Day 02 · Build An Investment Strategy AI Agent Inside Claude Projects

Why this matters

Most retail investors who ask Claude for an investment strategy paste three sentences and accept whatever comes back. That output is generic by design.

The model has no idea what capital you actually have to deploy. It has no idea what timeline you're underwriting against, or what risks you'll live with through the cycle.

When you set Claude up as a dedicated Project and let it interview you before generating anything, the output changes shape. You get a strategy tied to your specific capital, goals, and risk profile. The output is only ever as good as the inputs you give it.

The strategy that comes back is only ever as good as the inputs you put in.

What you're building

The end result

A personalised investment strategy as an editable HTML page. It covers your capital deployment plan, the strategy archetypes that fit your situation, sample target deal profiles, and a 12-month execution roadmap. Every figure is tied to the inputs you gave Claude in the interview.

This is the layer above a deal underwrite. It defines which deals are even worth running through Day 01's model in the first place.

What you need

Claude Pro at minimum. Projects is gated to the paid tier, so the free plan won't get you past step i.

The other requirement is five minutes plus your investor profile written down. Capital available now, monthly contribution capacity, time horizon, primary goal, risk appetite.

Step 01 · Set up the Project

Three clicks. One minute. Once.

You only do this once. The Project stays in your Claude sidebar afterwards, and every new chat inside it inherits the context you give it.

i.Open Claude and go to Projects in the sidebar.App

ii.Click New Project. Name it "Investment Strategy AI Agent".App

iii.Add a one-line description so the Project remembers what it's building. Example: "Personalised property investment strategy for a long-horizon investor."Done

Heads up

Projects only ships on Claude Pro and above. If your sidebar has no Projects link, upgrade your plan inside Settings before starting. The free tier won't get you past step i.

Step 02 · Give it your context

Paste your investor profile.

Open a new chat inside your Project. Paste the template below into the chat box, swap each bracketed field for your own situation, and send. The output you get later is only ever as good as the inputs you put in here.

01

The Template

Investor profile

Four lines. The structure stays the same for every investor. Only the bracketed inputs change.

Copy & Paste

I am a [your age]-year-old [your nationality] and current [country of residence] resident. I have [your available capital today] available today, plus a monthly investing capacity of approximately [your monthly capital]. My primary goal is [your investment objective, e.g. long-term capital appreciation, monthly cash flow, hybrid]. I am [your risk appetite, e.g. comfortable with construction risk and leverage / risk-averse, ready-to-move only].

How I filled mine in

I am a 30-year-old British national and current UAE resident.

I have AED 350,000 available today, plus a monthly investing capacity of approximately AED 15,000.

My primary goal is to grow my net worth through long-term capital appreciation.

I am comfortable with construction risk, market cycles, and leverage.

Step 03 · Let it interview you

Don't rush this part.

Claude will read your profile and start asking clarifying questions. Treat them as the strategy. The questions Claude can't infer are the ones that decide whether the plan that comes back is actually yours.

i.

The kinds of questions it asks

Whether you have existing property exposure, and in which markets
Whether your available capital is fully liquid or partly tied up
Whether your monthly contribution is post-tax and net of other commitments
Whether you have a target portfolio size, or whether the strategy should propose one

ii.

What a strong answer looks like

Be specific. Ranges are fine if you tell Claude which end is realistic.
Flag any constraint Claude doesn't know about (visa status, tax residency, dependants).
Where you don't know, say "propose a reasonable default and flag it". Claude will do that and label it.

How to read the interview phase

Treat the interview as the strategy. The artefact Claude generates afterwards is only as good as the answers it pulled out of you.
Don't end the interview early. If Claude says it has enough context, ask if there are three more questions worth asking. Usually there are.
Read it back. Once you're done, ask Claude to summarise what it now knows about you. Correct anything wrong before generating the strategy.

Step 04 · Generate the strategy tool

One prompt. A working tool, not a doc.

