Bring your own telephony, on purpose

DialSplice does not place calls. It does not record them, and it does not transcribe them. When an agent clicks Call, the number opens in the agent’s own phone app (RingCentral today), the call happens there, and DialSplice records the outcome as a disposition. If that sounds like a gap in the product, it is a deliberate one. This post is about why.

What “bring your own telephony” means concretely

Your phone system stays your phone system. IVR menus, ring groups, number inventory, voicemail boxes, and recordings all live in the provider. Predictive or power dialing, if your provider offers it, stays the provider’s. DialSplice sits on top: it decides which agent gets which work item, in what order, under which policy, and what the next step is when the call ends.

The connector reads call events (ringing, answered, missed, duration), SMS delivery and receipt events, voicemail-left events, number inventory per location, and aggregate metrics. It writes click-to-call intents to the agent’s own app, outbound SMS under policy, transfer handoff requests, and the location’s caller ID for outbound work. It never receives call audio, transcripts, or message bodies, and it never sends provider tokens to the browser.

Reason one: the audio is where the risk is

Recordings and transcripts are the densest form of regulated content a call team produces. A recording contains everything the contact said, including things they were never asked. Every system that holds a copy is a system in scope for the review, the BAA, the retention policy, and the breach plan.

By never carrying the call, DialSplice never holds the audio. The ledger stores events and hashes. The call log exports to CSV without message content. The Security page lists recordings, transcripts, and message bodies from the telephony adapter under “We never return”, and that is a structural fact, not a configuration setting. There is no toggle that would let a future admin turn recordings on inside DialSplice, because the pipe does not exist.

For a compliance lead, that removes an entire class of question. Where is the audio? In the provider, where it already was.

Reason two: credentials stay on the server

Click-to-call works by handing a number to an app the agent is already signed in to. The agent’s phone app authenticates the agent; DialSplice does not need to. Provider tokens live server-side and never reach the browser. There is no embedded softphone in the agent’s tab holding a credential that could be extracted.

This is also why DialSplice never originates the call. Originating would mean acting as the caller, with a credential, on the agent’s behalf. Handing off the number keeps the agent, and their authenticated device, as the party that dials.

Reason three: dialing regulations are about who dials

Predictive and power dialers exist to save agents the seconds between calls. They also put a machine in the position of initiating contact, which is exactly what consent and dialing regulations are written around. DialSplice prioritizes and prepares work; agents place calls in your phone system. A person, holding a lease on one work item, having seen the preflight, presses Call.

This does not make the process slow. It makes the preparation fast and leaves the initiation human. The contactability preflight has already told the agent which channels are open. The guidance panel already shows the approved script beside the contact. The disposition list is already there for the outcome. What the agent does in their phone app is the one thing that should stay theirs.

Reason four: transfers and outcomes still work

Bring your own telephony does not mean the phone system is invisible to DialSplice. A transfer hands off inside the provider, and DialSplice journals both legs and keeps the work item with its owner. A missed inbound call arrives as a routed event and lands in the Inbound view for the same location, with its return window attached. Voicemail-left and no-answer outcomes come back as events and advance the retry cadence or hold at the cap.

The distinction is between the call and the decision about the call. The provider carries the call. DialSplice makes and records the decision: who, when, under which policy, with what outcome.

What it buys a regulated team

Keep the tools your agents already know, and the recordings and QA scoring already in them. Add one layer that owns the queue, the preflight, and the ledger. When the compliance review comes, the answer to “what does DialSplice hold” is short: work items, scope, timestamps, dispositions, event hashes, and operational telemetry. The answer to “what does it never hold” is shorter: the audio, the transcript, the message, the token.

It also keeps the vendor list honest. You are not replacing a phone system you already pay for and your agents already know. You are not migrating recordings. If you change providers later, the connector changes; the queue, the policy, and the ledger do not.

If your telephony is not RingCentral, the connector is built during onboarding when the vendor exposes an API. The shape stays the same: events in, intents out, no audio.

What to ask your vendor

• Does the middleware carry the call, or hand it to the agent’s own phone app?
• Does it ever hold recordings, transcripts, or message bodies, even temporarily?
• Do provider credentials reach the browser, or only the server?
• Who initiates the call: a person with a work item, or a dialer?
• When a transfer or missed call happens in the provider, how does the outcome get back to the queue?