Planning family money before a loan is needed

Reference build. HMR designed this system for a SEBI-registered advisory firm or a bank's wealth team. The household below is fictional, and every figure on this page is calculated from the inputs and assumptions shown. None of it is a client result.

The problem: families find gaps too late

Most households do not plan badly. They plan in pieces. Salary lands in one bank, mutual funds sit in an app, EPF lives on a government portal and the school fee arrives as a surprise every April. Nobody sees the whole picture, so a shortfall shows up only when a bill is due. By then the easiest fix is a loan.

A personal financial advisor should close that gap. It looks at what the family has, what it spends and what is coming, then shows the shortfall while there is still time to fix it cheaply. The hard part is not the maths. It is getting complete data with consent, keeping every number correct and keeping a qualified person in charge of advice.

HMR insight: A plan is only useful if the family can check every number in it. So the AI explains the plan, and a rules engine does the arithmetic.

The household in this model

The Rao household in Pune is invented for this case study. Two earners, one child aged seven, a home loan and savings spread across six places.

Item	Amount
Take-home pay, both earners	₹2,20,000 a month
Household spending	₹95,000 a month
Home loan EMI	₹42,000 a month (₹38,00,000 outstanding at 8.5%)
Existing SIPs	₹20,000 a month
Left over each month	₹63,000, sitting in a savings account
Savings account and fixed deposits	₹4,10,000 and ₹6,00,000
Mutual funds	₹14,50,000
EPF, PPF and NPS	₹24.4 lakh in total

Three costs are coming. The child starts an undergraduate degree in 11 years, which costs ₹20 lakh at today's prices. The family car needs replacing in two years, at ₹10 lakh today. School fees of ₹1,80,000 fall due every April.

What the first plan showed

The advisor's first run found three things the family had not seen.

1. A monthly shortfall of ₹7,600. Funding the car, the degree and the fees from the monthly surplus needed more than the family had left over.
2. Cash doing nothing. An emergency fund of six months' spending and EMIs needs ₹8,22,000. The savings account and fixed deposits held ₹1,88,000 more than that.
3. A car loan on the way. Without a plan, the family would finance the car. A five-year loan for ₹11.03 lakh at an assumed 9.5% costs ₹23,155 a month and ₹2.87 lakh in interest.

Moving the spare ₹1,88,000 into the car fund cut the monthly set-aside from ₹43,000 to ₹34,700. That closed the gap, and the car can now be bought outright.

The plan, month by month

Goal	Target	Monthly set-aside	Assumption
Car in 2028	₹11.03 lakh	₹34,700	5% price rise, 6.5% on deposits
Degree in 2037	₹46.6 lakh	₹12,600	8% cost rise, 10% return, ₹6 lakh of funds set aside
School fees each April	₹1,80,000	₹15,000	Fees unchanged this year
Buffer		₹700	What remains of the ₹63,000

The emergency fund stays at ₹8,22,000, untouched. Their home loan runs its course in about 145 months, and the family's EPF, PPF and NPS grow to about ₹1.14 crore by age 58 at an assumed 8%, before any new contributions.

What if the assumptions are wrong

Every plan rests on assumptions, so the advisor shows how the answer moves when one changes.

If this happens	The degree set-aside becomes
Returns average 8%, not 10%	₹15,700 a month
The ₹6 lakh is spent on something else	₹19,900 a month

The family sees these before they agree to the plan, not after.

The HMR architecture

1. ConnectBank, fund and insurance data arrives through Account Aggregator, only with the user's consent. Cash, EPF and planned costs are typed in.
2. CalculateA rules engine works out every figure: goals, set-asides, loan costs and gaps. Same inputs, same answer, every time.
3. ExplainThe assistant explains the plan in plain language and runs what-if questions through the engine. It never does the maths itself.
4. ApproveAnything that names a specific product goes to a SEBI-registered adviser, who approves, edits or rejects it.

Reminders then arrive before each set-aside, fee or renewal is due, so no one learns about a gap on the day it matters. We split the system so that each part does one job and can be tested on its own.

