Salesforce MVP Hall of Fame (Shivanath Devinarayanan) · Founded 2022

Chatbot to human handoff

One question on every turn: should a person take over now. The bot stops guessing past its depth.

Founded 2022 · Founder in the Salesforce MVP Hall of Fame · Dubai HQ · Kochi studio

  1. 01Models · Claude, Gemini, OpenAI, Bedrock
  2. 02Agentic systems · Agentforce, CrewAI, ADK, A2A
  3. 03Agentic memory · Neo4j, Cognee, IndyKite
  4. 04Automation and voice · n8n, Retell
  5. 05Deployment and governance · identity, evals, policy, audit

The stack

How we would assemble it.

The honest no: if the work is a software factory with a named delivery date, we are the wrong partner.

The decision

The decision, drawn.

Noul per turn 'escalate to human' with threshold

Jev · Noul

Should a person take over this conversation now?

  • yes
  • no

Customer: This is the third time. I want to speak to someone who can actually cancel this.

Noul · yes 0.96

Noul · about $0.0168 per 1,000 decisions, estimated

The flow

One chat, and the turn a person steps in.

  1. Each turn

    The customer writes to the bot

    On the chat platform or in Agentforce.

  2. Jev decides

    Should a person take over now

    Yes or no, with a probability, on every turn.

  3. Below the line

    The bot keeps going

    Claude or Gemini carries the conversation while Jev says no.

  4. Above the line

    A person takes over

    With the conversation so far on the case in Service Cloud.

The job

What has to be true.

Act at 0.85 or higher. Block at 0.15 or lower. Between 0.35 and 0.65, a person decides.

The figure

Builder's figure. Not reproduced by us.

Reference example

TypeSafe quickstart example ('Escalate to a human now?')

www.jevtypesafeai.com/how-to-use

Write your house rules into the request. When rules were left out, Jev got 5 of 24 right with high confidence (Huryn). Validate thresholds on labelled samples.

Who runs it

Who would run this, and what changes.

  • Example

    Head of customer experience, Dubai online retailer

    Today: The chatbot loops on cancellation requests. Customers type "agent" again and again, then leave.

    After: The bot hands over when a person should take it, not when the customer gives up.

    Retail and e-commerce
  • Example

    Guest services manager, Dubai hotel apartments

    Today: Website chat handles check-in questions well. It also tries to handle complaints, and makes them worse.

    After: A complaint or refund request reaches a person on the turn Jev flags it.

    Hotel apartments
  • Example

    Front desk lead, Abu Dhabi clinic

    Today: Patients ask the WhatsApp bot about pain and results, not just appointments.

    After: A question your house rules say the bot should not answer goes to the front desk on that turn.

    AI agents on WhatsApp
Examples of the work, not client stories.

The brief

Name the decision your team makes a thousand times a day.

We read the process and write down where Jev fits, where it does not, and where the line sits.

Where are you?

Founded 2022 · Founder in the Salesforce MVP Hall of Fame · Dubai HQ · Kochi studio

All 100 decisions