5 Savvy product samples Unofficial concept, synthetic data
Unofficial concept, not affiliated with Savvy. All data synthetic. Built only from public information. Happy to take it down on request.
Product samples for the Senior Product Manager role, November 2026

Five product samples for one operating RIA

Savvy is three businesses now: an adviser platform, a custodial platform on Fidelity's rails, and a back-office AI line sold to other firms. I built one product in that shape, five surfaces, one client record, and one approval and evidence trail. Each surface maps to one lever the business actually pulls.

Synthetic data only Built in two days with AI coding agents Live model calls, recorded No custodian connection Not affiliated with Savvy
Recruit Live

Transition Copilot

Turns a signed adviser's book into a structured record, a signature and NIGO plan, and an approval file.

AI pattern: Document extraction and workflow orchestration

109 to 152 hours returned per 100-client transition, modelled Open →
Serve Live

Client Concierge

Answers a client's own questions from their record, with citations, and routes anything that is advice to the adviser.

AI pattern: Retrieval over the household record and grounded generation

3.3 adviser hours returned per month, modelled Open →
Grow Live

Book IQ

Ranks the next dollar in every book with evidence, and drafts the play for the adviser.

AI pattern: Reasoning and ranking over the book, plus outreach drafting

USD 4.48M modelled net new assets across 12 households Open →
Comply Live

Supervision in the loop

Drafts adviser marketing inside a pre-approval workflow with an evidence file per piece.

AI pattern: Generation plus claim classification and policy checks

Review cycle 3.31 to 1.87 days per piece, modelled Open →
Scale Live

LP Ops

Takes private markets paperwork from intake to capital call without adding headcount per dollar.

AI pattern: Document extraction and exception detection

5.25 hours returned per LP per year, modelled Open →
Method Always current

How it is built

The model, the prompts, the recorded runs, the evaluations and what they got wrong. Every number on the method page is read from the modules, not written by hand.

Model: DeepSeek Flash by API. Recorded calls replay by default.

Prompts, recordings, evals, failures Open →

How to look at this in five minutes

  1. Start with Transition Copilot. It is the finished one: a signed book becomes a record, a signature plan, and an approval file.
  2. Then Client Concierge. Same client record, now answering a client question with citations and an advice boundary.
  3. Book IQ ranks the next dollar in every book. Supervision puts generated marketing through a pre-approval file. LP Ops takes private markets paperwork from intake to capital call.
  4. Read How it is built to see the model, the prompts, the evals, and what the models got wrong.

Why these five

They are the five levers in Savvy's own operating model: recruit advisers, grow their books, serve their clients, keep marketing compliant, and scale operations without adding headcount per dollar.

Every surface reads the same synthetic book and writes to the same audit trail. That is the product argument: one client record, many surfaces, one set of controls.

The suite was built in two days. AI coding agents wrote most of the code. I directed the product decisions, prompts, and controls, reviewed every module, and ran the QA. That workflow is the part I would bring to the role.

All metrics on this site are modelled estimates from stated assumptions. Nothing here is a measurement.

What is real and what is simulated

Real

  • The application code, the model calls, the prompts, the evals, and the approval and audit logic.
  • Every model output was produced by a live call to DeepSeek and is replayed from the recording by default.
  • The synthetic book is generated deterministically and is internally consistent.

Simulated

  • All clients, households, accounts, custodians, LPs, and adviser data are invented.
  • There is no custodian connection, no filing, no trading, and no delivery of any message.
  • The business metrics are modelled estimates, not measurements from a live firm.

The business case, in one line per lever

LeverSampleWhat it movesEstimate, modelled
Recruit Transition Copilot Turns a signed adviser's book into a structured record, a signature and NIGO plan, and an approval file. 109 to 152 hours returned per 100-client transition, modelled
Serve Client Concierge Answers a client's own questions from their record, with citations, and routes anything that is advice to the adviser. 3.3 adviser hours returned per month, modelled
Grow Book IQ Ranks the next dollar in every book with evidence, and drafts the play for the adviser. USD 4.48M modelled net new assets across 12 households
Comply Supervision in the loop Drafts adviser marketing inside a pre-approval workflow with an evidence file per piece. Review cycle 3.31 to 1.87 days per piece, modelled
Scale LP Ops Takes private markets paperwork from intake to capital call without adding headcount per dollar. 5.25 hours returned per LP per year, modelled