01Overview
The agency’s advisers spend their best hours with families. The workspace sits on top of the tools they already use and files every lead, email, call and document to the right family, so the file and a brief are ready before each meeting. Answers cite the policy terms they came from, and the follow-up is drafted after the call.
The family workspace
One file per household: its policies, conversations, tasks, proposals, claims and a brief.
The knowledge and AI layer
Documents read and searched in Hebrew, every answer cited, every model call recorded.
The finance desk
Cash flow, commission statements and their gaps, policy milestones and new-offer matching.
02The challenge
Preparing took longer than the meeting.
Client data sat in the agency’s client system, with portfolios and cash flow in spreadsheets, mail in one place and the calendar in another. Preparing a meeting meant collecting all of it by hand. Answers were dug out of policy documents, each call was written up several times, and commission statements were checked line by line.
What the client wanted to achieve
Prepared before every meeting
The family’s file and brief, ready without collecting anything by hand.
Answers with their source
Every answer cites its passage, or says it has none.
Write each call once
One summary becomes four follow-ups.
The owner stays in control
AI proposes and a person approves, with one switch to pause it.
How it worked before
- A family’s history across systems, spreadsheets and the inbox
- Answers dug out of policy documents by hand
- Each call written up several times
03How we built it
- 01
Foundation and permissions
A Hebrew workspace, staff roles enforced on the server, every source kept as its original, and an approvals queue.
- 02
Knowledge and customer context
A read-only copy of the client system, Hebrew document reading, and search that cites its answers.
- 03
Intake and operations
Leads from forms, mailbox rules, call transcription, cases and tasks, the calendar and the family file.
- 04
Professional outputs
Exact calculations, comparisons, proposals, a claims workspace and Hebrew PDF reports.
- 05
Finance and the policy lifecycle
Cash flow imports, commission statements and their gaps, policy milestones and new-offer matching.
- 06
A controlled assistant
An assistant across the whole workspace, with a screen that pauses all AI work at once.
01Foundation
2 days02Knowledge
3 days03Intake and operations
9 days04Professional outputs
11 days05Finance
4 days
04How it works
Step 01Arrive
Leads and mailKept exactly as they were received.Calls and documentsRecordings transcribed, documents read.Step 02File
Matched, then confirmedA shared phone number only suggests a family; a person confirms it.Step 03Prepare
The family filePolicies, calls, tasks and proposals in one place.Cited answersFrom the policy terms, with their source.Step 04Follow up
Four draftsWritten from one summary of the call.Approved, then sentSent once, and only after approval.
A summary for the family
Ready to copy, never sent on its own.
The meeting email
Insurance, pension, finance and next steps; a topic that was not discussed says so.
An entry ready to file
For the agency’s client system.
A handover
The decisions and the tasks that came out of the call.
Who stays in control
- Everything sent to a customer waits for approval, and goes once
- Only a person can start an email to a customer
- One switch holds all AI work while intake keeps running
- No customer document ever carries a commission or cash flow figure
- A phone number alone never identifies a person
05The outcome
Everything that arrives is filed to the right family. Before a meeting the file and the brief are ready, questions are answered with their source, and after the call one summary becomes four follow-ups that go out only once someone approves them.
| Before | After |
|---|---|
| A family’s history across systems, spreadsheets and the inbox | One family file, with a brief before each meeting |
| Answers dug out of policy documents by hand | Questions answered with the passage they came from |
| Each call written up several times | One summary and four drafted follow-ups |
| Commission statements checked by hand | Gaps and missing payments flagged |
| Renewals and reviews remembered by people | Reminders per policy and per family |
06Key figures
07Technology
What the system runs on, for the technical reader.
- Cloud
- AWSContainersManaged PostgreSQLFile storage and a CDN
- AI
- Claude and OpenAI, behind one gatewayRAG with hybrid searchLocal multilingual embeddings
- Documents and speech
- OCRHebrew speech-to-textHebrew PDF reports
- Integrations
- The client system, read-onlyThe mailboxOnline formsThe calendarCall and video recordingsInsurers’ commission statements
- How we build
- TypeScriptReactPostgreSQLA background job queueA walkthrough at the end of every stage
08What comes next
Next, the assistant across the whole workspace, security and backups, then a first trial on the agency’s own records.
09About
About the client
An insurance and financial-planning agency that looks after families’ insurance, pensions and savings. We name a client only with their permission.
About MYBB
MYBB builds systems around how a business already works. We diagnose it with MYBB ROIQ, build the system, launch it with the team and keep running and improving it.