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 →
Claude maps remedies to policy conditions and tiers, then drafts conditional wording. You check the matrix, then the wording.
Collect the case files
Map the remedies
Check the matrix
Draft the wording
Check the wording
Use: You
The draft customer reply, requested remedy lines, approved remedy policy and agent approval-limit table, and case eligibility facts.
Check the policy and approval table are the supplied approved versions. Never paste credentials or full payment details.
Use: Claude
Stage 1 in one Claude chat: upload the files and ask it to map each proposed remedy to policy conditions and approval tiers.
What to paste into the tool
Task: Draft remedy matrix with proposal IDs, values, clause citations and tiers. Using only the attached files, build a matrix with one row per remedy line in the draft reply: value and currency as supplied, each eligibility condition with the clause quoted, case fact that meets it or still to confirm, and the approval tier from the table. Missing facts must not count as eligibility. Use only the supplied case evidence and rules. Do not request or reproduce passwords, authentication codes, session tokens or payment credentials. Do not send, change records or accounts, assign work or commit any remedy or deadline. Mark every missing or contradictory fact as still to confirm instead of inferring it.
Use: You
Check every condition, value and tier in the matrix against the original tables and correct it.
Check each policy condition, monetary value and authority tier against the supplied original tables; missing facts do not count as eligibility.
Use: Claude
Stage 2 in the same chat: ask for eligibility and approval-limit gap findings and conditional wording that implies no approval.
What to paste into the tool
Task: Gap findings and policy-grounded response alternatives, with no remedy approved; Unresolved conflicts and explicit human decisions; no downstream action taken. From the corrected matrix only, list eligibility gaps and approval-limit gaps, then draft alternative wording that offers nothing the policy and tier do not support and implies no approval. Use only the supplied case evidence and rules. Do not request or reproduce passwords, authentication codes, session tokens or payment credentials. Do not send, change records or accounts, assign work or commit any remedy or deadline. Mark every missing or contradictory fact as still to confirm instead of inferring it.
Use: You
Check the wording implies no eligibility, exception or approval; no onward action.
Check every proposed remedy has an eligibility-condition result and value-based approval-tier result or an explicit open item; no refund, replacement or exception is approved. Check the wording does not imply eligibility, exception or financial approval; any proposed offer still needs separate authorised approval.
ChatGPT compares values with the tier table and drafts the findings. You map policy by hand and check everything.
Enter the spreadsheet
Compare the tiers
Map the policy
Draft the findings
Check the wording
Use: You
The draft reply, remedy lines, policy and case facts. Enter each remedy line, its value and the tier table into one spreadsheet, one row per remedy.
Check each entered value, currency and tier threshold against the supplied originals.
Use: ChatGPT
Stage 1 only: upload the sheet and ask ChatGPT to add a column with the approval tier whose limit covers each value.
What to paste into the tool
Task: Table with the tier result per remedy row. Using only the uploaded sheet, add a column with the approval tier whose limit covers each value; if the value, currency or tier boundary is missing or ambiguous, write still to confirm. Use only the supplied case evidence and rules. Do not request or reproduce passwords, authentication codes, session tokens or payment credentials. Do not send, change records or accounts, assign work or commit any remedy or deadline. Mark every missing or contradictory fact as still to confirm instead of inferring it.
Use: You
Map each remedy to the policy eligibility conditions by hand and check the tier column.
Check each policy condition, monetary value and authority tier against the supplied original tables; missing facts do not count as eligibility.
Use: ChatGPT
Stage 2 in the same chat: ask for the gap findings and conditional alternatives from the completed matrix, with no remedy approved.
What to paste into the tool
Task: Draft gap findings and response alternatives, with no remedy approved. From the completed remedy matrix I paste here, draft: 1) the eligibility and approval-gap findings for each remedy, with clause citations, and 2) policy-supported conditional wording alternatives for the reply. My completed matrix: [paste the completed matrix]. Approve no remedy, exception or refund; leave unresolved conflicts and needed human decisions explicit. Use only the supplied case evidence and rules. Do not send, change records or accounts, assign work or commit any remedy or deadline.
