05 / Products
Callcenter/Agentic Customer Service
One customer. One conversation. Every channel.
Callcenter connects voice, WhatsApp, SMS, email, Teams, Slack and Salesforce through an agentic service layer that can understand customer needs, retrieve knowledge, take permitted actions and hand complex interactions to people without losing context.
Conversation
Customers arrive on many channels, stay in one conversation, and the work either resolves or hands off.
The problem
Customer service has become fragmented. A customer may send an email, follow up on WhatsApp, phone later, receive an SMS, speak to an adviser, and have those fragments recorded separately in Salesforce.
Traditional systems often treat each of those as a separate interaction. The customer repeats themselves. Context is lost. Cases are duplicated. Handoffs are thin. Handling times rise and the service feels inconsistent.
Callcenter is designed to treat them as one service journey.
- 01
Send an email
- 02
Reply over WhatsApp
- 03
Phone later
- 04
Receive an SMS
- 05
Speak to a human adviser
- 06
Have the same information recorded separately in Salesforce
- Repeated questions
- Lost context
- Unnecessary transfers
- Poor handoff
- Duplicate case work
- Frustrated customers
- Higher handling times
- Inconsistent service
Three pillars
Resolve what AI can. Route what it cannot. Give people everything they need to finish the job.
Callcenter is an omnichannel agentic customer-service system: one conversation across channels, actions where policy allows, and a complete handoff when a person should take over.
01
Universal Conversation
The channel changes. The conversation does not.
One customer interaction can move from email to WhatsApp to SMS to voice to a human adviser without losing shared state.
02
Agentic Resolution
Answers are useful. Permitted actions finish the work.
Callcenter is designed to retrieve approved knowledge, consult connected systems and take permitted actions — not only to answer FAQs.
03
Warm Handoff
The conversation transfers, and the context transfers with it.
When automation reaches a boundary, a person receives identity, history, summary, actions already attempted and a suggested next step.
The workflow
Understand. Retrieve. Act. Resolve or hand off.
A Callcenter interaction enters through a channel, establishes who is speaking, determines what they need, retrieves the relevant knowledge and records, then either completes permitted work or prepares a person to finish it.
Figure · From channel to resolution
- 01
Channel
- 02
Identity
- 03
Intent
- 04
Retrieve
- 05
Act
- 06
Resolve
- 07
Route
- 08
Handoff
Figure · Many channels. One conversation.
Channels
- Voice
- SMS
Service layer
Universal Conversation
Agent Team
Customer identity and context travel with the interaction.
Outcomes
Resolve
Human handoff
Channels
One service layer. Multiple channels.
Callcenter already integrates with voice, WhatsApp, SMS, email, Microsoft Teams, Slack and Salesforce. Each surface has a job: customers ask for help, people collaborate, and Salesforce holds the case.
Human operators should be able to work from the tools they already use — escalation into Teams or Slack, case updates in Salesforce, a conversation summary pushed into the CRM, and takeover from an internal collaboration interface.
Figure · One service layer. Multiple channels.
Customer channels
The surfaces a customer already uses to ask for help.
Voice
Inbound calls for status, changes, triage and the work that still needs to be spoken.
WhatsApp
Messaging that can continue a case, carry a document, or resolve an exception without a queue.
SMS
Short, timely contact — updates, links, confirmations and callbacks.
Email
Longer written contact that should still join the same conversation, not open a disconnected ticket.
Human collaboration
Where service teams already work when a person needs to take over or approve.
Microsoft Teams
Escalation, specialist consultation, supervisor alerts and takeover from an internal collaboration interface.
Slack
The same collaboration pattern where Slack is the operational workspace.
Customer and case systems
The system of record for who the customer is and what has already been done.
Salesforce
Contacts, accounts, cases, notes and previous service history — retrieved and updated as part of the journey.
Universal Conversation
The channel can change. The conversation stays the same.
At 09:10 a customer emails that a delivery has not arrived. Callcenter investigates. At 10:24 they follow up on WhatsApp. The system is designed to recognise the same customer and the same issue. At 11:03 they phone. The voice interaction already has who they are, what has been checked, which systems were consulted, and what has already been attempted.
