The platform
One pipeline from bounceto paid instalment.
Collections Hub is not a voice agent with a dashboard around it. It is the operating system for a collections floor: it decides which accounts to work, on which channel, at what hour and under whose authority, then carries that decision out and keeps the evidence.
The problem
The profit lever isdelay, not dialogue.
A customer who misses an instalment on Day 1 and hears from you on Day 12 has already reclassified the debt in their own mind, and in your book. Every hour between the event and a structured intervention raises the chance of rolling into the next bucket, and almost none of those hours are spent talking.
| The shift | What the hours go on today | What happens instead |
|---|---|---|
| Before the shift | Log in, take last night's allocated list, skim the broken promises and callbacks. | A bounce at 00:12 opens a case straight away, and the written notice is queued for the first permitted hour. Nothing waits for an allocation meeting. |
| Dialling | Progressive dialler. Verify identity, disclose purpose, state the amount and the due date. | Contact is planned to an hour the customer has historically answered, inside the permitted window, under the cap. |
| On the call | Hang-ups, I-will-pay-Saturday, fee arguments, already-paid, waiver requests, a language switch. | Identity comes first, then the promise. Anything outside policy goes to a person with the account packet already open. |
| After each call | Notes, promise amount and date, callback, dispute flag, CRM update: the after-call work. | The outcome is filed as a code, not a note, and the promise and its pay link are sent in writing during the call. |
| Afternoon | Retry the no-answers, chase today's promises, take the inbound you-called-me. | The promise carries a pay link and a reminder. When it breaks, the next action is decided that hour. |
| End of shift | Assemble the report: attempts, contacts, promises, collected. Tomorrow's meeting asks why roll-forward moved. | Recovered amount, promises kept, time to first touch and compliance flags are live on the dashboard, not assembled overnight. |
The chore column is the job as it is actually run on a collections floor. Killing after-call work is what lets the next attempt happen now rather than tomorrow.
Positioning
Voice is a channel.The decision is the product.
Voice AI is becoming a commodity, and a product priced by the minute has to want more minutes. The defensible layer is the one that decides whether a contact is worth making at all. After that, adding a channel, a mandate re-presentation, a person or a field visit changes nothing upstream.
Decide
Which accounts, which action, which channel, which hour, scored in money against doing nothing.
Gate
Consent, calling window, contact caps across every channel, holds, identity and stale data.
Execute
Voice, WhatsApp, a written notice, a person on the phone, a pay link, a mandate retry, or deliberate silence.
Evidence
Rules in force, every rule consulted, score, recording, transcript. One record per action.
Learn
Rates re-learned every night, a comparison group, and a promotion gate that says no until the evidence says yes.
The loop
From a failed mandateto a written promise.
One account, one morning, with nothing waiting for a person to notice it.
- 01
The event arrives, not the batch
A mandate fails and the bounce reaches Collections Hub as a signed event, with its return reason. The case opens immediately. Account and payment data still arrive in your regular bank feeds, and if a feed is late, contact waits instead of guessing.
- 02
The engine picks an action
Ten candidate actions, from waiting to a legal notice, are each planned to a specific moment, then vetoed, then scored in money. Waiting scores exactly zero, so anything that happens has to beat silence.
- 03
The gates decide whether it may happen
Consent, calling hours, contact caps, third-party contact, holds and stale data are checked before anything is dialled, and the calling window is checked again at the dial. A vetoed action cannot be revived by a high score.
- 04
The agent has the conversation
Identity first, then the amount and the due date, then the actual conversation. Anything outside policy, every settlement request included, goes to a person with the account packet already open.
- 05
The promise becomes an artefact
Amount and date are captured on the call, confirmed in writing with a pay link while the customer is still on the line, and a reminder is set for the day it falls due.
- 06
The loop closes itself
Kept or broken is worked out from the ledger, not from anyone's honour. A broken promise goes back to the engine the hour it breaks, with a cap of five attempts and a twelve-hour backoff, so the ladder stops instead of grinding.
The scoreboard
The numbers a collectionshead is actually judged on.
Not conversation quality. Coverage, speed, and whether the promise turned into money.
