FLZR
Sales agents pick up every chat the AI assistant cannot answer
We built FLZR a platform where each organization runs its own AI sales chatbots and its own sales team. Admins set agent hours per chatbot, and agents pick up handed over chats in a shared inbox.
Retail & E-commerceMulti-tenant chatbot platformBuilt from zero
The challenge
Sales teams wanted product questions answered automatically, with a clean handoff to a person when the bot could not help, set up by each organization itself.
What we did
A multi-tenant platform with four roles: admins set up chatbots, knowledge and agent hours, and the assistant hands chats it cannot answer to agents in a shared inbox.
The outcome
We delivered every item in the MVP scope, including agent sign in, dashboard, chat inbox and chat with an agent, and reviewed it with the client.
- MVP scope
- Every scoped item delivered, for all four roles
- Roles
- Super admin, organization admin, agent, end user
- Agent hours
- Set per chatbot, by weekday and time zone
- Handoff
- Assistant to sales agent, in a shared inbox
The starting point
FLZR, a Berlin company that runs point of sale sales services, asked us to build an AI sales assistant platform as a new product. Many organizations use one platform, each with its own chatbots, product knowledge and sales agents.
What had to change
Product questions had to be answered automatically, with a clean handoff to a human sales agent when the bot could not help. Every organization had to configure its own chatbots, knowledge and agents.
What could not break
- Several organizations share one platform, each with its own chatbots and knowledge
- Four roles, from super admin to end user, each reaching only what the role allows
- A shopper the assistant cannot help reaches a human sales agent
- Outside business hours, shoppers see the organization's own message instead of waiting
The calls we made
We treated the assistant as half of the product: the other half is the people who set it up and the agents who take over, so tenancy, roles and the inbox were in the wireframes before any AI feature was built.
- 01
Scope agreed before the build
A scoping and design phase produced wireframes for all four roles, walked through with the client to find gaps in user flows, then Figma designs, user stories and a system flow chart, so both sides agreed the full scope before the build.
- 02
Tenancy and access control first
The first sprint set up the schema, middlewares and access control with super admin and organization admin sign in, so every later screen respected organizations and roles.
- 03
The human side designed with the bot
Agent availability, a shared inbox and a dashboard were in the first wireframes, and an escalation flow in the conversation service hands a chat it cannot answer to a person.
- 04
Agent hours set per chatbot
Admins turn on business availability for a chatbot, pick a time zone, set hours for each weekday and assign agents. Overlapping hours cannot be saved, a chatbot needs at least one agent, and off hours shoppers see the admin's own message.
- 05
Setup each organization runs itself
Organization admins create a chatbot in six steps, from customization, pre chat form and agent availability to knowledge base, quick prompts and website integration, and test it in a live chat before copying the embed script.
How the system fits together
What we built
Knowledge sources per chatbot
Admins upload documents, add URLs, import CSV files or connect a database, see each source's processing status and attach every source to one or more chatbots.
Inbox and dashboard
Agents and admins see every conversation grouped by chatbot, filtered by date or searched by name or email, and a dashboard tracks conversations, users and escalations over time.
Organizations, teams and access
Super admins manage organizations and users, organization admins manage their teams and profiles, and sign up runs on email verification with one time codes.
Chatbot setup
Organization admins create chatbots in six steps: customization, pre chat form, agent availability, knowledge base, quick prompts and website integration, which ends with an embed script and a test chat.
Agent availability
Per chatbot business hours by weekday and time zone, the agents assigned to it and an editable message shoppers see when no agent is on shift.
How it works
Set up
An organization admin creates a chatbot, attaches its knowledge, assigns agents and sets their hours.
Embed
The admin tests the chatbot in a live chat, then adds its embed script to the organization's website.
Chat
A shopper fills in the pre chat form and asks by text or voice.
Hand over
When the assistant cannot answer, the chat goes to an assigned sales agent in the shared inbox, or the shopper sees the off hours message.
Follow up
Agents and admins work every conversation from the inbox and dashboard.
What changed
We delivered every item in the MVP scope. The final milestone covered the organization admin's notifications, sign in, dashboard and chat inbox, the agent's sign in, dashboard, inbox and account, and the shopper's pre chat form, voice input and chat with an agent. It was deployed over SSL, and the client reviewed it with us in a feedback session.
- Every item delivered, for every role
- MVP scope
- Super admin, organization admin, agent, end user
- Roles delivered
Under the hood
- Datics role
- Product built by Datics
- Access
- Middlewares and access control for super admins, organization admins, agents and end users, set up in the first sprint before any chatbot feature.
- Escalation
- An escalation flow in the FastAPI and LangGraph conversation service hands chats the attached knowledge cannot answer to the agent inbox.
- SMTP sends verification emails and one time codes from our mailbox, with Azure SMTP recommended for the client's own sender.
- Deployment
- Frontend and backend over SSL on a temporary subdomain, ready to move to the client's own subdomains, with AWS S3 for media and a separate dev environment.
AI and models
Cloud and infrastructure
Apps and APIs
Also in the stack
How it shipped
Scoping and design
A proposal and feature breakdown, wireframes for all four roles walked through with the client, Figma designs and user stories, then the client confirmed the build.
Foundation sprint
Schema, middlewares and access control, a reusable component library, the email service, sign in for super admins and organization admins, and user and organization management.
Chatbots, knowledge and inbox
Chatbot management with pre chat forms and quick prompts, agent availability, website integration and knowledge base management, then the inbox, dashboard, teams and notifications on a dev server.
Agent and shopper features
Agent features and the shopper's chat with quick prompts, product comparisons, voice input and chat with an agent, then AI engineering on the escalation flow and the embed widget.
Regression and delivery
A regression sprint, delivery of the full MVP scope over SSL with a demo video, and a feedback session with the client.
Sprints with weekly reviews
At kickoff we set out deliverables for each two week sprint, and we reviewed progress with the client in weekly meetings.
Sample data and a recorded demo
We built sample data to test the chatbots, kept regression sheets for every release, and sent the client a recorded product demo with the final milestone.
Related work
Have a product with the same kind of constraints?
Tell us what already exists and what you want it to do. We will tell you what it takes.