How AI Can Help in Sales: A Gift Shop's Bundle and Follow-Up Plan

How AI can help in sales for a gift shop comes down to three small jobs: finding which products customers already buy together so you can price bundles, drafting follow-up messages for people who agreed to hear from you, and offering gift-note wording at checkout. Start from your own sales export, not from software built for sales teams.

Most advice on AI in sales is written for companies with pipelines, forecasts and reps. A gift shop has baskets, seasonal spikes, repeat givers and a till full of data nobody has read. Everything below runs on that data. The figures are illustrative, and every price, discount and message is your decision; the AI counts, drafts and checks.

Follow me on Instagram@sagnikteaches

Where a one-till shop's extra sales come from

Skip lead scoring and AI cold outreach. A gift shop has three levers, and each maps to one job for the AI:

Connect on LinkedInSagnik Bhattacharya

Export the sales history the AI will read

Use last year's busy period, since that is the one your December plan will resemble, and ask for line items, not order totals. Each row should say which receipt or order it belongs to, the date, the item, the quantity and the price. Unit cost helps if you have it; you can add it later. Most tills and shop platforms can export this as a CSV file. On Shopify, the Orders page has an Export button, and the export is effectively one row per item, with extra items on separate rows. A selection of up to 50 orders downloads straight away; anything bigger, such as a whole December, is emailed to you.

Subscribe on YouTube@sagnikteaches

Before you upload anything, delete the customer name, email, phone and address columns. Pair analysis needs none of them, and a chat tool is the wrong place for them. Then deal with messy item names. "Candle med 220g", "Candle - Medium" and "CANDLE M" will be counted as three products unless someone merges them, and if the export is full of such variants, cleaning messy data with AI shows how to standardise them first. Claude accepts CSV uploads, and XLSX files too when code execution and file creation are switched on; ChatGPT and Gemini take spreadsheet files as well, within each plan's limits.

Find the pairs: attach rate and lift on last December's baskets

The question is which items travel together more often than chance. Two measures answer it. The attach rate is the share of baskets containing item A that also contain item B. Lift compares how often A and B appear together with how often you would expect that if shoppers picked each at random; 1.0 means no relationship, 2.0 means twice the chance level. Attach rate tells you how big the opportunity is, and lift tells you whether it is real.

Attached is last December's sales as a CSV, one row per item sold, with columns receipt_id, date, item, quantity, price.
1. Treat each receipt_id as one basket.
2. Count baskets containing each item, and baskets containing each pair of items.
3. For the 15 pairs with the most baskets, report: baskets with both, attach rate each way, and lift (share of all baskets containing both, divided by the product of each item's share).
4. Ignore gift cards, carrier bags and delivery charges.
5. Tell me how many baskets there are in total, and how many contain two or more items.
Show the working for the top 3 pairs so I can check it in a spreadsheet. If a number can't be computed from the file, say so rather than estimating.

The last two lines matter more than they look. If you paste a summary instead of the file and ask for "likely pairs", a chat assistant will happily invent counts that sound right. Make it compute from the data, and spot-check one pair yourself in a pivot table; it takes three minutes. Here is the kind of output you would get for an invented shop with 2,600 baskets last December (illustrative):

Read the last row first. Cards and wrapping paper appear together in 162 baskets, which sounds good until you see the lift: 1.0. Both are bestsellers, so they meet in plenty of baskets by chance, and a bundle discount would only hand money to people who were buying both anyway. The notebook and pen set is the opposite: a small pair, 98 baskets, but shoppers pick them together 4.5 times as often as chance would predict. And the two top rows (candle with matches, mug with hot chocolate) combine a decent attach rate with a lift above 2, which is what you want a bundle to rest on.

Two checks before you trust any table like this. Ask whether last year was normal: a stock-out of one item, a viral post or a half-price promotion will distort the pairs, and you should tell the AI about it in the prompt. And ask whether the pair makes sense to a human. If it doesn't ("padlock + birthday balloons"), it is probably a staff habit at the till, not a customer preference.

