---
title: "zAIn — Maximilian Dahl"
description: "Product design case study by Maximilian Dahl: zAIn, the AI assistant in zvoove One. Chat experience, rules for AI behaviour and deliberate limits, live since August 2026."
url: https://max.does.systems/en/work/zain/
lang: en
---

Product Design Case Study · AI assistant · Human-in-the-loop

# An AI assistant that brings the context.

zAIn is the AI assistant in zvoove One. It reads operational data straight from the ERP and answers in natural language. I designed the vision, the UX concept and the whole chat experience.

zvoove · staffing software · 2025 – 2026 · 8 min read

Read the case

Jump to results

Role in the project · Concept & experience · zvoove Germany

Team · PM, AI lead, production engineering · PM: scope and phase model · AI lead: RAG and technology

Responsibility · Vision to chat UI · North star, UX concept, prompting, confirmation and reflection cards, messaging

As of September 2026 · Live since 10 Aug 2026 · +3 · As of September 2026 · Live · Since 10 Aug 2026: GA v1 for existing and new cloud customers · Read-only · 106 of 154 endpoints released, writing only on test and staging data · Test round · 16 customers, August 2026 · Open · Productive write access: legal review open · and 3 more states

As of September 2026

**Live · +3**

As of September 2026

- Live
   Since 10 Aug 2026: GA v1 for existing and new cloud customers

- Read-only
   106 of 154 endpoints released, writing only on test and staging data

- Test round
   16 customers, August 2026

- Open
   Productive write access: legal review open

At first, an agent answered by percentage. zAIn is not a chat window waiting for questions but an assistant with context and clear limits. What was designed was the behaviour, not just the interface.

zAIn

## At a glance

- 16 customers
   in the August 2026 test round
   **16 customers**in the August 2026 test round · GA v1 since 10 Aug 2026 for existing and new cloud customers

- 106 of 154
   endpoints released, read-only
   **106 of 154**endpoints released, read-only · Writing only on test and staging data

- 5 decisions
   from early agent to an assistant with context
   **5 decisions**from early agent to an assistant with context · The five decisions from “At a glance”

1. Problem
   The first agent answered by percentage
   An early chat above the old Launchpad: confidence decided which tier answered.

2. Intervention
   The behaviour is designed, not just the surface
   Chat experience, the three-layer pattern and a handbook of rules for AI behaviour.

3. Outcome
   Live since August 2026, read-only and bounded
   106 of 154 endpoints released for reading, writing only on test and staging data.

Role in the project**Concept & experience**zvoove Germany

Team**PM, AI lead, production engineering**PM: scope and phase model · AI lead: RAG and technology

Responsibility**Vision to chat UI**North star, UX concept, prompting, confirmation and reflection cards, messaging

02 · From agent to navigator

## The first agent answered by percentage.

zAIn was first called zvoove Assistant. The PM lead, the tech lead and I had the idea independently of each other; we got it started together. I set up the initiative team, connected the design to UnityUI, built the first proof and connected the knowledge base via MCP, tested internally first, including as an assistant for customer support.

Before the launchpad there was an early “zvoove Agent”: a chat window over the old launchpad, with a greeting, quick actions and a visible confidence level.

![Early prototype of the zvoove Agent with quick actions and a confidence badge](https://max.does.systems/assets/cases/zain/early-agent.png)

Early prototype: the “zvoove Agent” in the old launchpad, with answer level and confidence (“L1 · 75%”).

Behind it sat a level model: confidence decided which level answers, from a direct answer to a handover to support.

Answer levels and routing · early agent

1. L0
   System and context
   Entry, orientation, setting expectations.
   greeting, quick replies

2. L1
   Known knowledge
   Answers known topics directly.
   confidence ≥ 0.8

3. L2
   Clarification
   Resolves ambiguous input with a follow-up question.
   confidence 0.5–0.8

4. L3
   Generative
   AI-generated help with a disclaimer when confidence is low.
   confidence 0.2–0.5

5. L4
   Handover
   Handover to support or a human for unknown topics or repeated failures.
   < 0.2 or 2–3 failures

Two things came of it later: the percentage left the interface in favour of interpretable uncertainty, and the chat got clearly separated jobs next to notifications and structured UI. The [flagship case](https://max.does.systems/en/work/platform-experience-ai/#experience) covers both.

The PM and I prioritised which cases zAIn had to handle first in around 20 qualitative interviews. In July 2026, customer tests at the innovation roundtable followed.

> zAIn: this could become a powerful tool.
> 
> Customer at the innovation roundtable, July 2026 · translated from German

03 · Chat experience

## Every answer gets a form that fits it.

I designed the complete chat UI and UX: the thinking state, the messages and display patterns for the whole range of answers, from rich text through tables and cards to images and video.

Answer forms · Examples · synthetic data

Text

zAIn · Orders · Cologne

1. How many orders are open in Cologne right now?A

2. There are **12 orders** open in Cologne, 3 of them start this week.

Write a message …

Table

zAIn · Dispatch

1. Who is available as a forklift driver next week?A

2. Three employees match the request:
   Available · forklift licence
   | J. Weber | Forklift licence | from Mon |
   | --- | --- | --- |
   | A. Yilmaz | Forklift licence | from Tue |
   | M. Nowak | Forklift licence | from Wed |

Write a message …

Card

zAIn · Assignments

1. Show me Jonas Weber’s current assignment.A

2. Here is the assignment:
   Assignment · Jonas Weber
   Customer · Nordhafen Logistik GmbH
   Period · 15–30 Sep 2026
   Shift · Early, 6 am–2 pm
   Status · confirmed
   Open in the ERP

Write a message …

Video

zAIn · Help

1. How do I create a new customer?A

2. There is a short video from the zvoove Academy for that:
   Create a new customer · 2:40

Write a message …

On top came confirmation and reflection cards, the prompting and the messaging script: how zAIn speaks and how it offers next steps.

