RMT Engineering Logo
Inbound contact centre

Inbound Contact Centre

Everything that happens to a query once the customer has initiated, it arrives on voice or any channels of customers preference, AI answers before the Client is put queue, and when the call is handed over to an associate the AI driven transcript takes over rather than asking the customer to begin again. This is a inbuild feature of OptiML CX Platform.

  • One queue across every channel
  • AI answers before the queue
  • The handover carries the context
We reply within one business day. No newsletter.
8+ conversation channels on one model, with voice as the ninth mode of communication
2x of interactions scored on 15 named dimensions, AI-handled and human-handled alike
99.9% report definitions, filterable by channel, team, agent, queue, campaign, date, sentiment, score and disposition
100% Of interactions scored, AI and human

Inbound Calls Smart Decision Making

Inbound call handing is where a Contact Centre needs to demonstrate maximum efficiency. This is where there could be solutions like IVR, Chat, Quality, Data Sheets are all holding together. The OptiML CX Platform treats the whole path as one record instead. Thus, arrival of a call, AI based response, handover of a call to live floor and wrap-up are separate modules all processed under on one conversation resulting an integrated output to the Client and the associates

How it fits together

The path a conversation takes through Inbound Contact Centre

An inbound contact moving from arrival through AI answering and handover to wrap-up, with one conversation record underneath all four stages 1 · Arrival voice, plus 8 chat channels Omnichannel 2 · AI answers grounded, cited, 20+ languages Voice AI · Knowledge 3 · Handover transcript, summary, sentiment Agent Desk · Supervisor 4 · After wrap-up scored, ticketed, reported Quality · Ticketing One conversation record unified customer key · phone, email, your own identifiers
By the numbers

Few Facts - at a glance

  • 8+

    chat channels on one conversation model, with voice as the ninth surface

  • 100%

    of interactions scored on 15 named dimensions, AI-handled and human-handled alike

  • 75+

    report definitions, filterable by channel, team, agent, queue, campaign, date, sentiment, score and disposition

The process

How It Works

The four stages, end to end

Four things happen to an inbound contact, and each of them belongs to a module that can be read about on its own. This is the order they happen in, and what is handed to the next stage.

  1. The contact arrives

    Voice, or any of the eight chat channels: WhatsApp, web chat, SMS, Messenger, Instagram, Telegram, Discord and email. All of them open the same conversation record inside one universal inbox with shared routing, memory and ownership. A unified customer key ties the person together across phone, email and your own identifiers, so the WhatsApp message at nine and the call at eleven are one history rather than two strangers.

  2. The AI answers first

    The customer says what they want in their own words, and intent is read from that instead of from a keypad tree reciting six options. Answers are grounded on your own content and cited back to it. Over 20 languages are covered, including 22 Indian languages through Bhashini, detected and switched mid-conversation rather than fixed at the moment the call connects.

  3. Handover to a person

    When a person is needed, a bundle moves with the conversation: the transcript, a summary, the extracted entities, the sentiment trajectory and the reason for handover. Assignment then runs on skill, priority, geography and availability, and presence carries per-channel capacity, so an agent holding three chats and one call is described accurately rather than by a single concurrency number.

  4. After wrap-up

    The disposition, the notes, the tags and the extracted entities come back drafted from what was actually said, and the CRM write-back is idempotent, so a retry after a dropped connection updates the record instead of creating a second one. A ticket is raised where the case outlives the conversation. Every interaction is then scored, AI-handled and human-handled, on the same 15 dimensions.

Features

Every capability you need in one module

1. One queue, every channel

The channel a customer picks is a detail of their morning. It should not become a fact about how your operation is structured.

Eight chat channels carry WhatsApp, web chat, SMS, Messenger, Instagram, Telegram, Discord and email, and voice sits beside them as the ninth surface. The two counts are worth keeping apart — they are not the same number, and a report that conflates them will not reconcile. All of it lands on one conversation model behind a single universal inbox, sharing routing, memory and ownership. The unified customer key resolves the same person across phone, email and custom identifiers, which is what stops somebody explaining a problem twice because they moved from chat to a phone call halfway through.

  • WhatsApp, web chat, SMS, Messenger, Instagram, Telegram, Discord and email
  • Voice as the ninth surface, on the same conversation record
  • One universal inbox with shared routing, memory and ownership
  • A unified customer key across phone, email and custom identifiers
WhatsApp Web chat SMS Messenger Instagram Telegram Discord Email Voice

Omnichannel Engagement, where the channels converge

The voice path is yours

Voice reaches the platform through LiveKit, Twilio, Telnyx or Vonage, or over SIP ingress and egress to a carrier you already hold a contract with, with a WebRTC media pipeline behind it. The numbers themselves stay on your carrier account.