Once Claude has read your context back correctly, send the prompt below. It does not ask for a written strategy. It asks for a self-contained interactive HTML tool: a form with your inputs, six market-assumption sliders you control, a residency-aware LTV calculator, a 3-phase roadmap, and a year-by-year projection table that simulates off-plan handovers, equity release, and reinvestment cycles.

The prompt is long by design. Every section in it closes a specific failure mode I hit when the earlier, shorter version of this prompt produced a tool that double-counted off-plan gains and ignored construction instalments. The calculation rules are the load-bearing part.

02

The master prompt

Strategy tool generator

Form spec, projection model rules, and design system in a single message. Claude returns one self-contained HTML file you can open in a browser and use.

Copy or Download

Download .md

# UAE Property Strategy Tool. Master Prompt. Generate an interactive UAE property investment strategy tool as a single self-contained HTML file. The tool has two parts: a form and a generated output. --- ## Form (four sections) 1. About You. Name, age, nationality, employment status, UAE residency status (pill selector: UAE Resident / Non-Resident / Planning to Relocate), and an open text field for goals and context. 2. Capital & Capacity. Available capital today (AED), monthly investing capacity (AED), existing UAE properties owned, existing mortgage debt. 3. Goals & Strategy. Target portfolio value (AED), investment timeline (years), primary goal (pill selector: Capital Appreciation / Rental Income / Balanced), risk appetite (pill selector: Conservative / Moderate / Aggressive), strategy preference (pill selector: Off-Plan Focused / Ready Properties / Mixed). 4. Your Market Assumptions. Six sliders the user controls entirely: annual capital appreciation %, gross rental yield %, off-plan launch-to-handover gain %, mortgage interest rate %, booking deposit %, and transaction fees %. No market data is hardcoded. All projections derive purely from the user's slider inputs. Below the form is a Generate My Strategy button. On click, the tool computes and renders: - A summary strip of four stat cards (total capital, LTV rate based on residency, portfolio target, leverage multiple required). - A 3-phase roadmap with dynamically generated text and bullet points drawn from the user's actual input values, not template copy. - A year-by-year projection table (Year, Cumulative Cash Invested, Properties Held, Gross Portfolio Value, Mortgage Balance, Net Equity, Estimated Rental Income) computed by a JavaScript simulation model that correctly times off-plan handovers, equity release via mortgage, and reinvestment cycles. --- ## Projection Model. Calculation Rules. The JavaScript simulation must follow these rules precisely. Each rule addresses a specific bug to avoid. ### Payment plan structure (per property) - Year 0 (booking): investor pays bookingPct% + feesPct% of purchase price from available cash. - Years 1 to N minus 1 (construction): investor pays 5% of purchase price per year in instalments, deducted from that year's monthly savings before any surplus is added to cash. - Year N (handover): remaining developer balance (= 100% minus bookingPct% minus 5% times construction years) is settled via a bank mortgage. Mortgage = handover value times LTV. Cash released = mortgage minus remaining developer balance. ### Off-plan appreciation. No double-counting. - Off-plan properties are held at purchase price in the model until handover. Do not apply annual growth during construction. - offPlanGain% is applied once only, at the handover event: handoverValue = purchasePrice times (1 + offPlanGain/100). - Annual growth (annualGrowth%) applies only to completed properties, starting the year after handover, compounding each year: value *= (1 + growth/100). ### Monthly savings. Net of instalment obligations. - Each year, the cash added from savings = max(monthly times 12 minus active construction instalments, 0). - Active construction instalments = sum of purchasePrice times 5% for each property currently under construction (bought but not yet handed over). - cumulativeInvested tracks total gross savings committed (capital + monthly times 12 times year) regardless of where the cash went. This is the investor's total cash outlay. ### New property acquisition. Dual affordability constraint. When deciding whether and how large a new purchase is, apply both constraints and take the minimum: - Deposit constraint: maxByDeposit = deployableCash / (bookingPct% + feesPct%). - Income constraint: maxByIncome = freeMonthlySavingsPA / 5% where free savings = monthly times 12 minus existing instalment obligations. - Only proceed if the resulting property value is greater than or equal to AED 400,000. - Always maintain a minimum AED 50,000 cash reserve. Never buy if doing so would breach the reserve. - Cap simultaneous off-plan properties under construction at 2. ### Net equity. Accounts for all outstanding obligations. Net Equity = Gross Portfolio Value minus Bank Mortgages minus Remaining Developer Balances. Remaining developer balance per off-plan property: - paidFraction = bookingPct% + 5% times min(yearsSincePurchase, constructionYears minus 1). - remainingDevBalance = purchasePrice times max(1 minus paidFraction, handoverBalancePct). - This balance reaches zero at handover (replaced by the bank mortgage). The Mortgage Balance column in the table shows bank mortgages only. Net Equity uses total debt (mortgages plus developer balances). A table footnote explains the difference. --- ## Design System - Background: #E8DFD0 (cream), #DDD4C5 (card fill), #D0C5B2 (alternate rows). - Headings: Instrument Serif (Google Fonts). Section titles, phase titles, output header. - Body: Satoshi (Fontshare). All paragraph and label text. - Data: JetBrains Mono (Google Fonts). All numbers, table cells, range values, form labels, toolbar. - Accent: #3E1F1F (burgundy). Key italic phrases (<em class="k">), borders, range thumbs, phase card left border, generate button. - Pill selectors styled as mono-font toggle buttons. Selected state fills burgundy. - Range sliders styled with a burgundy thumb and live value display. --- ## Behaviour - Output section hidden on load. Appears and scrolls into view after Generate is clicked. - All phase text and projection numbers update from inputs on each generation. - Residency status automatically sets LTV (75% resident, 50% non-resident). - No external API calls, no hardcoded market data, no clipboard export required.