Price three December bundles without giving the margin away

Take the three strongest pairs with a lift above 2 and give each a third item or a free extra that costs you little. Then ask for the sums, with the limits written into the prompt:

Here are three candidate bundles. For each item I've given the shelf price, unit cost and stock on hand. For each bundle: show the separate total, the cost (add $1.50 for gift wrap), the margin at 10% and at 12% off, and the most bundles I can sell from current stock. Flag any bundle whose margin falls below 45%. Never suggest a price below cost. Then suggest a name and one plain, warm line of description.

The shop in this example settled on three bundles at about 11 per cent off. The cost column includes $1.50 for a box and ribbon, which the shop gives away rather than discounting further (illustrative figures):

Sold separately, those same items earn margins of 57 per cent, 61 per cent and 65 per cent, so every bundle gives up some percentage margin in return for a bigger basket. Whether that is a good trade depends on who takes it. Suppose 160 customers buy Cosy Night In in December, and you estimate that 70 of them would have bought the candle and matches anyway, while 90 would have bought only the candle:

• The 70 who would have bought two items pay $31 for four things (the candle, the matches, the tin and the wrap) instead of $26 for two. Profit per sale goes from $15.60 to $14.60, so the bundle costs you $1.00 each, $70 in total.
• The 90 candle-only buyers go from $10.80 of profit to $14.60, a gain of $3.80 each, $342 in total.
• Net: about $272 more profit from 160 sales. The bundle pays only because of the customers it pulls up from a single item.

That is the number to watch in December: not bundle sales, but whether the candle-with-matches attach rate rises above last year's 33 per cent. If it stays flat, the bundle is mostly discounting people who would have paid full price. A deeper discount makes this worse. On the Cosy Night In items, which earn 57 per cent sold separately, a 10 per cent discount needs about 21 per cent more units sold just to keep gross profit level, and 15 per cent needs about 35 per cent more. So hold the discount near 10 per cent and add a free extra that costs you little. Here that is a box and ribbon at $1.50, a smaller bite than another $2 off.

Cap each bundle by its scarcest part

Ask for the "most bundles I can sell" figure, because a bundle sells only as many times as its scarcest component. If the shop has 90 matches sets, Cosy Night In stops at 90 however many candles are on the shelf, and the 91st customer finds a bundle that can't be fulfilled. Decide in advance what happens then: swap in a named substitute of equal cost, or take the bundle off the website the moment its scarcest item drops below ten. The party supply store bundles tutorial covers seasonal stock planning from the other side, and AI help with small-shop prices and promotions covers when a discount helps at all.

Gift-note wording customers can pick at checkout

A customer standing at the till, or at the note box on your website, often hasn't thought about what to write. Offer a few short options and more notes get written; a blank box is easy to skip, or to fill with just "Happy birthday". Ask the AI for options in your shop's voice, and give it three of your own existing cards as examples so it doesn't sound like a greetings-card factory:

Write gift-note options for a small gift shop's checkout. Occasion: birthday. Tone: warm, a little playful, never cheesy. Maximum 110 characters each including spaces, because our card holds about two lines of handwriting. Write 6 options. Rules: don't assume the relationship between giver and receiver; don't mention age, health, religion or money; use [name] as a placeholder in 2 of the 6; at most 2 exclamation marks across all six. Then write 3 more in a more formal tone for a colleague or a client.
Style examples from our own cards:
[paste three]

A typical set (illustrative) includes lines such as these, and each one needs a human read:

• "Happy birthday, [name]. May the day be full of cake and the year full of good surprises." Fine for almost anyone, 88 characters.
• "Wishing you a birthday as lovely as you are." Cut: warm between partners, odd from a colleague. The AI doesn't know who is buying.
• "For [name]: a small something for a big day." Rewrite: "big day" can read as a milestone birthday, which the prompt asked it not to hint at.
• "Treat yourself today." Keep, but only in the casual list; it's a bit pushy for a client.