Cloud Telephony, the voice plumbing underneath

2. Answer before the queue

The cheapest contact is the one that never needed a person, provided the answer it got was right.

Callers state the reason for the contact in their own words and intent is read from that. Answers come from your own knowledge sources with citations attached, and three checks sit between the model and the customer: a groundedness gate, a hallucination grader and a citation auditor. Sources attach per agent with no organisation-wide fallback, so a billing agent cannot quietly answer out of the HR handbook because nothing better was to hand. Over 20 languages are supported, 22 Indian languages among them through Bhashini, and the language is detected and switched mid-conversation rather than chosen once at the start.

  • Natural-language intent instead of a keypad tree
  • Answers grounded on your own content, with citations attached
  • Groundedness gate, hallucination grader and citation auditor
  • Sources attached per agent, with no organisation-wide fallback
  • 20+ languages, including 22 Indian languages through Bhashini
  • Language detected and switched mid-conversation
Grounded answers Citations 20+ languages Bhashini No blanket source fallback

Voice AI, the agent that answers the call

Knowledge and RAG, where the answers come from

Replacing an IVR menu with a question

Tier-1 automation, in more detail

3. Routing and assignment

A contact reaches a person on skill, priority, geography and availability in combination, rather than on whoever has been idle longest. Presence carries per-channel capacity, so an agent can be full on chat and still able to take a call, and a supervisor can tell which of the two is actually full. What the agent opens is not a blank screen: the handover bundle arrives with the conversation, and the customer history sits under the same key. On the way out, the wrap-up is drafted from what was said and written back to the CRM idempotently.

  • Assignment on skill, priority, geography and availability
  • Presence with per-channel capacity, not one blanket concurrency number
  • Handover bundle: transcript, summary, extracted entities, sentiment trajectory, reason for handover
  • Wrap-up drafted as disposition, notes, tags and entities, with idempotent CRM write-back
Skills Priority Geography Availability Per-channel capacity

Agent Desk, the screen the conversation lands on

How this looks on a contact centre floor

4. The live floor

Queue depth, agent state and every live conversation, redrawn on a two-second cycle, in grid, list, wallboard or manager-dashboard views. When one of them turns, there are three ways in: monitor silently, whisper to the agent alone, or take the conversation over. All three work on voice and on chat. Every monitor, whisper and takeover is recorded with the supervisor who did it and the time they did it, which is the difference between supervision and listening in.

  • Two-second refresh across queues, agents and live conversations
  • Grid, list, wallboard and manager dashboard views
  • Silent monitor, whisper and full takeover, on voice and on chat
  • Every intervention recorded with supervisor and timestamp
2-second refresh Wallboard Silent monitor Whisper Takeover Audited

Supervisor Console, the live floor in full

Analytics, once the shift is over

5. After the conversation

Two things outlive an inbound contact: a judgement about how it was handled, and a case somebody still has to close.

Scoring covers 100% of interactions, AI-handled and human-handled, voice and chat, against 15 named dimensions: per turn while the conversation is open, then across the whole conversation at wrap-up. It reads the transcripts already stored, so there is no second recording pipeline to keep alive beside the one you have. Where the case outlives the conversation it becomes a ticket, with SLA targets set per channel and per priority, escalation triggered on an SLA breach, a sentiment drop, a keyword or a repeat contact, and a resolution email carrying a CSAT or NPS survey.

  • 100% of interactions scored on 15 named dimensions, AI and human alike
  • Scored per turn while the conversation is open, then per conversation at wrap-up
  • No second recording pipeline: scoring reads the transcripts already stored
  • SLA targets per channel and per priority, with escalation
  • Triggers on SLA breach, sentiment, keyword and repeat contact
  • Resolution email with a CSAT or NPS survey
15 dimensions 100% coverage SLA per channel Escalation triggers CSAT or NPS

Quality Management, and the 15 dimensions

Ticketing and case management

6. Staffing the queue

Forecasting runs Erlang-C by interval, channel and skill at 15-minute intervals, adjusted for the AI containment actually measured rather than the containment quoted in a business case. That adjustment is the part most inbound forecasts get wrong once automation lands, because the arrival pattern reaching a person stops matching the arrival pattern at the front door. Real-time adherence then compares the roster with what the floor is doing and raises a deviation alert when the two part company.

