Claude Prompt Chain: Check Planning Data Against Statements and Build the Missing-Documents List

What This Builds

Before you enter a new household into the planning software, you need to know whether the statements in hand match the accounts on the intake fact sheet, who is missing a document, and which statements are too old to use. This chain does that sorting in three prompts. Claude turns the statement text you paste into a table, compares the table with the fact sheet, and produces a missing-documents list by household member plus a data-entry checklist.

Claude never supplies the numbers that go into the planning software. You enter every figure from the statement itself. The table is a map that shows you where to look.

Prerequisites

• A Claude account (Pro, $20/month), in the workspace or plan your chief compliance officer (CCO) has approved for client financial data
• Redacted statement and benefits summary text, one document at a time, each headed with a label you type
• The household fact sheet from intake, using roles and account labels
• Your own staleness rule, for example a number of days after which a statement is too old for your purposes

Total ongoing cost: one Claude plan ($20/month). Use Opus 5.5 or the strongest model your plan offers.

Before You Paste: Client Data Rules

Statements show balances, holdings and account identifiers, which is some of the most sensitive data a firm holds.

1. Ask your CCO whether Claude is approved for client financial data and in which plan or workspace. If it is not, do not paste anything from a statement.
2. Check the data settings on your plan: whether chats are used for training and how long they are kept. A free consumer plan may use chats to train models unless a setting is off.
3. Before you paste, strip out names, account numbers, Social Security numbers, dates of birth, addresses, employer names and beneficiaries' names. Use "Client A", "Client B" and "Joint", and account labels such as "Client A 401(k)". Keep health details out entirely.
4. Paste only what you need: account label, as-of date, value and page. Leave the holdings detail behind.

The Concept

This automates the first pass of a sanity check that a careful paraplanner already does by eye. Does each account on the fact sheet have a statement? Does each statement belong to an account on the sheet? Is each one recent enough? Claude does the first pass of matching, and you do the verifying. Three prompts keep each step small enough to check against the pages. After each prompt you stop, check, and correct before the next one builds on it.

You state the staleness rule yourself. Claude counts and flags. It does not decide what counts as too old.

Build It Step by Step

Part 1: Prepare the Source Text

Paste the statement text into a document of your own with a header line before each statement, for example DOCUMENT 1: Client A 401(k) statement. Mark each page break as PAGE 2 and so on. The page markers let Claude cite where each figure came from. Write the fact sheet with one account per line.

Part 2: Prompt 1, Extract the Table

Start a new conversation. Paste the statement text where the square brackets are:

Below is redacted text from several statements and benefits summaries. Each document starts with a line beginning DOCUMENT, and pages are marked PAGE.

[paste the statement text]

Extract one row per account into a table with these columns: Account label | Owner role (Client A, Client B or Joint) | Account type | As-of date | Value as printed | Source document and page.

Use only what is printed. Write "not shown" for anything that is missing. Copy values exactly as printed. Do not calculate, estimate or add accounts.

Checkpoint 1. Open each statement and check every row. Is the as-of date the statement's own date? Is the value exactly as printed? Is the page right? Is every account on the statements in the table, and nothing else? Correct the table in the chat before moving on.

Part 3: Prompt 2, Compare With the Fact Sheet

Paste the fact sheet, today's date and your rule:

Here is the household fact sheet from intake:

[paste the fact sheet]

Today's date is [yyyy-mm-dd]. My staleness rule is [N] days, counted from the as-of date to today's date.

Compare the fact sheet with your table and give me four lists.
1. Accounts on the fact sheet with no statement in the table.
2. Statements in the table that are not on the fact sheet.
3. Accounts where the owner role or the account type differs between the fact sheet and the statement.
4. Accounts whose as-of date is more than [N] days before today's date, with the number of days for each.

Show your date counts. Do not decide whether the rule is right and do not suggest a different rule.

Checkpoint 2. Recount each age in days yourself. In a spreadsheet, put today's date in one cell, the as-of date in another and subtract. Models make date arithmetic mistakes, so do not trust the counts. Check each mismatch against the fact sheet and the statement, and check that the stale list uses your rule and no other.

Part 4: Prompt 3, Missing Documents and Data-Entry Checklist

Using the corrected lists from this conversation, produce two things.

