Ambient AI for financial stability

Balansa predicts money problems before they happen.

An intelligence layer over your whole financial life — every account, continuously watched — so a warning arrives while there is still room to act.

Financial stress over time

Cost zoneRetention spend · literacy programmes · social services loadCrisis thresholdHighLowTimeEarly signalWithoutWith Balansa
Above the threshold, the cost stops being the household’s alone — banks absorb it as retention and financial-literacy spending, and the socio-economic sector takes the rest.
A signal fired early enough does not just hold the line — it gives a household room to reverse the trend.
Lean Canvas

The whole business, on one page

Nine blocks. Each one opens into the detail, and each one separates what we have actually validated from what comes next.

  • 65% of Europeans cite money as a top stressor, and 40% cannot cover a €1,000 emergency.
  • Financial lives are split across banks, cards, subscriptions and platforms — no single view exists.
  • Budgeting apps report what already happened and require manual effort to stay useful.
  • Chatbots can answer financial questions but cannot see your accounts, so they cannot warn you.
  • Financial stress compounds into health and social crises long before anyone intervenes.

What we've validated

  • Every bank and operator conversation so far has recognised the problem without being sold on it.
  • Published research consistently links lower socio-economic position to worse mental-health outcomes.
  • Dutch and Finnish agencies are already funding budgeting tools to address it.

What's next

  • Quantify how early a signal has to fire to actually change an outcome.
  • Test the framing with caseworkers, not only with banks.
  • Retail banks and banking groups that already fund financial-wellness programmes.
  • Municipal social services and public-health agencies in the Netherlands and Finland.
  • Financially vulnerable households under social stress — the people the signals are for.
  • Employers and benefit providers as a later, higher-volume channel.

What we've validated

  • Banks are actively spending on this problem and are willing to take meetings about it.
  • Dutch and Finnish agencies have a structural gap: no real-time early-warning signal exists for them.

What's next

  • Confirm who inside an agency actually holds the budget for a tool like this.
  • Establish whether employers are a distinct segment or a channel into households.
  • Not a chatbot, not a dashboard: ideally a checkbox inside the banking app you already use.
  • Cross-bank by default, because a single institution only ever sees part of the picture.
  • Continuous monitoring rather than a monthly report you have to remember to open.
  • Early-warning signals framed as help, not as another chart to interpret.

What we've validated

  • The full pipeline already produces a structured financial health read-out from real account data.

What's next

  • Prove that a signal fired early enough measurably changes what a household does.
  • Guardian — real-time protection: unexpected charges, duplicate payments, cashflow risk.
  • Coach — scheduled briefings: what is coming, what changed, what it means.
  • Advisor — conversational answers to "can I afford this?" grounded in actual accounts.
  • Builder — automation hooks for people who want to wire their own triggers.
  • Underneath all four: PSD2 ingestion, cross-bank normalisation, and our MCP reasoning layer.

What we've validated

  • PSD2 to normalisation to automation platform to MCP workflow runs end to end on live accounts today.
  • The public stress check demonstrates the read-out that the modes are built on.

What's next

  • Move the monitoring engine from scheduled runs to genuinely continuous.
  • Decide which mode leads for an agency audience — Guardian and Coach are the likely pair.
  • Direct partnership conversations with banks and banking groups.
  • Public-sector and agency procurement in target markets.
  • Accelerator and programme routes such as Mastercard Lighthouse.
  • A public stress check that lets anyone experience the output before any commitment.
  • Writing openly about what we are building, to compound inbound interest.

What we've validated

  • Warm introductions have consistently converted into real meetings.
  • The public analyzer works as a low-friction way to show rather than tell.

What's next

  • Learn the actual procurement path for a Dutch or Finnish municipality.
  • Establish whether a maker and open-source channel produces qualified demand.
  • Per-seat or per-customer licensing to banks embedding the signals.
  • Agency licensing priced against crisis-intervention cost avoided.
  • Consumer tiers indicatively €8–€12/month standard and €15–€25/month for real-time protection.
  • API access for partners building on the intelligence layer.

What's next

  • No revenue yet — first paid pilot is the milestone that matters.
  • Test whether agencies buy per-caseworker or per-household-monitored.
  • Confirm consumer willingness to pay at the indicated tiers.
  • Open-banking data access and aggregation fees per connected account.
  • AI inference for reasoning over financial context.
  • Infrastructure and orchestration for continuous monitoring.
  • Compliance, security review and legal work required before any regulated pilot.
  • A small engineering team — currently very small.

What we've validated

  • Running the full pipeline on real accounts has kept infrastructure cost negligible so far.

What's next

  • Model per-account cost at pilot scale rather than at single-user scale.
  • Scope what a bank-grade compliance review actually costs.
  • Lead time: how far ahead of an event a signal fires.
  • Signal precision: how often a warning was worth sending.
  • Action rate: how often a person or caseworker responds to a signal.
  • Coverage: how much of a household’s financial life is actually visible.
  • Retention past 30 days — the point where budgeting apps lose more than 70% of users.

What we've validated

  • The pipeline produces the underlying data these metrics require.