That is stronger than saying the product is omnichannel. Omnichannel often means many inboxes. Universal Conversation means one journey that your customer does not have to restart.
Figure · Universal Conversation
Universal Conversation
One shared conversation state. Changing channel should not open a new case.
09:10
Email
My delivery hasn’t arrived.
10:24
WhatsApp
Any update?
11:03
Voice
Customer calls
11:07
Human handoff
Adviser receives the full context
Shared state
- Customer identity
- Intent
- Interaction history
- Retrieved knowledge
- CRM context
- Actions attempted
- Pending work
- Authentication state
- Escalation status
Identity
Know who you are serving.
Callcenter can conceptually reconcile a phone number, a WhatsApp identity, an email address and a Salesforce contact into a common customer context. Identity resolution is subject to your organisational policy. Recognising a number is not sufficient for sensitive actions.
Different interactions require different levels of trust. Some requests need little authentication. Others require a stronger check before Callcenter can expose information or execute an action. The levels are configurable policy, not a single invented authentication technology.
Figure · Identity and trust
Identity resolution
- Phone number
- WhatsApp identity
- Email address
- Salesforce contact
Customer context
Subject to your organisational policy and appropriate authentication. A recognised number is not a completed check.
Step-up verification
01
Public information
Published status, store hours, and other material that does not identify a customer.
02
Recognised customer
A contact identity has been reconciled. Recognition on its own is not authorisation.
03
Authenticated customer
Your configured checks have been satisfied for ordinary account work.
04
Step-up verification
A stronger check before exposing sensitive information or preparing a restricted action.
05
Sensitive action
Work that policy treats as high-risk: a person, an extra check, or both.
Intent and context
Understand more than the last message.
Callcenter should read a contact against previous interactions, account context, urgency, issue complexity, Salesforce data, business rules and model confidence. Sentiment is one signal that may help prioritisation or escalation. It is not treated as a reliable reading of how a person feels.
- Customer intent
- Previous interactions
- Product or account
- Urgency
- Issue complexity
- Salesforce data
- Business rules
- Customer status
- Preferred channel
- Model confidence
- Sentiment signals
Voice
Handle routine calls without the queue.
A Callcenter voice interaction may answer immediately, determine intent, identify the customer, authenticate where appropriate, retrieve information and knowledge, invoke tools, perform permitted actions, confirm the result, and escalate when necessary. The value is resolution. The system is not designed to impersonate a human adviser.
Figure · Voice interaction
- 01
Answer immediately
- 02
Determine intent
- 03
Identify the customer
- 04
Authenticate where appropriate
- 05
Retrieve customer information
- 06
Retrieve organisational knowledge
- 07
Invoke tools
- 08
Perform permitted actions
- 09
Confirm the result
- 10
Escalate when necessary
- Delivery status
- Booking changes
- Appointment management
- Account questions
- Service requests
- Support triage
- Order enquiries
- Basic troubleshooting
- Case creation
Call deflection
Deflect the call by solving the problem.
Call deflection should not mean hiding the phone number, making a human difficult to reach, forcing people through a frustrating bot, or pushing customers into static FAQs.
Deflection should come from resolution, not from making contact harder.
A customer might begin on the website and finish the task in an automated conversation. Or Callcenter may send information by WhatsApp or SMS that prevents the later call. The goal is fewer reasons to call, not simply fewer calls accepted.
Callback
Keep the context while the customer leaves the queue.
A customer can request a callback rather than remain waiting. When the callback takes place, the adviser should already have customer context, the reason for contact, conversation history, a summary, attempted actions and the relevant records. The callback continues the same conversation. It is not a fresh interaction.
Channel switching
Move the conversation to the right channel.
Callcenter can transition a conversation between channels while retaining context. The agent may recommend a different channel if it is more appropriate for the task — “I can send that form to you by WhatsApp now” — without restarting the case.
Figure · Channel switching
Voice
SMS link
Voice
WhatsApp document
Email
Voice callback
WhatsApp
Human voice
SMS
Web workflow
Proactive customer service
Solve problems before customers need to call.
Callcenter is not only inbound. A business event — a delivery delay, a failed payment, an outage, a changed appointment — can open an SMS, WhatsApp or email conversation and let the customer resolve the situation there.
