Compare the asset register with an equipment count
A draft matched-equipment table and separate exception lists with source row IDs for the register and the count.
Choose one of three approaches. The steps for that option appear below.
Choose an option above. On a phone, swipe the tabs left or right to see all options →
ChatGPT: matches the lists in Python, you check the code
ChatGPT runs a code-backed match on a structured export. You inspect the code and recalculate samples.
You
Get CSVs, rules ready
ChatGPT
Match in Python
You
Check and save pack
Get the CSVs and rules ready
Use:You
Have ready: asset-register and physical-count CSVs (tag, description, location, custodian, count date where known) and approved tag-normalisation and duplicate-handling rules.
Check
Stop if upload permission is absent. Verify date, currency, completeness and the supplied rules; do not replace missing rules with guesses. Keep inputs in an access-controlled local draft.
Draft the matched table and exceptions
Use:ChatGPT
Attach the CSVs and paste the rules. Use ChatGPT data analysis for the draft; it shows the Python code, row counts, assumptions and exception rows.
Details
What to paste into the tool
Task: Produce a draft matched-equipment table and separate exception lists with source row IDs for the register and the count. Inputs: [ATTACH: the approved asset-register and physical-count CSVs from step A-1] [PASTE: tag-normalisation rule] [PASTE: duplicate-handling rule]. Use only these inputs. Show the Python used and a row-level audit trail with source IDs. Match on approved asset tags only. Do not merge near-matches automatically. Keep missing tags, duplicate tags, location differences, uncounted assets and unregistered counted items separate. A missing count is not proof of loss. Do not choose write-offs or disposal entries. Never silently drop or repair rows. Report missing facts as not stated. Do not access accounts or external sources, send messages, post entries, change the master ledger, sign, publish or pay. Produce a draft for human review only.
Check the output and save the pack
Use:You
Check the output, correct or reject it; save locally, marked draft. Account for every row from both sources; link every exception to its rows.
Check
Check counts against each input; test leading zeros, blank tags, duplicate tags and approved normalisation. Never join duplicate tags many-to-many. Do not send, sign, pay, publish, post journals or update master records.
Copilot in Excel: suggests formulas, you build and test
Copilot in Excel suggests formulas from synthetic sample columns only. You apply checked formulas locally to the real draft.
You
Get CSVs, rules ready
Copilot in Excel
Suggest formulas
You
Test, build, check
Get the CSVs and rules ready
Use:You
Have ready: asset-register and physical-count CSVs (asset tag, description, location, custodian, count date where known) and approved rules.
Check
Stop if upload permission is absent. Copilot must not modify the workbook or make a finance decision.
Suggest formulas and explanations
Use:Copilot in Excel
Ask Copilot in Excel chat mode for formulas and explanations, using synthetic sample rows and column names only, no real figures.
Details
What to paste into the tool
Task: Suggest spreadsheet formulas and test cases for a private draft worksheet, not a finance decision. Inputs: [PASTE: 5-10 synthetic example rows for the register and the count, no real assets] [PASTE: column names] [PASTE: approved tag-normalisation and duplicate-handling rules]. Use only these inputs. Match on approved asset tags only. Do not merge near-matches automatically. Keep missing tags, duplicate tags, location differences, uncounted assets and unregistered counted items separate. A missing count is not proof of loss. Do not choose write-offs or disposal entries. Explain references, date boundaries, sign handling and error conditions. Give formulas as text, with sample expected results; do not edit files. Missing inputs must produce a 'not stated' note or an explicit exception. Do not choose accounting policy or recommend action. Do not send, post, sign, publish or pay.
Test formulas, build the draft, check it
Use:You
Test each formula on synthetic normal and edge cases before applying it. Build the local draft yourself and account for every row and exception.
Check
Save locally, marked draft. Do not send, sign, pay, publish, post journals or update master records.
Claude: explains your checked result, you match by hand
You do all matching and arithmetic first. Claude only summarises a small, checked text note.
You
Match by hand, write note
Claude
Summarise the note
You
Check and save pack
Match by hand and write a note
Use:You
Have ready: both CSVs and rules. Match by hand on approved tags; group exceptions with row IDs; infer no disposals. Note the checked results.
Check
Stop if upload permission is absent. CSVs need asset tag, description, location, custodian, count date where known.
Summarise the note and list open questions
Use:Claude
Paste only the note of checked results and source references, not the data file. Ask Claude to summarise it and list open questions without recalculating.
Details
What to paste into the tool
Task: Write a count-review checklist grouped by missing tags, duplicate tags, location differences and unmatched items, using the human-checked exception list. Input: [PASTE: the human-checked note from step C-1]. Use only this note. Preserve every amount, unit, date, label and source ID exactly. Do not recalculate, infer missing facts or add a judgement. Match on approved asset tags only. Do not merge near-matches automatically. Keep missing tags, duplicate tags, location differences, uncounted assets and unregistered counted items separate. A missing count is not proof of loss. Do not choose write-offs or disposal entries. Say 'not stated' for missing information. Separate facts from open questions. Do not draft an external message or instruct anyone to act. Do not access accounts, send, post, sign, publish or pay. Return internal draft text only.
Check the summary and save the pack
Use:You
Check every sentence and number against your note; remove unsupported interpretations and keep not-stated facts visible. Account for every row and exception.
Check
Save locally, marked draft. Do not send, sign, pay, publish, post journals or update master records.
Compare the three approaches
Option A: ChatGPT: matches the lists in Python, you check the code
Best when: A structured export needs repeatable batch matching you can inspect.
What it does: ChatGPT runs the match in code; you recalculate samples.
Strong at: Shows the Python code and row-level audit trail.
Limit: Free plan limits file uploads and data analysis.
Free plan: Yes, limited (file uploads, data analysis)
Option B: Copilot in Excel: suggests formulas, you build and test
Best when: You want to build and own the formulas.
What it does: Copilot in Excel suggests formulas; you build and test them.
Strong at: Explains formulas and gives sample expected results.
Limit: On iPhone and Android, chat mode only.
Free plan: Paid plan needed
Option C: Claude: explains your checked result, you match by hand
Best when: You want to do all matching and arithmetic yourself.
What it does: You match by hand; Claude summarises the checked note.
Strong at: Preserves every amount, date and source ID exactly.
Limit: Usage is capped in 5-hour and weekly windows.
Free plan: Yes, limited (usage capped in 5-hour and weekly windows)
Before you paste into any AI tool, remove any confidential information. Put it back at the end.
Check matched and unmatched row counts against each input; test leading zeros, blank tags, duplicate tags and approved normalisation. Verify location differences against the original count notes.
Reconcile completeness and totals, preserve open questions, and save only a draft.
Check the result against your original inputs. Approve it before sharing or acting.
Availability notes
Based on official product documentation checked on 2026-10-29.
What still needs checking
Results on your own data depend on your inputs and rules. Always check the output yourself.
Owner-approved inputs and policy conventions must be supplied at use time.