![Assistant in the context of a product view with suggested next steps](https://max.does.systems/assets/cases/zain/assistant-context.jpg)

Assistant in context: zAIn recognises where someone works and suggests fitting next steps.

![Notification center handing a case to zAIn](https://max.does.systems/assets/cases/zain/notification-handover.jpg)

Handover from the notification center, live: the case arrives at the assistant as a prepared prompt.

When zAIn is meant to act, the three-layer pattern separates what it understood, what it intends to do and what came out. Briefly in the [flagship case](https://max.does.systems/en/work/platform-experience-ai/#vertrauen), in depth in the [UWX case](https://max.does.systems/en/work/uwx/#pattern).

Three-layer pattern

1. Message
   Understand
   What zAIn understood becomes visible before it plans anything.

2. Intent
   Plan
   The steps are shown before they run; for write actions, approval sits here.
   approval

3. Result
   Result
   The result stays attached to the case and traceable.

04 · Rules, not gut feeling

## A handbook makes AI behaviour testable.

So that product, design, engineering and AI build the same behaviour, I set up the AI Assistant Handbook, an internal playbook (as of February 2026).

Four layers

1. System
   Hard limits no one overrides.

2. AI logic
   How zAIn reasons and decides.

3. Interaction
   Tone and structure of the conversation.

4. UI
   The visible presentation.

Higher layers constrain lower ones: if the system blocks a request, the UI shows no “try anyway” button.

Every behaviour is written as a rule card: trigger, rule, rationale, example. If a dataset holds more than 500 items, for instance, it is not rendered in full. On top come patterns for recurring situations, such as ambiguity: two or three clear readings as a choice, more than three as a follow-up question. And a catalogue of edge cases, each mapped to its layers.

Rule card · Pattern · ambiguity

Trigger · An input allows more than one reading.

Rule · Two or three clear readings as a choice, more than three as a follow-up question.

Example · Two readings, so a choice.Show me the open orders for Müller.A · Do you mean one of these customers?Müller Logistik · Müller Bau

![AI Assistant Handbook, page “How to Use This Handbook”](https://max.does.systems/assets/cases/zain/handbook-how-to-use.jpg)

One handbook for every role: what product, design, engineering, AI and management need from it.

![AI Assistant Handbook, page “Edge Cases &amp; Exceptions”](https://max.does.systems/assets/cases/zain/handbook-edge-cases.jpg)

Edge cases mapped to the layers, such as conflicting filters or permission boundaries.

I built the AI Assistant Handbook on my own in Figma Make in February 2026; it became the AI Product Playbook.

Learning

A handbook for AI behaviour is a design system for decisions. It is meant to save the discussion that would otherwise start over in every feature.

05 · Limits

## What zAIn cannot do is designed.

At GA, zAIn reads only: 106 of the 154 classified endpoints are released, and write actions run on test and staging data only. The legal review for productive writes into the ERP is still open (as of September 2026).

Three deliberate limits

1. Refuse honestly instead of hallucinating
   v1 deliberately refuses analyses across the whole tenant, because without a search index the answers would be unchecked.

2. Sensitive data technically out of reach
   Salaries, invoices, bank details and reasons for sickness are out of reach in v1, not just hidden. For this I built the three-layer pattern to GDPR rules.

3. Visibility over convenience
   Queries across branches without an explicit visibility check were stopped in the test round.

Declining honestly · Example · synthetic data

zAIn · Reports

1. Which customer had the most assignment hours last quarter?A

2. I cannot answer that reliably in this version yet: reports across the whole tenant need a search index that does not exist yet.

3. I can show you the assignment hours for a single customer.
   Choose a customer

Write a message …

The limits follow the [autonomy boundary](https://max.does.systems/en/work/platform-experience-ai/#vertrauen): read before prepare before change.

06 · After launch

## Measured honestly after launch.

After go-live, the team and I analysed the conversations of the first weeks: recorded prompts, thumbs ratings and free-text feedback. We classified every prompt by whether zAIn answered, asked back or blocked.

Context

What the analysis showed

A large share of the problematic answers came down to a few concrete, fixable mechanics.

Learning

An AI assistant is not finished at launch. The most honest source for the next decisions is the real conversations afterwards.

07 · Outcome

## What is live and what stays open

Go-live, GA v1 · existing and new cloud customers · 10 Aug 2026

Endpoints released for reading · of 154 classified · 106

First test round · August 2026 · 16 customers

### What stays open

1. Write actions in the real system
   Productive writes need the legal sign-off and the human gate in operation.

2. Analyses across the whole tenant
   Need a search index before zAIn may answer them.

3. The fixable mechanics
   The next iterations follow from the post-launch analysis.

fin (for now)

**Case notes · How to read this case.**

Evidence · Confluence documentation on scope and GA, my own prototypes and the AI Assistant Handbook.

Images · Prototypes and product screens; names and data in them are synthetic.

Team · Together with the PM, AI lead and production engineering. The post-launch analysis was team work.

Last updated · As of September 2026