Put the six on a small printed menu by the till and in a drop-down on the website, with a free-text box beside it. Occasions that need more care should stay with you:

Who writes the card is the real cost. At about two minutes a handwritten note, 300 December orders with notes is ten hours of someone's time. Decide in advance whether the note is free, whether it is a small paid extra, or whether the shop prints it in a script typeface and hand-writes only the ones customers ask for. And whatever the customer typed or chose goes on the card exactly as they wrote it. The AI's job ended when it supplied the options.

Four follow-up messages, and who may receive each

A follow-up is worth sending when it is useful to the person receiving it at that moment. The four below are the ones that earn their place in a gift shop, with the messages as a starting point (placeholders in square brackets; wording illustrative):

1. Order received, with a note reminder (web orders, within the hour). This is about their order, so it isn't marketing.

   Hi [first name], thanks for your order [order number]. If it's a gift, reply with a note of up to 110 characters before [date] and we'll handwrite it. Delivery is expected by [date]. [Shop name]

2. Last order date (past customers who agreed to hear from you, two weeks before your cut-off). This is marketing, so it goes only to people on your consented list.

   Hi [first name], our last day for orders that arrive in time is [date]. Last year you chose [item] for [occasion]; this year our Cosy Night In bundle is $31, gift-wrapped. You can unsubscribe at the link below at any time.

3. Did it arrive? (three to five days after delivery, to everyone).

   Hi [first name], we hope your order arrived safely and the gift went down well. If anything wasn't right, reply and we'll sort it. If you'd like to leave a review on Google, the link is [link]. Honest feedback, good or bad, helps other shoppers.

4. Same time next year (only if they ticked a reminder box at checkout).

   Hi [first name], last March you bought a birthday gift from us. You asked us to remind you, so here we are: [link]. Reply STOP and we won't contact you again.

Message 3 is the one with rules attached. Google's contribution policy bars offering incentives such as discounts or free goods in exchange for reviews, and bars selectively soliciting positive reviews from customers, so ask everyone, not only the people who emailed to say thanks. If the consent rules where you sell count a review request as marketing, send message 3 to everyone on your consented list instead; what the policy bars is choosing people by how happy they seem. It also bars asking staff to hit a set number of reviews. Reviews belong to the customer, so don't quote one on your shop window without asking. The policy page sets out the exact wording, and local SEO for a retail shop shows where the reviews end up.

Ask the AI to draft all four in one go, with a line saying which are order messages and which are marketing:

Write four short customer messages for a gift shop in a warm, plain voice, each under 60 words, using square-bracket placeholders for anything personal. 1: order confirmation with a reminder they can add a gift note. 2: last-order-date reminder to past customers who agreed to hear from us. 3: check the order arrived and invite a Google review from everyone, with no reward and no mention of star ratings. 4: ask whether they'd like a reminder next year for the same occasion. Label each "order message" or "marketing message".

The AI can't know who has consented; your list does. Keep it as a simple sheet with name, email, date, and where they agreed (till tick box, website, in writing), and add an unsubscribe route to every marketing message. A shop sending a few hundred messages a season can start on a free email plan: as of October 2026, Mailchimp's Free plan allows up to 250 contacts and 500 sends a month. Where the rules on marketing consent differ from place to place, ask your adviser which apply to you. The general method for follow-ups that don't read as pushy is in how to automate sales follow-ups with AI.

One December, week by week

Here is the plan as it runs for the invented shop, with the time each step takes. Set-up is about six hours spread over a week, and after that it is roughly an hour a week plus the notes themselves (illustrative):

Seven December numbers to compare with last year

Choose the measures before December so that you can't pick the flattering ones afterwards. Last year's figures come from the same export you already have (a pivot table gets them in minutes), and the weekly view in a retail KPI dashboard built from till data keeps them up to date:

Compare like with like. December trading days, weather, a competitor's sale or a shop-window refit all move the numbers, so write one line about anything unusual and keep it with the results. One month is a single data point, and it is the data you will give the AI next year.

