WealthTech / Enterprise Document AI2025An "Ask your vault" AI layer for wealth-management advisors
FutureVault sits at the centre of how RIAs organise client documents — estate, tax, insurance, identity. We built a retrieval-grounded AI layer on top, letting an advisor ask plain-English questions across an entire household's document history and get cited answers in seconds instead of digging through folders.
- Document lookup
- 3.2× faster
- Citation accuracy
- ~96%
- Households indexed
- 50K+
- Pilot → GA
- 11 weeks
The brief
The advisors who use FutureVault are not document-search experts. They are wealth professionals who need to answer a client question — "what did we decide about the cottage in 2019?" — in the middle of a meeting, without leaving the conversation to scroll through a folder tree.
The bet: a retrieval-grounded AI layer on top of the existing vault would convert a tedious lookup task into a one-line query and let advisors operate at a different pace.
What we built
A purpose-built RAG pipeline tuned for legal-financial documents — wills, trusts, beneficiary designations, policy schedules. The retrieval layer combines dense embeddings with a domain-tuned keyword scorer, then runs a cross-encoder reranker over the top 30. Each answer is cited back to the specific page in the specific document, and the citation is clickable: tap and the advisor is reading the original artifact, with the cited section highlighted.
We did the unglamorous work too: a PDF parsing pipeline that survives form scans, an OCR fallback for image-only documents, and a household graph so the system understands that the will mentioning "the cottage" and the property survey for 1422 Lakeshore Road are the same asset.
Every advisor query is logged, scored against a curated eval set, and the judge model is itself evaluated quarterly against advisor preferences to keep drift in check.
Outcome
Document-retrieval time dropped 3.2× in the median case, with the biggest wins on cross-document questions where the legacy flow required opening multiple files. Citation accuracy held at ~96% on the evaluation set — high enough for advisors to stop double-checking by default.
The pilot went from first design partner to GA in 11 weeks, with 50,000+ household vaults indexed at launch. FutureVault now positions the AI layer as a category-defining feature in advisor demos.
"We were one of the first WealthTechs to put a real retrieval-grounded AI in front of advisors. The difference is people use it every day."
Keep exploring.
Book a call
Insurance / Financial ServicesA modern policyholder portal for a 75-year-old insurance co-op
Co-operators is one of Canada's largest insurance co-operatives — 600+ locations, generations of customer trust, and a legacy self-service experience built around PDFs and phone trees. We rebuilt the policyholder portal around real-time claim status, digital ID cards, and a single-tap "file a new claim" flow that absorbs roughly a third of the inbound contact-centre volume.
- Inbound call reduction
- ~28%
- Time-to-claim filed
- 4 min
EdTech / Non-profit LearningAdaptive mastery-streak nudges for learner retention
Khan Academy's mastery streak is one of the most loved features in online learning — and one of the easiest to break the moment a learner has a hard week. We built an adaptive nudge layer that personalizes lesson length to a learner's energy curve, lifting 14-day retention 19% for opted-in cohorts.
- 14-day retention
- +19%
- Learners enrolled
- ~1.2M