Migration on the Mulberri platform
Migration is a product, not a project
Connect your legacy policy admin systems and rating sources through prebuilt connectors on day one. Berri AI finds what you run, builds what’s missing, and proves the result against your own book before anything goes live.
Your legacy stack9 systems, 3 rating sources, 14 years of records
The market
Two ways to buy this problem. Neither one fixes it
One camp sells you a program. The other rents you a layer. Your book stays where it was.
The project
The traditional suites
A replacement measured in fiscal years. Architecture reviews of your existing code, mapping decks written by hand, an integrator billed by the hour, and a cutover weekend at the end. The platform you bought waits behind the project you funded. Afterward, every rate change goes back through the vendor.
The layer
The new AI players
Intelligence rented on top of the systems you already have. The demos are impressive. But the record underneath stays fragmented, the layer is one more system to integrate, and the moment you want to change what sits below it, you are back in the project.
The fabric
The suite carries its own migration
The record, the logic, and the agents live in one place, and moving in is part of the product. Connectors keep your current systems live while Berri AI carries what you choose to move. No single cutover to bet the company on, no layer to reconcile, and you change the logic yourself in plain English.
Discovery
Know your scope before you commit
Point read-only connectors at what you run today: the policy admin system, the rating spreadsheets, the agency management system, the document stores. Berri AI takes inventory. Which systems hold the book. Where the rating logic actually lives. Which records conflict. You get a readiness report with a proposed scope, acceptance criteria, and a decision date, before you have signed anything.
- Read-only by design. Discovery never writes to your systems.
- Every finding cited to the system and file it came from.
- Conflicts surfaced up front: two rating sources, disagreeing class tables, orphaned records.
- The readiness report becomes the migration plan. Nothing gets re-scoped later.
Berri AI · Discovery
Inventory of the systems you run today.
- Scanned 6 connected systems
- Found 3 rating sources, 2 in conflict
- Mapped 14 years of policy records
- Readiness report ready for review
Read-only throughout. Nothing was written to your systems.
How it works
Find, build, prove
Your team checks each step, and nothing goes live without sign-off.
Step 1
Berri AI reads what you have
Rating sheets, appetite guidelines, authority grids, forms, policy registers, bordereaux, loss runs. Any file, any format, plus everything discovery already inventoried. A source citation on every value.
Reading the inputs3 files, every value cited to its page
Rating_Model_2026.xlsxSheet 3 · 18 factors extracted
UW_Guidelines_v4.pdfPages 4 to 11 · 9 rules mapped
Policy_Register.csv14 years · records structured
Step 2
The mapping shows itself before anything moves
You get a mapping report, not a promise. Every source field, every target, every assumption Berri AI made, laid out for your team to confirm or correct. Nothing lands on the fabric until you approve the map.
Mapping report26 source fields · 25 confirmed, 1 for your call
Insured name to record.insured
Class code to record.class
Payroll basis to rating.payroll
Step 3
Berri AI builds what's missing
Discovery always finds gaps. A claims system with no connector. Rating logic that lives in one person's spreadsheet. A form format nobody can parse. Berri AI builds the fix itself: a connector from a sample export, the rating plan rebuilt in UWBerri as formula steps and factor tables you can open and test, an extraction model trained on the form. Configured, not coded, so every fix stays yours to change.
Gaps and fixesFound in discovery, built by Berri AI
Connector built from a sample exportGap: No connector for the claims system
Rating plan built in UWBerri, test run passedGap: Rating logic in a spreadsheet
Extraction model trained on the formGap: Unreadable form variant
Step 4
Proven against your own book
Berri AI reruns your historical cases through the rebuilt logic and compares the output to what your legacy system produced. Acceptance criteria are agreed at kickoff. The line goes live when the replay meets them, and both systems run in parallel until then.
Replay against the bookCommercial Property · 150 historical cases
127
match legacy
2
within tolerance
0
outside
Three ways in
Stay, move, or launch. The fabric doesn't force the choice
- 01
Stay on your stack
Coexist through connectors. Your policy admin system stays the record while the fabric connects to it and Berri AI works at the edges: intake, screening, preparation. Structured submissions hand off to your external system by API. Books can move later, one line at a time.
- Your PAS stays the record
- Berri AI works at the edges
- Move later, one line at a time
- 02
Move your book
Berri AI carries the book through the fabric to the acceptance criteria agreed at kickoff. Full history comes along: policies and versions, endorsements, claims, certificates, financials. Migration is included, with no separate fee.
- Acceptance criteria agreed at kickoff
- Full history comes along
- No separate fee
- 03
Launch a new line
Describe the line in plain English and attach a rater. Berri AI builds the rating plan, the appetite rules, and the forms, live in the sandbox the same day. No release cycle to wait on.
- Described in plain English
- Rating plan, rules and forms built
- Live in the sandbox the same day
What moves
Years of work, carried over whole
Rating logic
Formula steps, factor tables, and variables into UWBerri rating configuration.
Appetite and triage
Guidelines become rules that read operations, not just class codes.
Authority
Binding authorities and referral thresholds, enforced live.
Forms and templates
Quote, policy, and endorsement documents attached to the record.
Policy records
Versions, endorsements, renewals, and cancellations with history.
Claims
Open and closed claims with reserves and payment history.
Certificates
Issued certificates linked to the policy versions behind them.
Financials
Invoices, commissions, taxes, and bordereaux by binder period.
Governance
Berri AI does the carrying. Your people make the calls
Every extracted value carries its source. Every mapping is confirmed before it lands. Every action is logged on the record. Every decision is confirmed by an underwriter.
On the recordEvery value cited, every mapping confirmed, every action logged
Territory factor 1.12cited to Rating_Model sheet 3, p.2
Appetite rule 14 mappedconfirmed by B. Parsons
Payroll basis selectionlogged 14:02
Cutover approvalunderwriter sign-off
Start connected
There is no integration phase. That's the plan
Bring your rater and your stack as they are. The fabric connects on day one, and Berri AI builds whatever the library doesn't cover. Built for MGAs, carriers, brokers, and reinsurers.
See the integrations