Where a December sales plan backfires

• Counts that came from nowhere. Paste a summary and ask for "likely pairs", and you get plausible numbers that nobody counted. Make the AI compute from the file, and check one pair by hand.
• Stacked discounts. A site-wide 20 per cent code applied on top of a bundle takes Cosy Night In from $31.00 to $24.80, a margin of 34 per cent instead of 47. Exclude bundles from codes, in writing, at the till and online.
• The wrong list. An email address handed over for a receipt isn't consent to marketing. If message 2 goes to everyone in the till's history, you will get complaints and unsubscribes. Send it only to people who agreed.
• Personalisation that feels like surveillance. "Last year you chose [item]" is warm to one customer and unnerving to another. Test messages on the five customers you know best before the list sees them.

Start with the export and the pair table; everything else depends on those numbers being real. If you would rather map where AI pays off across the whole shop, the 1:1 AI implementation consultation I offer starts from exactly that question.

Gift shop owners' questions about AI for sales

Do I need email software to send follow-up messages?

Not at the start. A spreadsheet of customers who agreed to hear from you, plus your own email or phone, handles a hundred messages. Once you send regularly you want a tool that manages unsubscribes. Mailchimp's free plan covers 250 contacts and 500 sends a month at the time of writing, so check the current limits on the vendor's page.

How many bundles should a small gift shop offer at once?

Three is plenty: one for each of your strongest pairs. Every bundle ties up stock in all its parts, and the scarcest part caps how many you can sell. More than three also makes the display and the website harder to read, and you can't tell which one worked.

Can AI write notes that look handwritten?

It can suggest the wording, and a person or your printer still has to produce the card. Treat a gift note as the customer's voice. Offer a few options at checkout, let them edit one or write their own, and use exactly what they chose, with no AI embellishment added afterwards.

Is it all right to message customers who bought once?

Messages about an order they placed, such as delivery or a note reminder, are one thing. Marketing messages generally need clear consent where you sell, so ask at checkout with an unticked box, keep a record, and add an unsubscribe link. Ask your adviser which rules apply to you.

Further reads

• How Small Shops Can Use AI to Forecast Stock and Reorder — Forecast stock and reorder before the busy month
• Customer Loyalty Programmes for Retail Stores: Can AI Design One? — Turn repeat gift buyers into a loyalty scheme
• Tea Shop Marketing Ideas: Tasting Notes and Gift Sets Written With AI — Gift sets and tasting notes for a food-and-drink shop
• AI for Boutique Owners: A Weekly New-Arrivals Email Routine — A weekly email routine once your consented list grows
• How to Build an Upsell and Cross-Sell Email Sequence With AI — Turn your strongest pairs into an upsell email sequence
• Examples of AI in Retail: Eight Jobs a Small Shop Can Copy — Eight other small-shop jobs AI can take on
• AI for Retail Store Design: Layout and Display Ideas — Plan a shop re-layout with AI, safely: measure, get three options, tape-test the best, check exits and access, then measure whether sales moved.
• Starting a Business Using AI: A First-Month Plan for a Shop or Cafe — Four weeks before opening: names, supplier emails, menu or range, a cash-flow draft and launch posts, with the checks a person must do.
• Baby Store Marketing Ideas: Gift Lists and New-Parent Offers With AI — A gift-list page, a four-email welcome series and milestone offers for an independent baby shop, with prompts and a claims check for safety wording.
• Stationery Shop Marketing Ideas: School and Corporate Orders With AI — A step-by-step plan for winning school and small-office orders: the price sheet, the buyer list, AI-drafted emails, follow-ups, samples and quotes.
• Yarn Shop Marketing: Fill Knitting Classes and Promote New Stock With AI — A class calendar, listing and caption prompts, a monthly newsletter template and a three-hour routine for a small yarn shop's marketing.
• Can AI Help a Convenience Store Drive Sales? Where to Start — A convenience store owner's guide to AI sales tests: find high-margin lines in your till export, run one bundle test first, and read the result honestly.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Shopify help pages (exporting orders); Google's contribution policy (review incentives and selective solicitation); Mailchimp's pricing page (Free plan limits). All sales counts, prices, costs and message results are illustrative figures for an invented shop, not benchmarks.