Designing AI-assisted suitability reports
Suitability reports are a mandatory part of financial advice, explaining why a recommendation is appropriate for a client's circumstances and objectives. At Ekorn, producing one meant leaving our platform for a third-party AI tool, editing the output in Word, then returning to continue onboarding.
I led product design across the project — from early workflow definition and user research through prototyping and validation — and worked with product and engineering to define how source documents, AI extraction and the report template connect.
Target metrics
Measured post-launch · Sep 2026reduction in average time on the report step
of started migrations reaching report sent
native Ekorn report adoption within 6 months
The problem
Producing one report takes eight steps and four different platforms.
Fragmented workflow
Every handoff interrupted onboarding.
No early validation
Users couldn't review the AI's assumptions before generation.
Limited control
Once generated, correcting mistakes meant leaving the workflow or starting again.
Research
Advisers want automation without losing control.
AI should extract, not make users repeat themselves
Existing client documents and data should do the heavy lifting.
Errors should be caught before generation
Users want to inspect conflicting values and correct important information before trusting the report.
The adviser remains accountable
AI can draft; humans need to review, edit and explicitly approve.
“Every time Saturn produces a charges table, he does it wrong. So you almost have to copy and paste the charges table in. Every time…”
Financial adviser · user interview
The new workflow
Every step now happens inside Ekorn.
Source control
Users choose which documents AI should consider rather than blindly feeding everything into the model.
Review before generation
Extracted information and inconsistencies are surfaced before drafting, so errors can be corrected early.
Human-editable output
AI produces the starting point, not the final decision.
Explicit approval
The adviser remains the final accountable human before completion.
Document upload
Asking an AI to “read everything” creates risk
A suitability report draws from documents created at different points in the client journey. Information overlaps, changes over time and conflicts — so the first design decision was which documents get read at all.
For MVP we're working from a single controlled report template. I mapped that template into variables and generated prompts, defining the information required and the likely source for each section.
<client-name> · <risk-level>
<current-plan-value> · <prompt-4>Excerpts from the mapped template. Yellow marks a variable extracted from a named source document; green marks a passage the model writes from the client's data. Working through the template this way also exposed which fields have no reliable source — the ones that later became the conflict-resolution step.
Key decision01
Give the AI a source hierarchy, not just more context
Rather than treating every uploaded document as equally reliable, we designed the extraction logic around recency and source relevance. The most recent information wins where circumstances have changed, while specialist documents stay authoritative for product data such as plan values, fees and charges.
Key decision02
Surface uncertainty instead of hiding it
A hierarchy reduces conflicts but can't eliminate them. When the system finds contradictory information, we don't want the model to silently pick an answer.
Show the conflict and its sources
What doesn't match, and where did each value come from?
Suggest, don't decide
The system can recommend the likely answer; the user stays in control.
Correct at the source
Once resolved, the verified information becomes the basis for generation.
End to end
Four steps, one platform, an adviser signature at the end.
Beyond MVP
Firms bringing their own templates
Today generation starts from one controlled template. The direction we're exploring lets a firm upload their own: the system identifies required variables, maps them to trusted client data, and produces a report that preserves the firm's structure and language.
Also at Ekorn
The migration journey: a seven-stage pipeline for moving whole client books onto the platform, with stage counts, approval queues and the data a client is missing surfaced before it blocks the transfer.
Illustrations and account opening: choosing a wrapper, pre-filling from existing plans, and building the investment strategy, contributions, withdrawals and charges an illustration needs.
The client dashboard: the client-facing side of the platform — portfolio value and growth, per-plan detail and documents — specified screen by screen against the API fields behind it.
