Planning retirement and later life for my family on one tested engine
Our household kept coming back to one question: when could we stop working and still afford the life we want? Every tool I tried either ignored UK tax or treated the future as certain, so I built my own. Then my parents started asking a different question, about later life, care and how they'd like to help the family. The same engine now answers that too.
- Company
- Side project
- Year
- 2026
- Role
- Research, product design and engineering
- Team
- Just me, using Claude Code
- 0
planners on one engine: our retirement, and my parents' later life
- 0
tests on the tax engine, including HMRC's own worked examples
- 0 weeks
of evenings and weekends, from the first spec to both planners
Try a simplified version of the planner, using an illustrative household. Move the sliders to see the chance of the money lasting, and how much your change moved it.
Chance the money lasts to 100
84%
Within reach
The problem
We wanted to know the earliest age we could retire at a given level of spending, and how likely the money was to last. Consumer planners model market ups and downs but ignore UK tax rules. Adviser software handles tax but assumes a flat rate and a single, predictable future. Neither could answer the question for us.
I soon realised the maths was the easy part. A plan only helps a household if everyone in it can read it, and not everyone enjoys spreadsheets as much as I do. So I set myself a simple test: would someone open it on their own, without me there to explain it?
Design principles
01
Lead with the answer
The page opens with the age you could retire and the chance of success. The workings sit underneath for anyone who wants them.
02
Show progress
Every change is read back against your current course, for example ‘retire two years earlier’, instead of as a shortfall.
03
Be honest about uncertainty
Results are shown as ranges and probabilities. A single precise figure would claim more than anyone can know about markets.
04
Design for the other person
The test for every screen was whether someone who doesn't enjoy spreadsheets would open it on their own.
Planner one: our retirement
- Shows the age we could retire against how much we spend and the chance the money lasts, and updates as we change how much we pay into pensions or how fast our income grows
- Simulates thousands of possible market histories to age 100, including the risk of a crash early in retirement
- Lets us test four habits we control, such as spending and saving, and shows each change against our current course
- Ranks where this year's savings should go (pensions, ISAs or overpaying the mortgage) and explains why
- Models life events such as remortgaging, moving house or a bonus, each compared with our current plan
Checking the maths
A planner you can't trust is worse than none. Each tax year's rules are stored as data, and every threshold records its source on legislation.gov.uk or in HMRC's manuals, whether it has been enacted or only announced, and when I last checked it. Every tax rule is checked against HMRC's own worked examples, in over a thousand tests.
The interface never calculates tax itself. It displays what the engine has worked out, and a test checks that the two agree to the penny.
Planner two: my parents' later life
My parents were starting to think about later life: whether they'd be comfortable, what care might cost, how they'd like to help the family, and how inheritance tax fits in. I built a second planner on the same engine to help them think it through together. It has a page for each of their big questions: Will we be comfortable? What do we have? How would we like to help the family? How much tax might be due? Which plan suits us best? What happens, and when?
You combine options into a plan, compare plans side by side and move through the years to see what changes. Every chart also has a written summary and a table, so nobody has to read a graph to get the answer.
What I learned
- Correcting a few household figures changed the answer more than any improvement to the model, so the planner now lists the information it's missing
- Writing a spec and a plan before each phase let me build something this thorough on my own, with Claude Code writing much of the code and me making the decisions