Illustrative workflow: the customer is told a delivery has been delayed and asked whether Tuesday would work instead. They confirm. Callcenter updates the appropriate system and confirms the revised delivery. Workflows of this kind are configured against your systems and policies.
- Delivery delay
- Appointment change
- Outage
- Failed payment
- Upcoming renewal
- Service interruption
- Missing information
- Order exception
Knowledge
Ground answers in your organisational knowledge.
Callcenter can use RAG pipelines and enterprise knowledge sources — help centres, policy, procedures, manuals, Salesforce knowledge and authorised account data. Retrieval should respect permissions, freshness, conflicting sources and policy constraints, and cite evidence where that is useful. This is the same retrieval discipline as Cognitecta Platform.
- Knowledge bases
- Help centres
- Product documentation
- Policy documents
- Service procedures
- Manuals
- Internal documentation
- Salesforce knowledge
- Authorised account data
- Permissions
- Citations where appropriate
- Versioning
- Freshness
- Conflicting information
- Policy constraints
Knowledge-gap detection
Find what your service operation does not know.
If customer interactions repeatedly fail because required information is missing, contradictory, obsolete or difficult to retrieve, Callcenter should be able to surface that. Conversations become signals for improving the knowledge base. The product does not automatically rewrite documents.
- Customers repeatedly ask this question, but no approved knowledge source answers it.
- Two policy documents provide conflicting instructions.
- Escalations frequently occur after retrieval failure.
Actions
Permitted actions, not answers alone.
Where authorised, Callcenter agents can use tools and integrations to perform work. The examples below are illustrative of the kinds of permitted actions a deployment may expose. Action boundaries are governed. Not every tool is available to every conversation.
- Check order
- Update case
- Create case
- Change booking
- Schedule appointment
- Send document
- Update details
- Trigger workflow
- Request approval
- Create follow-up
- Checking an order
- Updating customer details
- Changing a booking
- Creating a support case
- Checking availability
- Retrieving account information
- Sending documents
- Scheduling appointments
- Logging faults
- Initiating permitted workflows
- Issuing routine service requests
- Triggering fulfilment actions
Autonomy and policy
Decide what AI is allowed to do.
Different operations should have different autonomy levels. Callcenter should not treat all tool access equally. You decide whether the system may inform, recommend, act after confirmation, wait for approval, or stay out of the way.
Figure · Autonomy and policy
01
Inform
AI may provide approved information.
02
Recommend
AI may suggest an action but not perform it.
03
Confirm
AI may perform the action after explicit customer confirmation.
04
Approve
AI prepares the action but a person must authorise it.
05
Human only
AI cannot perform the operation.
Intelligent routing
Route based on the problem, not just the menu option.
Traditional call routing asks the customer to press 1 for sales, 2 for support, 3 for billing. Callcenter can route on richer context — intent, product, urgency, account, language, skills, previous contact, channel and why escalation is happening. Where appropriate, the route should be explainable. This is configured policy and available signals, not an invented routing algorithm.
- Customer intent
- Product
- Issue type
- Urgency
- Account
- Language
- Customer status
- Skills required
- Previous contact
- Channel
- Escalation reason
- Team availability
Warm handoff
When a person takes over, do not make the customer start again.
If automation reaches a boundary, Callcenter transfers the interaction without discarding context. The human receives a structured handoff: who the customer is, what they need, what has already been checked, and why a person is required.
The conversation transfers, and the context transfers with it.
Figure · Warm handoff
Illustrative handoff — not a live product screenshot
- Customer
- John Smith
- Current issue
- Order #82381 has not arrived.
- Conversation history
- Email → WhatsApp → Voice
- Identity
- Verified
- Summary
- Customer contacted support this morning regarding a failed delivery. Tracking indicates the parcel is likely lost.
- Reason for escalation
- Replacement value exceeds automated approval threshold.
- Suggested next action
- Approve replacement.
- Customer sentiment
- Frustrated
- Actions completed
- Tracking checked
- Delivery address verified
- Carrier status retrieved
- Replacement eligibility checked
Human takeover
Take over when judgement matters.
A human operator should be able to assume control through Teams, Slack, a contact-centre interface, Salesforce, or another configured operational surface. After takeover, conversation history remains, AI context remains available, tools remain accessible according to role, and the customer stays in the same journey.