Use: You
Check the drafted findings and wording line by line against the matrix and the originals; correct them; no onward action.
Check every proposed remedy has an eligibility-condition result and value-based approval-tier result or an explicit open item; no refund, replacement or exception is approved. Check the wording does not imply eligibility, exception or financial approval; any proposed offer still needs separate authorised approval.
Gemini Notebook pulls the policy clauses with citations. You do the money comparison by hand. Claude drafts the wording.
Load the policy sources
Quote the clauses
Compare the tiers
Draft the wording
Check the wording
Use: You
The draft reply, remedy lines, policy, tier table and case facts; load the policy and facts into a notebook, keeping the tier table out.
Check the policy is the approved version. Never paste credentials or full payment details.
Use: Gemini Notebook
Stage 1, conditions only: ask source-bound questions to quote the policy clause for each eligibility condition and the case fact that meets it.
What to paste into the tool
Task: Answers with citations to the policy and case sources. Question template: from the selected sources only, quote the eligibility conditions for this remedy and the case fact that meets each; if a fact is still to confirm, say so. Use only the supplied case evidence and rules. Do not request or reproduce passwords, authentication codes, session tokens or payment credentials. Do not send, change records or accounts, assign work or commit any remedy or deadline. Mark every missing or contradictory fact as still to confirm instead of inferring it.
Use: You
Compare each remedy value with the approval-limit table by hand and finish the matrix.
Open each citation against the original; no citation means still to confirm.
Use: Claude
Stage 2: paste the finished matrix into a Claude chat and ask for conditional wording that offers nothing the matrix does not support.
What to paste into the tool
Task: Gap findings and policy-grounded response alternatives, with no remedy approved; Unresolved conflicts and explicit human decisions; no downstream action taken. From the pasted matrix only, list gaps and draft alternative wording that offers nothing the matrix does not support and implies no approval. Use only the supplied case evidence and rules. Do not request or reproduce passwords, authentication codes, session tokens or payment credentials. Do not send, change records or accounts, assign work or commit any remedy or deadline. Mark every missing or contradictory fact as still to confirm instead of inferring it.
Use: You
Check the wording implies no eligibility, exception or approval; no onward action.
Check every proposed remedy has an eligibility-condition result and value-based approval-tier result or an explicit open item; no refund, replacement or exception is approved. Check the wording does not imply eligibility, exception or financial approval; any proposed offer still needs separate authorised approval.
Best when: One assistant reads the policy and writes alternatives
What it does: Claude maps remedies and drafts; you check
Strong at: Matrix and originals stay in the same chat
Limit: Tool tier comparison not evidenced; re-check values
Free plan: Yes, limited (5-hour and weekly caps)
Best when: The tier table is numeric; the tool does tables
What it does: ChatGPT checks tiers and drafts; you map policy
Strong at: You choose the columns, calculations and groupings
Limit: Manual data entry errors carry through
Free plan: Yes, limited (uploads, images, voice, analysis)
Best when: Clause lookup needs citations; tiers stay with you
What it does: Gemini Notebook quotes clauses; Claude drafts wording
Strong at: The tier table itself never goes to a tool
Limit: Tool and person hand-offs add copying steps
Free plan: Gemini Notebook: Yes; Claude: Yes, limited
Before you paste into any AI tool, remove any confidential information. Put it back at the end.
Deciding between ChatGPT and Claude for this? See the comparison.
Never paste credentials or full payment details.
Billing remedy reviewer compares the evidence worksheet with the supplied originals before the second stage starts.
No customer secrets in any prompt or handoff; cut down customer data before upload.
Billing remedy reviewer confirms the final draft; the draft is preparation only and no downstream action is taken.
Check the result against your original inputs. Approve it before sharing or acting.