WhatIf Money

Turn a plain-language money question into confirmed assumptions and reproducible low/mid/high projections—never advice.

Comment

Business Viability

People ask connected “what if” questions that single calculators isolate: what happens if I overpay my mortgage, how does buying compare with renting under my assumptions, or what does a four-day week do to household cash flow? Spreadsheet work is tedious, while general assistants may calculate inconsistently, hide assumptions, or drift into recommendation. WhatIf Money is a versioned projection workspace: the user describes a scenario, confirms every mapped input, compares two or three cases, and sees deterministic low, mid, and high outcomes with methodology and version visible.

The initial buyer is a UK consumer exploring a concrete life or money decision. The current £9 monthly plan unlocks additional saved comparisons after one free account-bound save. Because usage is episodic, the commercial model will also test annual or one-off access rather than forcing retention. Acquisition begins with educational search pages for mortgage overpayment, buying versus renting, and reduced hours; assumption-sensitivity explainers; carefully moderated personal-finance communities; and employer wellbeing referrals that never expose employee data. Activation is a confirmed comparison with assumptions opened. Retention is saved scenarios revisited or new scenarios compared—not time spent on the site.

Production includes self-serve accounts, versioned saved comparisons, deterministic engines, a live Stripe Checkout path, and an operator view for AI, billing, and compliance evidence. There is currently no verified independent product revenue or paying customer in the evidence pack. The submission therefore focuses on a functioning business system and transparent validation plan. The next proof is formula-reviewed user testing, completion and correction rates, paid conversion, support burden, and measured cost per scenario.

The official cash-basis P&L records $0 revenue and $0 product-specific cash expenses. Founder time and shared Yensi hosting, subscriptions, and product-factory resources predated this entry; they are disclosed and are not represented as WhatIf Money-specific cash payments.

We started WhatIf Money on 3 August 2026, only two weeks before the submission deadline. The late start shaped a narrow goal: ship a reproducible, reviewable production decision tool and disclose its limits. Revenue is £0 and paying customers are 0; the entry does not present the short development period as a mature business history.

AI-Native Operations

Gemini reduces the hardest input friction: mapping a natural-language question into candidate scenario parameters. For example, a user can describe mortgage balance, payment, rate, proposed overpayment, and time horizon in one sentence. Gemini 3.5 Flash returns bounded candidate fields with provider, model, confidence, prompt hash, scenario type, and status evidence. The user must confirm every value before calculation.

The numerical boundary is absolute. Gemini never returns the projection series, selects a “best” scenario, recommends a product, or changes an entitlement. Versioned deterministic Python and TypeScript engines own formulas, range generation, dates, sensitivity, chart data, and reproducibility. Every result repeats “Projection, not advice” and carries scenario, engine, and assumption versions. If Gemini is unavailable, the interface falls back to clearly labelled manual input; fallback output must never be recorded as a successful Gemini decision.

The submission audit caught exactly that observability risk: the production adapter used an obsolete model alias and recorded fallback defaults under the provider name “Gemini.” The final revision migrates to the supported Google Gen AI SDK and Gemini 3.5 Flash, records successful and fallback provider/status separately, and marks operator health as degraded when the latest extraction fell back. Twenty-eight backend checks and the production frontend build pass. This correction is important evidence of AI-native operations: model calls are useful only when their provenance and failure state are trustworthy.

Category Impact

WhatIf Money aims to make complex financial trade-offs more understandable without pretending to replace regulated advice. The theory of change is that natural-language mapping lowers the barrier to getting started; explicit confirmation catches misunderstandings; side-by-side sensitivity ranges reveal dependence on assumptions; visible methodology makes outputs reproducible; and neutral language leaves the decision with the user.

Impact will be measured through scenario start, mapping correction rate, confirmation, comparison completion, assumption-panel views, sensitivity interaction, saved/reopened comparisons, comprehension feedback, and paid continuation. Formula accuracy is protected with golden-master fixtures and immutable engine versions. The product will not claim better financial outcomes, certainty, or personalised recommendations. If users interpret the experience as advice, that message or scenario is stopped and reworked.

Shared Yensi product-factory contracts and runtime rails existed before the challenge. The WhatIf Money workflow, UI, calculation engines, persistence, Gemini extraction decision, compliance boundary, and billing path were built during the competition period and are disclosed separately.

Built With

Share this project:

Updates