Pulse: Customer Support AI for Nucleus BankOS
Support AI that can see the account.
Most support AI is a chatbot bolted to a help desk. Pulse is a blended human + AI workforce running on the bank’s own core — where the AI can actually see the account.
Capacity from AI. Control by design.
Support · Blended workforce
One cockpit · human + AI
18
42
Live sentiment
Three years of bolting chatbots on
A deflection widget with a confidence problem.
The bot deflects the easy questions, fumbles the hard ones, hands off to a human with no context, and leaves compliance with no record of what it told the customer. The agent still does the real work — now with an extra layer of cleanup.
A chatbot bolted on
- Deflects the easy questions, fumbles the hard ones
- Hands off to a human with no context
- Leaves compliance with no record
- The agent still does the real work — plus cleanup
A blended workforce
- AI and human agents work the same queues, same core
- AI handles what it should, copilots the rest
- Clean handoff the moment a person is needed
- Every action grounded, scored, and logged
Pulse turns on top of the bank’s baseline support module — subscribe and get the AI workforce; don’t, and keep the standard desk. Same data, same cases, more capacity.
What it does
A support workforce — part human, part AI, all governed.
Command center
A command center for a blended workforce.
One cockpit shows human and AI queues side by side, with a real-time sentiment overlay — so supervisors can see not just how many cases are open, but how customers feel, and route the tense ones to the right place.
- Human + AI queues, one view
- Real-time sentiment overlay
- Route tense cases to people
- Live capacity at a glance
Support · Blended workforce
One cockpit · human + AI
18
42
Live sentiment
AI agent fleet
Customer-facing bots and human copilots.
Pulse runs a roster of AI agents in two roles: customer-facing bots that resolve routine cases end to end, and copilots that ride alongside human agents. A live-assist monitor lets supervisors watch the copilots work in real time and override when needed.
- Customer-facing resolution bots
- Copilots for human agents
- Live-assist supervisor monitor
- Override any time
Teller Bot
Customer-facing
Card Bot
Customer-facing
Dispute Copilot
Human copilot
Fraud Copilot
Human copilot
Human-in-the-loop
Human-in-the-loop, enforced.
When an AI agent reaches the edge of its competence, it escalates. Those interventions are a first-class surface with a built-in guardrail — the handoff carries full context, and a person owns the outcome. The AI never quietly makes a call it shouldn’t.
- Escalation is a first-class surface
- Handoff carries full context
- A person owns the outcome
- No quiet, unsupervised calls
“This needs a human — escalating with full context.”
Full transcript, account, and intent carried over.
Human-in-the-loop guardrail, built in.
Automation playbooks
Automation playbooks that actually resolve.
Beyond chat, Pulse runs automated resolution playbooks — triage, disposition, customer notification — with a full run history. Repetitive, well-understood cases get handled start to finish; novel ones route to people.
- Triage → disposition → notify
- Full run history
- Routine cases resolved end to end
- Novel cases routed to people
Resolution playbook · run history
Grounded knowledge, watched for drift
A RAG knowledge base grounds every AI response in the bank’s own policies and procedures — and a drift monitor flags when the knowledge, or the model’s use of it, starts to wander, so answers stay accurate as products and rules change.
AI quality assurance on every case
Instead of QA sampling 2% of calls, Pulse auto-scores case review across the board, surfaces deviations, and feeds agent coaching — for both the human and AI sides of the workforce.
Analytics tuned to the AI tier
Deflection vs. resolution, intervention rates, sentiment trends, automation coverage, and quality scores — the measures a blended workforce actually needs.
The proof point
Deploying a support chatbot, or operating an AI-native workforce.
Because Pulse runs on the core — and the guardrails are built in — the bank gets AI capacity without giving up control.
It sees the real account
The AI agent answering a customer sees the real account, the real transaction, the real relationship — not a sanitized snippet passed through middleware.
Human in the loop by design
The handoff carries full context and a person owns the outcome. The AI never quietly makes a call it shouldn’t.
Grounded and monitored
Every answer is grounded in the bank’s knowledge and watched for drift, so accuracy holds as products and rules change.
Scored and logged
Every automated action is scored for quality and logged — capacity from AI without giving up control.
Part human, part AI, all governed.
Pulse is the AI-augmented tier of your support operation — it turns on over Nucleus BankOS’s baseline support desk, on the same data and the same cases. See it on your own data, or join the founding-partner cohort.