Hours
Time to first touch after a failed payment
Tracked on the dashboard as a median, from the bounce to the first compliant touch. Days is the industry norm.
Right party
Contacts that reached the actual borrower
Verified on every call before any account detail. Attempts per connect and right-party rate are measured, not guessed.
100%
Of conversations scored
Voice Studio calls get the full grading, and every other conversation still gets a rules-based card, so the complained-about call is on record.
Kept
Promises, not promises made
Capture without a keep rate is vanity. Keep and break are computed from the ledger.
Every recovery call we made in March, with the rule set that authorised it, the number it came from, who answered, what was said, what we promised, and whether we kept it. As one query.
Who does what now
Replace the volume.Keep the judgement.
Automation belongs where the work is high-volume and low-discretion. Hardship, disputes, settlements and complaints are not those, and the platform is built to route them to a person quickly rather than to argue with them cheaply.
- Tele-collections
- Replaced at volume in pre-due and early delinquency. Coverage stops being a sampled list and becomes the whole book.
- Floor lead
- Augmented. Floor command lets a lead listen to, whisper into or take over any live call, so supervision goes to the exceptions instead of the walk-around.
- QA analyst
- Sampling replaced with complete scoring. Analysts calibrate the rubric and coach the flags instead of listening to a fraction of the day.
- Consent and do-not-contact desk
- The spreadsheet replaced by hard gates. Frequency is capped across voice and messaging together, so persistent contact is not reachable.
- Back-office clerk
- The diary replaced by shared queues with owners and SLA timers: broken promises, callbacks, document requests and disputes.
- Hardship, legal and field specialists
- Not replaced. Fed the same hour, with the transcript, the promise history and the clock that matters.
The product
39 modules.One governed pipeline.
Every layer writes to the same customer record and the same evidence log, which is the difference between a platform and a folder of tools. See what each module does.
Live operations
What the floor is doing right now, on every channel at once.
- My workspace
- Conversation inbox
- Handoff hub
- Floor command
CRM and resolution
The account, its history, and every path to a cure.
- Executive dashboard
- Customer 360
- Promise to pay
- Disputes queue
- Document desk
- Callbacks
- Upsell and leads
- Decision intelligence
Compliance and QA
Proof, on demand, for every action the platform took.
- Audit trail
- Compliance risk
- Consent and do-not-contact
- Redaction and export
- QA scorecards
- Bot analytics
Voice Studio
Build, rehearse and release the voice and WhatsApp agents. Drafts can come from Claude or any MCP client; a person still publishes.
- Voice agents
- Campaigns
- Agent runs
- Knowledge base
- Call recordings
- Audio library
- Tools
- Models
- Guardrails
- Checks
- Telephony
- Agent routing
- Releases
- Reports
- Developers
- Studio settings
- Help
Bot configuration
The plumbing around the agents: what connects, what it costs, and who may see it.
- Webhooks
- Billing and usage
- Roles and access
- Settings
Questions
Before you bookhalf an hour.
Is Collections Hub a voice AI product?
No. Voice is one channel. Collections Hub decides which accounts to work, on which channel, at what hour and under whose authority, then carries that decision out: by message, by voice, by a person, or by doing nothing at all. Positioned as a voice product, the most valuable behaviour in the system would be the one it punishes: not calling.
How long does an implementation take?
Collections Hub reads your book through one contract for accounts, instalments, mandates, payments, consent and the contact ledger. Mapping your systems onto it is the long pole, usually four to six weeks. Agents, rules and scorecards are configuration, so they move at the speed of your compliance review, not ours.
Does it replace our collections CRM?
It can, and most deployments let it. Collections Hub ships the queues a floor actually runs on (promises, disputes, callbacks, document requests and audit), so the alternative is a second system of record and a reconciliation problem. Where a CRM has to stay, Collections Hub sends it signed events for promises, disputes, payments and call outcomes.
What happens to our existing collectors?
They stop dialling the early book and start taking the work that needs judgement: hardship, disputes, settlements and complaints. Those arrive with the transcript, the ledger and the account's limits already on screen.
See a bounce becomea conversation.
Thirty minutes, your book, your rules. We run one account end to end: the trigger, the gates, the decision, the conversation and the record it leaves behind.