AI Knowledge Assistants

A Dutch telephony provider: an AI customer-service platform for calls and email

Client: A Dutch business telephony provider · AI platform: agents + dashboard

How Wishfy built a multi-company platform for a Dutch business telephony provider: AI voice and email agents that answer from each company's knowledge, and a dashboard where every customer manages its agents, conversations, escalations and knowledge.

Headline result: AI voice and email agents for several companies, each with its own knowledge, dashboard and data

The customer dashboard: a demo account with sample data, logo removed.
Channels
Phone (voice agent) · Email
Languages
Replies in the customer's language
Platform
Multi-company dashboard, five roles
Knowledge
Texts and PDFs, kept separate per company
Location
Netherlands

The challenge

The provider wanted to offer AI agents to its own business customers, not just use one itself. Each customer needed agents that answer calls and emails from that customer's own information, a place to see every conversation and escalation, and a way to update what the agent knows, without touching the AI workflows behind it.

How a request flows

  1. A call or an email comes in

    Calls go to the company's voice agent; emails arrive in its mailbox. Both reach the same AI assistant.

  2. Non-customer email is set aside

    Sales pitches, vendor offers, auto-replies and research requests are recognised and not answered as if they were customers.

  3. The right company is recognised

    Each company has its own knowledge, agents and history, so the assistant always answers from the right source.

  4. The answer is found

    The assistant searches that company's knowledge base, during the call or before replying. Photos in emails are read too.

  5. Answered, or escalated

    It speaks or writes the answer, or raises an escalation with a priority and alerts the team with a short summary.

  6. Visible in the dashboard

    Calls with transcripts, email threads, escalations and analytics appear in the company's own dashboard.

Diagram: phone calls and emails go to one AI assistant that answers from the knowledge base, replies or alerts the team, and shows everything in the customer dashboard.
How calls and emails flow through the platform.

Who does what

AI agent

  • Answers phone calls through a voice agent, in the caller's language
  • Replies to customer emails from the company's knowledge base
  • Sets aside sales pitches, auto-replies and other non-customer email
  • Reads photos and attachments sent by customers
  • Indexes new knowledge texts and PDFs as soon as they are added
  • Logs every conversation, transcript and attachment per company

Your team

  • Handles escalations, with the summary already written
  • Adds and updates knowledge in the dashboard
  • Invites colleagues and sets their role
  • Tests an agent with a web call before changes go live

Effort moved to the AI agent

Estimate
TaskTypical manual timeNow done by
Reading the email or taking the call~2 minAI agent
Looking up the answer in documents~5 minAI agent
Writing or giving the answer~4 minAI agent
Logging the conversation and attachments~2 minAI agent
Questions outside the knowledge basevariesYour team
≈ 13 minutes of admin per customer question now handled by the AI agent

How we estimate: Our estimate of the typical time each step takes when done by hand for one customer question. It is not a measurement: real times vary per question.

The customer dashboard

  • Overview and analytics: the numbers that matter for each company's agents.
  • Agents: the status of each voice and email agent, and a test call straight from the browser.
  • Conversations: every call with its transcript and every email thread, in one place.
  • Escalations and tickets: what needs a person, by priority, and change requests to the provider.
  • Knowledge: entries and PDF uploads that the agents start answering from automatically.
  • Settings: users, invitations, roles and API keys.

Data kept separate

  • Every company's data is isolated in the database itself; a user never sees another company's conversations or knowledge.
  • Five roles with fine-grained permissions decide who can see and change what.
  • If the knowledge base has no answer, the agent says so and escalates, instead of guessing.

Built with

Next.js and TypeScript for the dashboard, a PostgreSQL database with row-level security and vector search, n8n workflows, OpenAI language models and embeddings, a voice agent platform for calls, Gmail, and error monitoring.

Start a project

Have a business problem AI could help solve?

Tell us what you're trying to improve. We'll help you identify a practical next step.

Discuss your project