  • Erlang-C by interval, channel and skill
  • 15-minute intervals
  • Adjusted for measured AI containment rather than forecast containment
  • Real-time adherence with deviation alerts
Erlang-C 15-minute intervals Containment-adjusted Adherence alerts

Workforce Management, forecasting and rosters

The hours nobody wants to roster

Overnight and weekend volume is the usual first place a team lets the AI answer alone, because the alternative is a rota nobody volunteers for. The same grounded answering runs at 03:00 as at 15:00, and anything it cannot finish queues for the morning with the transcript attached.

After-hours and overflow cover

Use Cases

Where Inbound Contact Centre delivers value

Voice AI with grounded answering

Retiring the keypad menu

A utility running six-option IVR

Scenario

Six options, two of them leading to submenus, and a steady stream of callers pressing zero to escape. The IVR tree is replaced by a question. The caller says and the intent is read from that, and the answer is grounded on the utility's own published tariff pages with the attached citation.

Outcome

Callers who wanted one fact get it without joining a queue. The ones who needed a person reach one with the reason for the call already written down, so the agent opens on the problem rather than on the menu path.

Knowledge-grounded answering with a groundedness gate

Tier-1 questions that never reach a person

A subscription retailer with a small support team

Scenario

Delivery windows, the returns policy, where an order has got to. The knowledge sources are attached to the support agent and to nothing else, so the answers come from the policy pages and the order lookup, with no organisation-wide fallback to wander into when a question is unfamiliar.

Outcome

The questions with a documented answer get one, cited. The questions without a documented answer hand over instead of being improvised, which is the behaviour that makes the automated ones worth trusting.

Containment-adjusted forecasting with escalation triggers

After hours, and the Monday spike

A regional insurer with a nine-to-six floor

Scenario

Overnight contacts on WhatsApp and web chat are answered by the same agent that works during the day. Anything it cannot finish raises a ticket with the transcript attached and an SLA timer set for the priority it was given. The Monday forecast is then built on containment as measured overnight, not as hoped for.

Outcome

Fewer people wait until nine to ask a question that had a documented answer at two. The roster is built against the volume that will actually reach a human, and adherence alerts show when the shift drifts away from it.

Language detection and switching mid-conversation

A caller switches language mid-conversation

A bank serving several states in India

Scenario

A call opens in English and moves into Tamil once the customer starts describing the actual problem. Language is detected and switched during the conversation instead of being fixed when the call connected, drawing on 20+ Indian languages.

Outcome

Nobody is transferred to a different queue because they changed language. If the conversation does hand over, the transcript carries what was said in both, and assignment can weigh geography alongside skill.

Security & Compliance

Enterprise security by design

Recordings made unreadable

Recordings are sealed with AES-256-GCM envelope encryption under a key per recording, with the tenant key held by you in your own KMS. Traffic is TLS in transit and mTLS between services. Custody is the part of a security review that cannot be argued around: either the data can be made unreadable without first raising a request with RMT, or it cannot.

Database Isolation

Tenant isolation is a PostgreSQL row-level security policy on 231 tables with fail-closed tenant context, rather than a WHERE clause each query is trusted to remember. For a BPO running several client programmes across one inbound floor, that is what makes an audit answerable one programme at a time.

Regional compliance packs

16+ regional compliance packs cover the jurisdictions an inbound floor is usually asked about, switched on per tenant rather than bolted on afterwards. The region the data rests in is chosen at deployment and does not move because a queue was added.

Forensic proof

Every supervisor intervention on a live conversation is written to an immutable audit trail with integrity proofs, carrying the name of the supervisor and the timestamp. Months later a reviewer can reconstruct what was done to a conversation rather than relying on somebody's account of it.