A. A missing-documents list grouped by household member (Client A, Client B, Joint). For each member list the documents needed, one line each, with the reason: no statement received, or statement older than my rule. Use short plain labels with no figures. I will use this list to write a document request.

B. A data-entry checklist for the planning software. One line per account in the table: account label, owner role, account type, as-of date, and the document and page to enter the value from. Leave the value itself out. Add a tick box at the start of each line.

Checkpoint 3. Read the missing-documents list against lists 1 and 4 from prompt 2. Nothing should be on it that you did not verify, and nothing verified should be missing. Then read the checklist against the statements. The checklist deliberately holds no values, because you enter each value from the statement page named on the line.

Part 5: Use the Output

The missing-documents list feeds the Level 1 chase email prompt. You write the email, the lead advisor approves it, and it goes out through the firm's archived email. Enter data in the planning software from the statements, line by line, ticking the checklist as you go. Keep the verified table with the household file only if the firm's policy allows.

Real Example: Household 112, Four Accounts

Invented example. Today's date is 2026-10-06 and the staleness rule is 120 days. All data is invented.

Prompt 1 table:

Three statements, total of printed values 312,000 plus 88,500 plus 145,200 equals 545,700. The total is for your own tie-out and goes nowhere in the planning software.

Fact sheet (invented): Client A 401(k); Client B Roth IRA; Joint brokerage; Client B 403(b).

Prompt 2 output:

1. No statement: Client B 403(b).
2. Statements not on the fact sheet: none.
3. Type mismatch: the fact sheet lists a Client B Roth IRA and the statement shows a Traditional IRA for Client B. Check which is right.
4. Older than 120 days: Client B IRA, as of 2026-03-31, 189 days.

Checking the counts by hand: from 2026-03-31 to 2026-10-06 is 189 days (30 to the end of April, 31 to May, 30 to June, 31 to July, 31 to August, 30 to September, then 6 more). The two statements dated 2026-06-30 are 98 days old, which is under 120, so they are not stale under this rule. A 90-day rule would flag all three statements. The rule is yours, and the fact that the answer changes with it is why Claude does not pick it.

Prompt 3 output:

A. Missing documents

• Client B: statement for the 403(b), none received.
• Client B: a current statement for the IRA, because the one received is older than the rule. Also confirm whether it is a Roth or a Traditional IRA.
• Client A: none.
• Joint: none.

B. Data-entry checklist

• Client A 401(k), Client A, 401(k), as of 2026-06-30, Document 1, page 2
• Client B IRA, Client B, type to be confirmed, as of 2026-03-31, Document 2, page 1
• Joint brokerage, Joint, Brokerage, as of 2026-06-30, Document 3, page 3

You would hold off entering the Client B IRA until the new statement arrives, and you enter each value from the statement page cited. The advisor decides how to treat the type question with the client.

Time saved: the matching of fact sheet to statements and the first draft of the request list are done in minutes, ahead of the careful checking that stays with you.

What to Do When It Breaks

• Claude invents an account. It appears in the table with a source you cannot find. Delete the row, and tell Claude to use only the text you pasted.
• Dates are off by a few days. This is common. Recount in a spreadsheet and give Claude the corrected counts before prompt 3.
• A value is transposed or rounded. Compare each value to the statement. Correct it in the chat, because later prompts reuse the table.
• The table is missing a statement. The text may have been cut off in the paste. Paste that document again in a separate message.
• A scanned statement gives unreadable text. Type the label, date and value by hand for that account.
• Pasted text still contains an identifier. Stop, start a new conversation, redact again, and tell your CCO if client data reached a tool that was not approved.

Variations

• Simpler version: run prompts 1 and 2 only and build the missing-documents list yourself.
• Extended version: keep the three prompts in a Claude Project to reuse. Ask your CCO before saving client material in a Project.

What to Do Next

• This week: run it on one new household and compare with your usual intake check.
• This month: write your staleness rule into your intake procedure so everyone applies the same number.
• Advanced: pair it with the daily document collection digest so each missing document gets a tracker row.

Advanced guide for paraplanner professionals. AI outputs can be wrong, so check every figure against the source document. The planning software values always come from the statements.