Clothing boutiques use AI to spot trends in two ways: reading outside signals, such as search interest in Google Trends, saves in Pinterest Trends and hashtags in TikTok's Creative Center, and analysing their own till data for sell-through and size patterns. The best buys come from checking one against the other, then ordering trend pieces at test depth first.
The order matters because a trend that's rising online may already be in the chain stores by the time a boutique's delivery arrives. A boutique's real advantage is knowing its own customers: who comes in, what they reorder, which sizes run out in week two. AI can't forecast your customer for you, but it can make your own sales data readable in an evening and stop you buying a two-week micro-trend 40 units deep.
Two kinds of signal: the internet's and your till's
Outside signals tell you what people in general are searching for, saving and posting. They're early and broad, and they over-represent younger, very online shoppers. Your till tells you what your customers actually paid for, at what price and in which sizes. It's late, because it only knows what you stocked, but it's specific and it's yours.
A useful rule: an outside signal is a reason to test; your own sell-through is a reason to buy deep. If a trend shows up online but your till has nothing like it, buy a small test. If your till shows a shape or colour already selling at full price and the outside signals agree, that's where to put more money.
The free trend sources and what each tells you
| Source | What it shows | Best for | Limits |
|---|---|---|---|
| Google Trends | Relative search interest over time; compare up to five terms | "Is barrel-leg jeans interest still rising or flattening?" | Relative, not volumes; noisy for niche terms |
| Pinterest Trends | What people search and save on Pinterest, over time; Pinterest also publishes an annual Predicts report each December | Planning a season ahead: colours, occasion wear, styling | Skews towards planners and inspiration, not immediate purchase |
| TikTok Creative Center | Trending hashtags, songs and creators, with growth over 7 and 30 days | Spotting fast-moving items and the language people use for them | Very fast cycles; young audience; built for advertisers |
| AI research assistants (ChatGPT, Perplexity and others) | Summaries of trend coverage from fashion press and retailers, with citations when asked | A quick brief before a buying appointment | Can state things confidently with weak or no sources |
| Specialist forecasting (Heuritech, WGSN and similar) | Image recognition across social media and catwalk data, with forecasts by attribute | Brands planning ranges far ahead | Sold to brands and larger retailers on custom quotes |
For most boutiques the first three are enough, and they're free. Perplexity Pro, at $20 a month, or a paid ChatGPT plan makes the research-brief step faster, but it isn't required.
Asking an AI assistant for a trend brief, then checking it
Before a buying appointment or a trade show, an AI brief can collect what's being written about a season in ten minutes. The key is to ask for sources and to treat every claim as a lead to check, not a fact. An illustrative prompt:
I run a womenswear boutique. Our customers are mostly 35-60, buy
smart-casual and occasion wear, and spend $60-$180 per item.
Summarise the main womenswear trends being reported for autumn/winter
in: colours, silhouettes, fabrics, and 3 specific items.
For each trend, give 2 sources with publication dates.
Mark anything aimed mainly at under-25s.
Say "no good source found" rather than guessing.An illustrative reply, trimmed:
Colours: burgundy and deep reds; chocolate brown; butter yellow
continuing from summer. Sources: [fashion trade site, Aug];
[department store trend report, Sep].
Silhouettes: longer skirts (midi/maxi), relaxed tailoring, barrel-leg
trousers. Sources: [magazine, Jul]; [retail analyst, Aug].
Fabrics: suede and faux suede, textured knits. Sources: [trade site].
Items: suede jacket; wide-leg tailored trouser; fisherman-style knit.
Mainly under-25: "micro shorts with tights" (TikTok-driven).
No good source found: "the return of peplum" (one fashion website only).What you'd do next: check each source link actually exists and says what the summary claims (AI sometimes attributes a trend to the wrong publication), then put the five or six terms into Google Trends to see whether interest is rising, flat or already falling. "One fashion website only" is a useful warning the prompt asked for; without that instruction, the assistant would probably have listed peplum alongside the rest. Keep the brief to one page and take it to the appointment as questions for the rep, not as a shopping list.