Layer	What it does	Built with
Consent and data	Pulls bank, fund and insurance data with consent, and takes manual entries	Account Aggregator, as a Financial Information User
Planning engine	Works out goals, set-asides, loan costs and gaps	Python rules engine, versioned formulas
Assistant	Explains the plan and answers what-if questions	Language model behind an agent gateway
Adviser console	Queue for any advice that names a product	Web app with named approvers
Reminders and audit	Due-date alerts and a complete log	Scheduler, PostgreSQL, OpenTelemetry

Account Aggregators are regulated by the Reserve Bank of India. They pass data only with the customer's consent, and they do not store or process it. That gives the advisor complete, current data without asking families to upload statements.

Why the numbers can be trusted

The assistant never calculates. When a family asks what happens if they delay the car by a year, the assistant sends the question to the planning engine and explains the answer it gets back. Each result stores its inputs, the formula version and the assumptions used, so an adviser or auditor can recompute it.

Before launch, the engine is checked against a library of hand-worked cases. A result that does not match to the rupee fails the build. When a formula changes, old plans keep the version they were made with, and the family is told when a new version gives a different answer.

Governance and regulation

• Consent first. Each data link has a purpose, an expiry and a one-tap revoke. Data is used for this plan only.
• Reproducible numbers. Every figure stores its inputs, its formula version and its assumptions, so anyone can recompute it.
• Adviser in charge. Specific product advice waits for a named, registered adviser. The assistant cannot send it alone.
• Full audit trail. Every question, calculation, approval and reminder is logged with who, when and why.
• No training on client data. Household data is never used to train models. Model providers are bound by contract to the same rule.

Three rules shape the design. Investment advice for a fee needs registration under SEBI's Investment Advisers Regulations, so specific product advice always goes through a registered adviser. Since February 2025, SEBI has made any regulated firm that uses AI solely responsible for the privacy of investor data and for the output of those tools. And India's Digital Personal Data Protection Act requires a clear purpose and consent for every use of personal data.

Each rule has a matching control. In the adviser console, approval is a step the software enforces. An audit trail shows who approved what, and a consent ledger records every purpose, expiry and revoke.

Delivery plan

1. Assess (3 weeks). Our Readiness Audit maps the firm's data sources, adviser workflow, Account Aggregator route and security gaps, and ends in a ranked plan.
2. Build (12 weeks). Weeks 1 to 4: consent, data links and the household data model. Weeks 5 to 8: the planning engine and its test library. Weeks 9 to 12: the assistant, the adviser console, reminders and security testing.
3. Pilot (8 weeks). A small group of consenting households uses the advisor with their adviser. Every plan is checked by hand before it goes out.
4. Run and improve. Widen access only when the pilot measures below hold.

A typical team is one HMR architect, three engineers, one quality engineer and a part-time security lead, working with the firm's product owner and a registered adviser.

What the pilot measures

We do not promise outcomes before a pilot has run. We agree the measures up front:

• Share of plans where the engine matches the hand-worked answer (target: all of them)
• Share of product advice approved by a named adviser before it reaches a family (target: all of it)
• Shortfalls found at least three months before the money is due
• Time an adviser spends preparing a plan, before and after
• Families who revoke consent, and why

Assumptions used in this model

• Car prices rise 5% a year; car fund earns 6.5% a year; car loan rate 9.5% over five years
• Education costs rise 8% a year; education fund earns 10% a year (8% in the sensitivity check)
• Retirement assets grow 8% a year, with no new contributions counted
• Monthly set-asides are made at the start of each month and rounded to the nearest ₹100
• Returns are assumptions for planning, not forecasts or promises

Moving forward

If you run an advisory firm or a wealth business and want to see how this would fit your data and your advisers, start with our three-week Readiness Audit. You can also read how we govern AI in our own work in our AI governance policy, or talk to an architect.

Sources

1. SEBI, Investment Advisers Regulations, 2013 (as amended)
2. SCC Online, SEBI introduces the concept of AI in its Intermediaries Regulations, February 2025
3. Sahamati, What is Account Aggregator?
4. Ministry of Electronics and Information Technology, Data protection framework