is the most predictive twenty minutes of the whole evaluation.
Every CRM records a call, and no two record it the same way — which is why "we log calls" is not a specification.
In Dynamics 365 a call is a Phone Call activity, related to a regarding record, with participants, direction and a status reason; wrap-up dispositions typically map to option-set fields you have to agree in advance. In HubSpot a call is an engagement on the record timeline, with an outcome from a defined list, and it needs to be created through the supported calling API if it is to behave like a native HubSpot call for reporting. In Pipedrive a call is an activity, and the interesting part is association: it must attach to the person and the right deal, or your pipeline reporting quietly understates activity.
Four questions worth settling before you sign:
A worked example of all of this on one CRM is in our Amazon Connect Pipedrive integration guide.
A CRM connector sits between two systems and therefore sees everything that passes between them. Four questions decide whether it is safe.
What does the connector store? The good answer is configuration and tokens, not conversations. Ask specifically whether call content, transcripts or customer records are persisted, and for how long.
Where does it run? Amazon Connect data stays in your AWS account and region. The connector's own broker may not. For Australian buyers with a residency obligation this is the question that matters most — ProUCX runs with residency options in Australia, Singapore, the USA and Europe, and you should expect any vendor to answer as precisely.
How are tenants isolated? Per-tenant configuration and credentials, not one shared configuration with a customer identifier in a query string.
Who can reach the administration surface? Portals should be authenticated, and they should be excluded from search indexing. If a vendor's customer portal turns up in a Google search, that tells you something about the rest of it.
The security page sets out how ProUCX answers these.
Building is entirely feasible. Amazon Connect's Streams API is public and well documented, every serious CRM has an embedding SDK, and a capable developer can demonstrate a working screen-pop in a fortnight. That fortnight is not the project.
The project is what follows: OAuth token refresh that survives long shifts; approved-origin configuration for every embedding surface; browser policy changes that break microphone access or third-party cookies without notice; CRM API version deprecations on the CRM vendor's schedule, not yours; multi-region behaviour; per-customer field mapping; the picker, the quick-create and the wrap-up mapping described above; and someone rostered to fix it the morning a platform update lands. Integration code is not a project with an end date, it's a permanent maintenance line.
Platform churn is the part most business cases miss. The industry is currently absorbing exactly that kind of change on the Salesforce side, with a deadline that is forcing an entire generation of connectors to be rebuilt — we looked at what that does to the market in the CRM integration market: where the connector gaps are. Whoever owns your connector owns that work.
Build when the CRM is bespoke or the workflow is genuinely unique to your business. Buy when the CRM is a mainstream product and you want the vendor absorbing platform change. And if you buy, buy something metered — ProUCX's Amazon Connect connectors for Pipedrive, HubSpot and Dynamics 365 are on AWS Marketplace at US$9 per agent per month, or US$15 on the premium tier, which is cheaper than the meeting you would hold to decide whether to build one.
If you're comparing platforms as well as connectors, the platform-level view is in Amazon Connect vs NICE CXone vs Genesys, for CRM integration.
Decide the depth first, because it determines the price and the value. Test screen-pop against real, messy customer data before anything else. Agree exactly which CRM object gets written and when. Settle residency and tenancy in writing. And treat build-versus-buy as a decision about who maintains the integration for the next five years, because that's the actual question.
Amazon Connect, NICE CXone and RingCentral connectors for the CRMs your team already uses.