What's next

  • No accuracy or lead-time figures measured yet; instrumenting this is the immediate priority.
  • Define what counts as an actionable outcome with an agency partner rather than alone.
  • A proprietary MCP workflow that lets an AI model reason over live, normalised, cross-bank data.
  • Cross-bank position: banks structurally cannot see beyond their own customers.
  • Built on regulated open-banking rails from the start, not scraped or screen-read.
  • A dual framing — industry good and social good — that neither pure fintech nor pure govtech occupies.
  • Deep systems and fintech experience behind the build.

What we've validated

  • The MCP workflow exists and runs against real accounts, not as a prototype sketch.

What's next

  • Turn the working pipeline into defensible models, which requires cohort data.
  • Establish whether the social framing opens doors that the fintech framing cannot.
Trajectory

Niche today, infrastructure eventually

The same intelligence layer serves a wider buyer and a heavier responsibility at each step. We are at the start of all three lanes.

Reach

  1. We are hereNiche
  2. Stage 2Mainstream
  3. Stage 3Systemic

Buyer

  1. We are hereBanks
  2. Stage 2Employers
  3. Stage 3Agencies
  4. Stage 4Governments

Capability

  1. We are hereStress scoring
  2. Stage 2Stress trajectories
  3. Stage 3National wellbeing infrastructure
Why Balansa matters

One intelligence layer, several audiences

The same signals serve a bank protecting its customers and an agency trying to reach a household before crisis. We are starting with two and designing for more.

Industry good

Financial institutions

Banks, banking groups and fintech partners

Banks already fund financial wellness but only see their own slice. We give them the cross-bank picture.

Read the detail →
Social good

Social services

Municipalities, public-health bodies and NGOs

Financial instability drives mental-health risk. Agencies have no real-time early-warning signal. That is the gap.

Read the detail →
Exploring

Makers & open protocols

Developers, automation builders and the MCP ecosystem

If financial context is available through an open protocol, people will build things we would never think of.

Exploring

Community funding

Mutual aid and community finance initiatives

Verified financial stress signals could route community support to the households that need it most.

Where we are today

What actually runs

Closed alpha — live PSD2 data through an automation platform and a proprietary MCP workflow, currently running on the founder’s own bank accounts.

Live

Bank APIs (PSD2)

Read-only account and transaction access over regulated open-banking rails.

Live

Cross-bank normalisation

Different banks, different formats, one consistent transaction model.

Live

Automation platform

Scheduled and event-driven workflows move data through the pipeline. Currently n8n, deliberately swappable.

Live

Proprietary MCP workflow

Our own Model Context Protocol layer lets an AI model reason over live financial context.

Building

Continuous monitoring engine

Turning a point-in-time read into an always-on watch.

Building

Early-warning signals

Rules and models that fire before a problem lands, not after.

Next

Consented external cohort

Gated on documented consent and data handling.

Next

Bank-ready compliance layer

What a regulated partner needs before a real pilot can start.

Next: a consented friends-and-family cohort, once the consent and data-handling process is documented.

Other people’s bank data is not something to extend access to casually. How we keep access minimal is set out in how we earn trust, and what we handle today is on the data and privacy page.

Evidence & validation

Labelled by how far it is actually proven

A meeting is not a pilot and an intention is not a metric. Separating them is the only way the validated claims stay worth anything.

Validated
Working today, verifiable now.
In progress
Live conversation or active build.
Planned
Committed next step, not yet started.

Validated

  • Full pipeline running on live bank data

    Validated

    PSD2 access, cross-bank normalisation, automation-platform orchestration and our MCP workflow run end to end against real accounts — not mock data, not a demo fixture.

  • Public financial stress check

    Validated

    Anyone can upload a bank statement and receive a structured health read-out. It has been live and usable since before this application.

  • Grounded in published socio-economic research

    Validated

    The stress model builds on established findings linking socio-economic position to mental health outcomes, including Dutch and Finnish public-health research.

  • Agencies are already building for this

    Validated

    Dutch and especially Finnish social bodies are actively developing budgeting tools to improve socio-economic position. The need is established and funded — what those tools lack is a real-time signal.

In progress

  • Baltic tier-1 bank

    In progress

    Positive first meeting, follow-up scheduled.

  • Second Baltic bank

    In progress

    Warm introduction made, conversation opening.

  • Nordic-Baltic banking group

    In progress

    Interest expressed, internal introductions underway.

  • Mastercard Lighthouse

    In progress

    Encouraged to apply; application in progress.

  • Nordic early-stage fund

    In progress

    Strong resonance with the thesis in early conversations.

  • Fintech operator

    In progress

    Independently validated the pain and the approach.

Planned

  • Consented friends-and-family cohort

    Planned

    The next step beyond self-testing. Blocked until the consent and data-handling process is documented — deliberately, not accidentally.

  • Early-warning precision measurement

    Planned

    Measuring signal precision against labelled outcomes. No accuracy figure exists yet, so we are not quoting one.

  • Agency pilots in the Netherlands and Finland

    Planned

    Target design: a caseworker-facing early-warning view validated against real intervention outcomes.

Pilot inquiry

Start a conversation

Whether you run a bank, a social service or something we have not thought of yet — tell us what you are trying to solve.