Then iterate

Open the generated HTML in a browser, run it against your real numbers, and feed edits back to Claude in plain English. Add a CSV download of the projection table. Add a second residency scenario side-by-side. Toggle AED to GBP. Pin the working file inside your Project so every future chat starts from your tool, not a blank slate.

The honest bit

What it still gets wrong.

Four real weaknesses of this workflow. Read them before you trust any output.

i.

Project context decays

After a few dozen messages in the same chat, Claude starts losing the earliest details you gave it. When the strategy stops matching your profile, start a fresh chat inside the same Project. The Project-level context survives the reset.

ii.

It invents market specifics

Ask for "the best Dubai community to invest in" and Claude writes something confident. It does not have a live feed of off-plan launches, payment plans, or service-charge changes. Treat every named project or community as a starting point you verify, not a recommendation.

iii.

Skip the interview, get a generic plan

If you ask for the strategy before Claude has interviewed you, you get the same advice it would give anyone. The interview is where the personalisation happens. Don't try to compress it.

iv.

One strategy at a time, not a portfolio comparison

This prompt produces a single coherent strategy tied to your profile. It will not run three competing strategies side by side, or compare your plan against a benchmark portfolio. Both of those are separate prompts in this series.

The workflow

How this stacks.

Day 01 was the install and the underwrite. Day 02 is the layer above: the strategy that decides which deals are even worth modelling. Every prompt in this series compounds with the others.

The workflow

Day 01 installed the underwriter. Day 02 builds the strategist that remembers you.

The underwrite tells you whether a deal works. The strategy tells you which deals are even worth running through the underwrite. Day 01 plus Day 02 is the funnel that builds a portfolio that holds together.

What "done" looks like

By the end of session one.

The session-one vision

A strategy document Claude knows you by name.

Total time, end to end, around seven minutes. The output is an editable HTML page that reflects your capital, your goals, your time horizon, and the risks you'll actually live with.

Every prompt in this series feeds back into the same Project. The strategy you build in session one is the one you'll keep refining for the next 30 days, against the real deals you screen.

Build anInvestment StrategyAI Agent.