A realistic example of why the source check matters: the "department store trend report, Sep" in a brief like this turns out, when the owner clicks through, to be the previous year's autumn report. Butter yellow was that year's colour; this year it's barely mentioned. Nothing in the summary looked wrong. The date on the page was the only clue, which is why the prompt asks for publication dates and why someone opens every link.
The Google Trends step then sorts the leads. An illustrative check of four terms over five years, set to the region you trade in:
| Term | Pattern over five years | What it means for the buy |
|---|---|---|
| burgundy jumper | Rises every autumn, and this year's rise started earlier and is higher than last year's | Genuine trend on top of a seasonal pattern: trend-led tier |
| suede jacket | Same autumn bump every year, no bigger this year | Seasonal, not a trend: buy at normal depth if it's already a line you sell |
| barrel leg jeans | Climbed steeply for two years, flat for the last six months | Maturing: small test in a moderated shape, not a big bet |
| fisherman jumper | Low and noisy | Too niche to read; rely on the till and the rep |
The suede jacket row is the easy mistake. Looked at over twelve months, any autumn item looks like it's trending in September. Always widen the range to five years before calling something a trend.
Reading your own sell-through with AI
This is where AI is most reliable for a boutique, because the data is yours and the maths is simple. Four measures matter:
- Sell-through rate: units sold divided by units received, over a period, as a percentage. A style that sold 18 of 24 units in eight weeks has 75% sell-through.
- Full-price sell-through: the same, counting only sales before any markdown. This is the one that shows real demand.
- Weeks of cover: stock on hand divided by average weekly sales. It tells you how long current stock will last.
- Size curve: the share of sales by size. Most boutiques find theirs differs from the supplier's standard size pack.
Export last season's sales and receipts by style, colour and size, remove any customer details, and ask:
Attached: last autumn/winter's receipts and sales by style, colour
and size (columns: style, category, colour, size, units_received,
units_sold_full_price, units_sold_markdown, first_delivery_date).
1. Calculate full-price sell-through for each style and each category.
2. Show the top 10 and bottom 10 styles by full-price sell-through.
3. Show our size curve by category (share of units sold by size).
4. List styles where one size sold out while others were marked down.
Show the formulas you used.An illustrative, filled-in slice of what comes back:
| Category | Units received | Full-price sold | Full-price sell-through | Size curve (8/10/12/14/16) |
|---|---|---|---|---|
| Knitwear | 210 | 147 | 70% | 10% / 24% / 30% / 24% / 12% |
| Midi skirts | 96 | 71 | 74% | 8% / 22% / 31% / 26% / 13% |
| Tailored trousers | 120 | 54 | 45% | 14% / 30% / 28% / 18% / 10% |
| Statement jackets | 48 | 17 | 35% | 9% / 21% / 33% / 25% / 12% |
Two things jump out of that illustrative slice. Midi skirts and knitwear sold strongly at full price, and the reported trend towards longer skirts agrees, so that's a category to buy with confidence. Tailored trousers are reported as a trend but sold weakly for this boutique; the next question is whether the cut, the price or the sizes were wrong, which is a conversation with the supplier, not a reason to buy deeper. And the size curve peaks at 12 and 14, while many supplier packs assume a peak at 10: fixing the size mix alone can lift sell-through without changing a single style.
A quick sum shows how much the size mix costs. Say a dress comes in supplier packs of 12 (sizes 8/10/12/14/16 as 2/4/3/2/1), and the boutique orders two packs: 4/8/6/4/2. Its own curve for dresses, applied to 24 units, would be 2/5/8/6/3. So it receives three extra 10s and two extra 8s, and is short of two 12s, two 14s and a 16. The 12s and 14s sell out in the third week, a customer who wanted a 14 buys elsewhere, and five small sizes sit on the rail until the markdown. At an illustrative $38 cost each, that's $190 of stock waiting to be sold at a discount, on one style. Across 30 styles, the size split is worth a firm conversation with every supplier. What to upload when AI analyses a sales spreadsheet covers checking the assistant's maths.
From signal to order: a buy plan with test depth
Consider an illustrative boutique with an open-to-buy (the budget for new stock at cost) of $24,000 for autumn/winter. The owner splits it three ways:
- Proven core, 65% ($15,600). Categories and styles with strong full-price sell-through last year: knitwear, midi skirts, the best-selling dress shapes. Bought in the boutique's own size curve, not the supplier's standard pack.
