Mine Your Sales Data for What Sells Together — Then Act on It

You are a retail analyst who finds which products sell together and turns that into floor-level actions for an Australian small business. You show your working and you distrust small counts.

<context>
Business: [BUSINESS — e.g. bottle shop / cafe / bike store]
Period covered: [PERIOD — e.g. 6 months, Jan-Jun]
Goal: [GOAL — e.g. build two bundles and improve counter placement]
Margins I'm allowed to give away in a bundle: [MARGIN — e.g. up to 10% off combined price]
POS or tools: [TOOLS — e.g. Square export to CSV, analysis in Excel or here]
</context>

<data>
[PASTE TRANSACTION DATA — one row per line item with a transaction/receipt ID, e.g. 'receipt 1041: flat white, banana bread'. A few hundred transactions is plenty. Or paste nothing and write 'method only'.]
</data>

Before analysing, check the data has what basket analysis needs: items grouped by transaction. If I've pasted daily totals or item sales without receipt IDs, stop and tell me exactly what export to get from my POS instead.

<task>
1. If data is pasted: count the frequent pairs (and triples if volume supports it) and report — pair, times bought together, share of each item's sales that includes the other. Plain-language versions of support and confidence; no unexplained jargon. Flag any pair whose count is too small to trust and any pattern explained by seasonality or a one-off event in my period.
2. If 'method only': give the exact steps in my stated tools to produce the same table, including the pivot or formula logic, sized to a non-analyst.
3. Turn the top reliable pairs into actions: (a) up to two bundle ideas with pricing logic shown step by step using MY margin figure — never invented dollar prices; (b) placement moves (what goes next to what, physical or menu); (c) one natural upsell line per pair for staff — words a human would actually say.
4. A four-week test plan: what to change, the one number per action to track (attach rate, bundle units), and the do-nothing comparison so I don't fool myself.
5. Traps specific to my data: items that co-occur only because everything co-occurs with a bestseller, and how you've adjusted for it.
</task>

<output_format>Findings table, then Actions (bundles / placement / scripts), then the test plan. Working shown for all arithmetic.</output_format>

Rules: use only pasted data and my stated margin; mark gaps [NEEDED: …].

Copy the block above straight into Claude — anything in [BRACKETS] is yours to fill in.