Agent assist
Give every human agent an assistant they remain in control of.
After a person takes over, Callcenter can continue to retrieve knowledge, surface history, suggest responses, recommend next actions, draft communications and prepare after-call notes. Suggestions are assistance. They are not always correct. The person retains control.
- Retrieve relevant knowledge
- Surface customer history
- Suggest responses
- Recommend next actions
- Summarise policies
- Find Salesforce context
- Suggest escalation
- Draft customer communications
- Provide an interaction summary
- Automate after-call notes
After-call work
Finish the administration automatically.
After a conversation — whether AI-handled or human-handled — Callcenter can summarise the interaction, update Salesforce, classify the contact reason, record disposition, generate follow-up tasks and identify what remains unresolved.
- Summarise the interaction
- Update Salesforce
- Classify the contact reason
- Record resolution or disposition
- Generate follow-up tasks
- Draft confirmation messages
- Record actions taken
- Identify unresolved items
- Prepare handover notes
Salesforce
Keep customer context connected to the CRM.
Salesforce is not merely another outbound channel. It is part of the customer context and the operational system of record. Callcenter can retrieve contacts and accounts, open and update cases, attach summaries, add notes and use account context when routing. Specific objects and automations follow your Salesforce model.
- Retrieve customer and contact data
- Retrieve account information
- Open cases
- Update cases
- Create activities
- Attach summaries
- Update disposition
- Add notes
- Trigger workflows
- Retrieve previous service history
- Route using account context
Microsoft Teams and Slack
Bring escalation into the tools your teams already use.
Teams and Slack can carry escalation notifications, takeover, approval requests, specialist consultation, conversation summaries and supervisor alerts. A refund that needs manager approval, for example, can appear as a structured request in the collaboration tool the team already watches.
- Escalation notifications
- Human takeover
- Approval requests
- Specialist consultation
- Internal collaboration
- Conversation summaries
- Supervisor alerts
Illustrative collaboration card — interface concept, not a live control
- Customer
- Jane Smith
- Request
- Refund £420
- Reason
- Confirmed failed delivery
- AI recommendation
- Approve
Multilingual service
Serve customers across languages.
Where enabled for a deployment, Callcenter can hold multilingual conversations across channels against a single underlying knowledge base: understand the customer’s language, retrieve approved information, respond appropriately, and hand over with translated context. A human adviser might receive an English summary of a conversation that took place in another language. This is not a promise of universal language coverage.
Sensitive data
Keep sensitive information under control.
Callcenter is designed to operate within configurable identity, permission and data-governance boundaries — PII handling, masking, redaction, logs, transcripts, retention, access and tool permissions. We do not claim compliance certifications here.
- PII handling
- Masking
- Redaction
- Logs
- Transcripts
- Data retention
- Access permissions
- Tool permissions
- Customer authentication
Quality assurance
Evaluate every interaction.
Traditional service quality programmes often review only a sample of conversations. Callcenter should support automated evaluation of AI and human interactions against configured criteria. The questions below are the kind of checks a deployment can run. They are not scores, and they are not a published accuracy claim.
Was intent understood correctly?
Was the answer grounded in approved information?
Was identity verified when required?
Was policy followed?
Did the requested action succeed?
Was escalation appropriate?
Did the customer have to repeat information?
Was the issue actually resolved?
Were unnecessary transfers introduced?
Were sensitive data rules followed?
Supervision
See the operation while it is happening.
A supervisor view can show live conversations and handoffs, emerging issues such as new intents or knowledge gaps, and performance indicators you actually collect. The categories below are a conceptual console. They are not live metrics.
Figure · Supervisor console
Illustrative supervisor view — not live operational metrics
Live
- Active AI conversations
- Active human conversations
- Escalations
- Queues
- Handoffs
- Channel mix
Emerging issues
- Unusual contact volume
- New customer intents
- Repeated failures
- Knowledge gaps
- Negative sentiment trends
- Tool or integration errors
Performance
- Automated resolutions
- Human resolutions
- Handoff rate
- Repeat-contact indicators
- Unresolved conversations
- Common contact reasons
Contact-driver analysis
Understand why customers are contacting you.
