تحلیل RFM مشتریان

Not all customers are worth the same, and one message for everyone is too much for a loyal customer and too little for a lost one. RFM asks three simple questions of each customer: when they last bought (recency), how often (frequency) and how much they spent (monetary) — and on that basis places customers in six actionable segments.

This tool reads your order list in this browser, reports data quality (bad rows, duplicate orders, returns), gives each customer three scores from 1 to 5, and delivers the output with a short guide for each segment and a CSV file. Your file is never sent to the server.

• Each customer has three figures: recency = days from last purchase to the reference date (if you do not give a reference date, the file's latest date); frequency = order count; monetary = sum of order amounts.
• Rows that share the same customer ID and order ID are one order: amounts are summed and the earliest date is kept. Without an order ID column, each row is one order.
• A negative amount is a return: by default it is subtracted from the customer’s total spend but is not counted as a purchase and does not affect recency; in “remove” mode it is discarded. A customer with only returns gets no score.
• Scores 1–5 for each dimension use rank quintiles: score = 1 + floor(5 × count of worse customers ÷ total customers). "Worse" for recency means more days; for frequency and monetary means lower value. Ties always get the same score (shared rank), so quintiles may not be exactly one-fifth each.
• With fewer than 20 customers, quintiles are not meaningful and the tool automatically (and with notice) uses fixed thresholds; you can also choose them: recency ≤30 days → 5, ≤90 → 4, ≤180 → 3, ≤365 → 2, more → 1; frequency 1 → 1, 2 → 2, 3–4 → 3, 5–9 → 4, 10+ → 5; monetary vs mean of customers in this file: under 0.25× → 1, up to 0.5 → 2, up to 1 → 3, up to 2 → 4, 2× and above → 5.
• Six segments are a full partition of scores, evaluated in order: Champions (R≥4, F≥4, M≥4); Loyal (R≥3 and F≥4, or R≥3 and F≥3 and M≥4); Potential (other R≥3); At risk (R≤2 and F≥3 or M≥3); Hibernating (R=2 with low F and M); Lost (R=1 with low F and M).
• File reading: BOM is stripped; delimiter (comma, semicolon, tab) is inferred from the first row; Persian and Arabic digits are accepted; period and "٫" are decimal, and comma, "٬" and space are thousands separators (only exception: multi-group three-digit dots like 1.250.000); dates with years 1200–1500 are read as Jalali and 1800–2200 as Gregorian. Rows with no customer, no date or an unreadable amount are reported with line numbers.

• RFM only summarises past purchase behaviour; it knows neither customer profitability, nor cost to serve, nor why they stopped buying. Labels describe patterns, not judgements about a person.
• Each segment’s suggestion is a “general guide” written the same for every business and not inferred from your data; the right action depends on your product, margin and team capacity.
• Quintiles are relative: "Champion" means best in this file, not best in the market. With a small file or short time window (under three months) segments are unstable and the tool reminds you.
• If customers appear in the file under several different IDs (guest and member, two phone numbers), one person is counted as several customers; unifying IDs before analysis is your job.
• Very large files (hundreds of thousands of rows) may slow the browser; the score table shows at most the first 5,000 customers but the CSV export is complete.
• This tool is a calculator, not advice: none of the segments is investment advice or a decision about individuals.