Coherent: Insurance tooling that let actuaries ship product changes in days, not quarters.

Coherent Spark converts the business logic inside Excel models into cloud API services. I designed the workflows that made that trustworthy for non-engineers — an in-Excel Modelling Center for mapping and testing models, and the publishing surfaces on the Spark web platform that added versioning, validation and approval without breaking the spreadsheet mental model.

Context

The spreadsheet was the product.

Respecting an expert-user model

Actuaries lived in Excel. Every insurance product change ran through a spreadsheet, then a manual engineering ticket. The design problem wasn't replacing that mental model — it was adding version control, review and audit without breaking the trust that made spreadsheets feel safe.

Actuaries author, engineers deploy — long cycle time
PR-style review felt alien to actuarial workflow
Cell-level trust is non-negotiable in regulated products
Every change needs a defensible audit trail

Excel spreadsheet with the Coherent Modelling Center add-in docked in a right-hand panel — The Modelling Center — docked inside the actuary's actual workbook

Diagram showing spreadsheet logic from actuaries, analysts and operations flowing through Spark into enterprise platforms — Spreadsheet logic in, cloud services out — the ecosystem Spark serves

01 · Research

What we needed to learn.

Contextual inquiry with three actuarial teams across APAC. Co-design workshops to test which review and approval rituals were welcomed versus resisted. Prototype benchmarking to quantify error rate against free-form editing.

3 embedded contextual inquiries
Co-design workshops on review patterns
Prototype benchmark vs free-form editing
Pilot analytics — cycle time & escalations

Hand-drawn wireframe sketches showing three screens: Modelling Center home, setup file screen, and mapping tool with cell address and input type fields — Early ideation — sketching the Setup file → Mapping tool flow with pilot actuaries

Story mapping board with user activities, tasks and MVP 1 user stories on colored sticky notes across offline and online flows — Story mapping — aligning product, engineering and actuarial pilots on scope for MVP 1

02 · Decisions

Finding → intervention.

Cell-level trust is non-negotiable

Every value carries provenance: who changed it, when, with what test result.

Review rituals must look familiar

Spreadsheet-style diff view with inline approve/comment — not PR-style.

Testing had to live inside Excel

The Modelling Center brings mapping, test-case generation and testbed runs into the workbook — no export, no context switch.

Cycle time dropped from quarter → days

Pilot customers ship product changes in a release window measured in days.

03 · Design

Three surfaces that carried the workflow.

Modelling Center add-in panel with Uploaded status, testing steps and a disabled Parameters option under Coming soon

Surface A

Testing without leaving Excel

The Modelling Center reads top-to-bottom in execution order — Mapping file → Generate test cases → Run Testbed — with the current step highlighted and upload status as a live signal. The layout is the workflow. Features not yet shipped ("Parameters") appear visibly disabled under "Coming soon" rather than hidden — honest roadmap signaling that compliance-minded users valued.

Surface B

Guided publishing with live validation

Publishing a model as a service is guarded by structured inputs — unique service names, semantic version numbers, character-limited labels with live counts, effective date ranges — each behaviour written as a GIVEN/THEN spec for engineering. Guardrails at authoring time, not error messages after.

New service modal on the Spark platform showing version label field at its 25-character limit with annotated validation spec

Spark web platform table listing model files with version, status, type, creator, location, approver and date approved columns

Surface C

Governance at a glance

The Current Overview gives administrators one auditable table across 248 files — version, status, creator, approver and approval date per model. Provenance isn't a detail view; it's the default view.

Impact

What moved, and by how much.

04 outcomes

≈95%

Faster product-change cycle (quarters → days)

−35%

Authoring error rate vs free-form editing

−40%

Support escalations during pilot

Actuarial teams onboarded across APAC

“The authoring surfaces made version control feel like a spreadsheet feature, not an engineering one — that's why the actuaries trusted it.”

Product lead — Coherent Spark pilot

Key takeaways

What I'd carry forward.

Familiarity is a feature

In expert tooling, borrowing the incumbent metaphor beats reinventing it.

Trust is cell-level, not screen-level

One un-audited change corrupts the whole tool's credibility.

Guardrails need selling

Actuaries accept structure only after the guardrail proves it helped.

Next case study

Healthcare · Patient experience