Structured interaction data can reveal product defects, delivery problems, confusing billing, broken website journeys, failed internal processes and documentation gaps. Customer service data becomes operational intelligence.
That creates a natural relationship with other Cognitecta products, without making combinations mandatory. Callcenter captures and structures the interactions. Stack can investigate the underlying data. Wargame can explore strategic responses where that is useful.
- Product defects
- Delivery problems
- Confusing billing
- Broken website journeys
- Failed internal processes
- Policy confusion
- Documentation gaps
- Recurring service issues
- Unexpected new topics
Root-cause feedback
Go beyond contact-centre reporting.
The product should be able to frame patterns of the kind below. These are illustrative readings, not claims from a live deployment.
- Contact volume regarding delivery delays increased significantly after the logistics change.
- Many calls classified as billing enquiries originate from one confusing line on the invoice.
- Customers are contacting support because the self-service workflow fails at step four.
Conversation memory
Remember the history that matters.
Subject to retention and data policy, Callcenter can use relevant previous interactions — for example, that this is the customer’s third contact about the same unresolved issue. That can affect prioritisation, escalation, routing and the recommended response. Memory is policy-controlled. It is not indefinite recall.
Multi-agent architecture
A specialist team, rather than a single chatbot.
Callcenter is easiest to understand as a set of specialist investigations: identity, intent, conversation state, knowledge, customer context, actions, policy, routing, translation, summary and quality. That is a way of explaining the operating model. It is not a claim that every deployment exposes these roles as named services.
Figure · Specialist agent model
Conceptual operating model — not a literal runtime topology
Identity Agent
Resolves who the customer is and what authentication state applies.
Intent Agent
Determines what the customer is trying to achieve.
Conversation Agent
Maintains conversation state as the customer moves between channels.
Knowledge Agent
Retrieves relevant approved information from the sources you have authorised.
Customer Agent
Retrieves customer and account context from connected systems.
Action Agent
Executes permitted tool calls inside the autonomy limits you set.
Policy Agent
Checks organisational constraints before information is disclosed or an action is taken.
Routing Agent
Determines when and where to hand the conversation off.
Translation Agent
Supports multilingual service where that is enabled for a deployment.
Summary Agent
Creates structured summaries and handoffs for the next person in the path.
Quality Agent
Evaluates completed interactions against configured criteria.
Architecture
Channels into one layer. Then resolve or hand off.
Voice, WhatsApp, SMS and email enter a universal conversation layer. Identity, intent and orchestration sit above an agent team connected to knowledge, Salesforce, tools, customer data and policies. The path then either resolves or hands off to Teams, Slack, Salesforce or a human adviser — surrounded by identity, permissions, observability, evaluation, audit and guardrails.
Figure · Callcenter architecture
Customer channels
- Voice
- SMS
Universal Conversation Layer
- Identity
- Intent
- Orchestration
Agent Team
Connected systems
- Knowledge / RAG
- Salesforce
- Tools / APIs
- Customer data
- Policies
Resolve
Complete the permitted work and confirm the result inside the same conversation.
Handoff
- Teams
- Slack
- Salesforce
- Human adviser
Surrounding controls
- Identity
- Permissions
- Observability
- Evaluation
- Audit
- Guardrails
The objective is fewer unresolved problems, not simply fewer calls.
Callcenter is designed to improve service outcomes, not simply to suppress contact volume. Immediate response, less queue dependence, better context, fewer repeated explanations and cleaner handover matter because they finish the work. Automate the routine. Keep human judgement where it is needed.
- Immediate response
- Reduced queue dependence
- Routine self-service
- Better context
- Faster resolution
- Fewer repeated explanations
- Better handover
- Reduced after-call administration
- Greater human capacity for difficult interactions
- More consistent service
Escalation
Not every conversation should be automated.
Organisations may choose immediate or early human handling for vulnerable customers, serious complaints, emotionally sensitive issues, uncertain model confidence, high-value or high-risk actions, regulated work, or an explicit request for a person. Escalation is configurable for your operation.
Controlled autonomy, with human judgement where it matters.
- Vulnerable customers
- Serious complaints
- Emotionally sensitive issues
- Uncertain model confidence
- Exceptional circumstances
- High-value decisions
- High-risk actions
- Regulated activities
- Explicit requests for a human
Trust
Automation your customers can understand.
