Use case · Customer-facing
A support agent on your documentation — web chat and WhatsApp — that hands off to a ticket
The agent is grounded on your documentation and, where allowed, your order or account data; it answers in the customer's language, cites the document it used, and says when it doesn't know. Anything outside its scope — a refund, a complaint, a question with no source — becomes a ticket in your helpdesk with the transcript and the customer's contact, and the customer is told a person will follow up. Every conversation is visible as a timeline; the metered unit is the conversation, not the token.
- Who it is for
- Companies and the agencies that serve them with repetitive first-line questions — order status, how-to, policies, hours — arriving on web chat and WhatsApp, and a small support team that wants the hard cases, not the easy ones.
- Systems involved
- ElectroNeek Chat UI / website widgetWhatsApp Business (Meta API)SharePoint / Google Drive docsZendesk / Freshdesk / HaloPSAOrder system (Shopify, ERP API or a bot)
How the run goes
5 steps · logged per run
- Trigger01
Message on chat or WhatsApp
Website widget or Business API
- Agent02
Answer from your docs
Grounded, cited, in the customer's language
- System03
Look up the order
API — or a bot if the system has none
- Person04
Hand-off when unsure
Ticket with transcript
- System05
Conversation logged
Timeline, rating, topics
How it runs, step by step
- 1
Messages arrive from the website widget or WhatsApp Business. A conversation groups them; the agent sees the history and the customer's identity when they've given it.
- 2
The agent answers only from the sources you connected — a documentation folder, a help center, product pages — and from live data tools you enabled: order status through the store API, or through a bot when the order system is a desktop application. Each answer includes the source it used.
- 3
Guardrails are explicit: topics it may not handle (refunds, legal, medical), a confidence threshold below which it hands off, and a maximum number of turns before a person is offered.
- 4
A hand-off creates a ticket in Zendesk, Freshdesk or the PSA with the transcript, the customer contact and the agent's own summary, and tells the customer when to expect a reply. Business hours and the team's language are respected per channel.
- 5
Conversations are reviewed from the timeline — what was asked, what source answered, where hand-offs happened — which is where the documentation gaps show up.
Limits — what this does not do
- The agent answers only from what you connected; a thin help center means a lot of hand-offs, which is useful information but not magic.
- WhatsApp requires an approved Business account; first-contact templates follow Meta's rules.
- It does not take actions on accounts (refunds, cancellations) in this use case; those are separate flows with approvals.
Questions about this process
Grounding on your sources, citations in every answer, a confidence threshold, and a short list of topics it must hand off. The timeline shows exactly which source each answer came from.
Yes — that is the ElectroNeek part: a bot runs the inquiry in the desktop or terminal system and the agent reports the result, with the lookup logged.
Per conversation, with your own LLM key for the model — not per token.
Related use cases
Build this on your systems in a guided sandbox.
Tell us the process; after a 30-minute call the sandbox is ready from 24 hours, and we build the first run with your team.