At a glance

Specification

The numbers and limits, without the sales copy

Specification for Inbound Contact Centre
Specification Detail
Channels Eight+ chat channels (WhatsApp, web chat, SMS, Messenger, Instagram, Telegram, Discord, email), plus voice as the ninth surface, all on one conversation record
Languages 20+, including 22 Indian languages through Bhashini, detected and switched mid-conversation
Routing Skill, priority, geography and availability, with presence carrying per-channel capacity
Interventions Silent monitor, whisper and full takeover on voice and chat, each recorded with supervisor and timestamp
Recording AES-256-GCM envelope encryption, a key per recording, customer-held tenant key in KMS, TLS in transit, mTLS between services
Quality coverage 100% of interactions, AI-handled and human-handled, on 15 named dimensions, per turn while open and per conversation at wrap-up
Reporting 76 report definitions, 6 export formats, filters on channel, team, agent, queue, campaign, date, sentiment, score, disposition and tag
Uptime 99.9% monthly commitment with a published credit schedule, multi-region failover, autoscaling, verified restores
Hosting The region you choose, PostgreSQL row-level security on 231 tables with fail-closed tenant context, 16+ regional compliance packs, immutable audit with integrity proofs
Telephony LiveKit, Twilio, Telnyx and Vonage, SIP ingress and egress, WebRTC media pipeline. You bring the carrier
FAQ

Questions,
answered

What teams ask us before they roll out Inbound Contact Centre — how it works, what it needs from your side, and what happens when it gets something wrong

Still not sure?

Talk to a specialist and get a straight answer.

Ask our team

It replaces the menu, not the routing. Instead of a keypad tree, the caller says what they want and intent is read from that, with the answer grounded on your own content and cited to it. The routing rules stay: assignment still runs on skill, priority, geography and availability. What goes is the part where somebody presses 3, then 2, then 0, and then describes the problem again to the person who eventually answers.

Yours. OptiML connects through LiveKit, Twilio, Telnyx or Vonage, or over SIP ingress and egress to a carrier you already have a contract with, with a WebRTC media pipeline behind it. We are the software layer above your voice path rather than a reseller of it, so the carrier relationship and the commercial terms attached to it stay where they are, and the numbers themselves stay on that carrier account.

It hands over, and the handover carries the work already done: the transcript, a summary, the extracted entities, the sentiment trajectory and the reason it stopped. Whoever picks the conversation up reads what the customer said rather than a paraphrase. The groundedness gate triggers this more often than a failure to understand does. If there is no source behind an answer, the AI is not permitted to produce one, and sources attach per agent with no organisation-wide fallback to reach for.

No. The OptiML CX Platform is 14 modules on one record, licensed per module, and inbound is a path through several of them rather than a product of its own. Most teams start with one module, usually the channel that is hurting most, and add the next once the first is running. There is no rip-and-replace step in the middle, because the modules already share one conversation model underneath.

In the region you choose at deployment. Tenant isolation is enforced inside PostgreSQL by row-level security on 231 tables with fail-closed tenant context, so the database declines another tenant's row whatever the application asks of it, and 16+ regional compliance packs cover the jurisdictions an inbound floor tends to be questioned about. RMT operates the platform against a 99.9% monthly uptime commitment with a published credit schedule, multi-region failover, autoscaling and verified restores.

On all of them. 100% of interactions are scored, AI-handled and human-handled, voice and chat, against 15 named dimensions: per turn while the conversation is open, then across the whole conversation at wrap-up. Scoring reads the transcripts the platform already stores, so nothing is uploaded and there is no second recording pipeline. A 2% sample scored a fortnight late tells you about the sample — this tells you about the operation.

Keep exploring

Additional supporting modules

Native modules that share the same record, knowledge and governance

Platform module

Voice AI Agents

The agent that answers before the queue: grounded answers, citations, and languages switched mid-conversation.

Explore
Platform module

Omnichannel Engagement

Eight+ conversation channels and voice on one model, with a unified customer key behind them.

Explore
Platform module

Supervisor Console

The live floor on a two-second cycle, with silent monitor, whisper and takeover, all of it audited.

Explore
Platform module

Workforce Management

Erlang-C by interval, channel and skill, adjusted for the containment you actually measure.

Explore
Talk to a specialist

Bring one inbound queue

Pick the queue that hurts most on a Monday and send us what it deals with: a recording, a chat transcript, the policy page the answers ought to come from. We will show what the AI would have answered with the citation attached, what would have handed over, and what the person taking it over would have had on the screen in front of them.

  • Start with one module
  • No rip-and-replace
  • Hosted in the region you choose

Book your slot

Leave your email and our team will come back to you within one business day.

or reach us directly
Call our team Email us

Your details stay private. We never share them.

This website uses cookies.

Cookies are small text files that allow us to create the best browsing experience for you on our site. By continuing to use this website or clicking "Accept & Close", you are agreeing to our use of cookies. To understand how we use cookies or how to manage them, please see our cookies policy.

Ask OptiML

Powered by RMT Engineering