Callcenter is not designed to trick customers into believing they are speaking to a person. Where appropriate, automated interactions should be identifiable as automated. Customers should have a way to reach human help. The product prioritises transparency, predictable behaviour, escalation, auditability and customer control.
Customer service that knows when to act, and when to hand over.
Relationship with Platform
A specialist product on a general-purpose foundation.
Cognitecta Platform is the general-purpose agentic foundation: orchestration, tools, retrieval, memory, permissions and production controls. Cognitecta Callcenter applies those same principles to customer service, support, contact centres and service desks — with the channel model, handoff and CRM context that work requires. That is a statement of family, not a claim that every deployment shares one runtime.
Capabilities
- 01
Universal Conversation
Keep one shared conversation state as a customer moves between email, WhatsApp, SMS, voice and a human adviser.
- 02
Agentic resolution
Use approved knowledge, tools, workflows and Salesforce context to complete suitable work — not only to answer FAQs.
- 03
Warm handoff
When a person takes over, they receive identity, history, summary, actions attempted and a suggested next step.
- 04
Voice handling
Answer routine calls, determine intent, retrieve context and resolve or escalate — without pretending to be human.
- 05
Call deflection
Remove the reason to phone by resolving the problem on a digital channel, rather than by hiding the number.
- 06
Channel switching
Move a live journey to a more suitable channel — a form, a document, a callback — without restarting the case.
- 07
Governed actions
Check an order, update a case, schedule an appointment or trigger a workflow inside configured autonomy limits.
- 08
Intelligent routing
Route on intent, product, urgency, account and skills — not only on a keypad menu.
- 09
Human takeover
Let an adviser assume control from Teams, Slack or another operational interface you already use.
- 10
Agent assist
After takeover, surface knowledge, history, suggested next actions and after-call notes. The person remains in control.
- 11
Salesforce context
Treat Salesforce as the operational system of record: contacts, cases, history and updates, rather than another outbound channel.
- 12
Service intelligence
Use conversations to find knowledge gaps, contact drivers and recurring operational failures — for people to act on.
Use cases
Service work the product is designed to carry.
01
Fragmented omnichannel service
The problem
Customers email, message and phone about the same issue. Each channel opens a new interaction. Advisers ask the same questions again.
How it is addressed
Callcenter holds a shared conversation state so the channel can change without the customer restarting the story.
02
Routine voice resolution
The problem
Queues fill with delivery checks, booking changes and account questions that already have a system of record.
How it is addressed
A voice interaction can identify the customer, retrieve the relevant record, take a permitted action and confirm the result — or escalate with the work already done.
03
Salesforce-connected service desks
The problem
The conversation happens in one place and the case lives in another. Context is copied late, or not at all.
How it is addressed
Callcenter can retrieve and update Salesforce records as part of the same journey, then attach a summary when a person takes over.
04
Proactive exception handling
The problem
Customers call because nobody told them about a delay, a failed payment or a changed appointment.
How it is addressed
A business event can open an SMS, WhatsApp or email conversation and let the customer resolve the exception there.
05
Knowledge and contact-driver improvement
The problem
The same questions escalate because the approved answer is missing, contradictory or obsolete.
How it is addressed
Repeated retrieval failures and escalations become signals for the knowledge base — and, where useful, for Stack to investigate the underlying operation.
Limits worth being clear about
Resolve more of the routine. Escalate better when it matters.
- 01
Not every conversation should be automated. Escalation is a designed outcome, not a failure of the product.
- 02
Authentication and organisational policy govern what can be disclosed or done. Recognising a phone number is not enough for sensitive actions.
- 03
Sentiment is one signal that may help prioritisation. It is not a diagnosis of a customer’s emotional state.
- 04
Conversation memory and retention follow your data policy. The system is not designed as indefinite recall.
- 05
Specialist agents describe the operating model. They are not a claim that every deployment exposes these roles as named services.
- 06
Resolution depends on approved knowledge, connected systems and the actions you have authorised.
Enterprise deployment
These products are delivered as engagements rather than self-serve licences. We start with your problem, your data, and the systems you already run.