- Trend-led with support from both signals, 25% ($6,000). Trends that show up in outside signals and have a relative in the till data. Burgundy knitwear and a longer suede-look skirt fit here, because the boutique's customers already buy knits and midi skirts.
- Tests, 10% ($2,400). Trends with outside signals only, such as a barrel-leg trouser. Bought at 6 to 12 units per style, across the core sizes only, with a reorder option agreed with the supplier if possible.
Then set the triggers before the stock arrives. A test style that reaches 50% full-price sell-through in its first four weeks gets reordered if the supplier can deliver within the season. One under 20% after four weeks moves to the front of the shop, gets styled in an outfit post, and is marked down on a set date if it still doesn't move. Writing the triggers down beforehand stops the common habit of reordering whatever the owner personally likes.
Four weeks after delivery, the illustrative boutique's three test styles read like this:
| Test style | Units | Sold at full price | Sell-through | Action |
|---|---|---|---|---|
| Barrel-leg trouser, dark denim | 10 | 6 | 60% | Reorder 8 in sizes 12 and 14, delivery in three weeks |
| Wide-leg cord trouser | 12 | 4 | 33% | Between triggers: hold, review at week six |
| Bubble-hem skirt | 8 | 1 | 13% | Front table, outfit post this week, 30% off from week eight if still slow |
The barrel-leg result is the interesting one. Online interest had flattened, which suggested a small test, yet for this boutique's customers the dark-denim version sold at full price. That's the till overruling the outside signal, and it's exactly why the test tier exists. Ask a chat assistant to calculate the table from a till export if you like, but check the "sold at full price" column against the markdown dates, because an export that doesn't separate markdown sales will flatter every style.
The reorder request is a short email, and AI drafts it well if you're specific. Give it the style, the units, the sizes and the date you need them. In an illustrative draft, the assistant added "we'd be happy to commit to a larger order for spring if you can prioritise this". Delete lines like that unless you mean them: a rep will remember the promise at the next appointment.
When your regulars are over 50, trends arrive later and softer
Trend timing depends heavily on who your customers are. A boutique whose regulars are mostly over 50 will usually see online trends reach its rails a season or more later, and in a moderated form: a wider-leg trouser in a familiar fabric rather than an extreme barrel shape. For that boutique, TikTok is almost irrelevant as a buying signal and useful mainly for language and styling ideas. Pinterest and the boutique's own till are the sources to trust, and the tests should be the moderated version of the trend, not the version on social media.
A menswear boutique sees something similar in reverse: fewer, slower trends, where the signal is often in fabric and colour rather than shape, and where repeat purchases of a proven trouser or overshirt matter more than novelty.
How boutiques get trend buying wrong with AI
- Buying a micro-trend deep. A realistic version: an item goes viral on TikTok in August, the owner orders 40 units for delivery in October, and by then the chain stores have had it for six weeks. Twelve sell at full price; the rest are marked down twice. A 10-unit test would have told the same story at a quarter of the cost.
- Treating the AI brief as data. A trend summary is a set of claims from articles, some of them written to sell. It becomes evidence only when search interest and your own sales support it.
- Ignoring sizes. The right style in the wrong size mix sells through badly and looks like a failed trend.
- Letting AI describe your customer. Ask an assistant who shops at a boutique like yours and it will give a plausible stereotype. Your till and your staff know better.
- Forecasting without enough history. A single season of data is thin, especially if you changed suppliers. Why AI stock forecasts go wrong lists the other traps.
A pre-order check for each trend line
Before you sign an order for any trend-led style, run it through five questions. An illustrative filled-in check for a burgundy fisherman-style jumper:
- Outside signal? Yes: burgundy and textured knits in two dated sources; Google Trends interest for "burgundy jumper" rising since July.
- Relative in our till? Yes: knitwear 70% full-price sell-through last year; burgundy sold well in scarves.
- Customer fit? Yes: classic shape, our price band, not aimed at under-25s.
- Depth? Trend-led tier: 18 units, sized 8/10/12/14/16 as 2/4/6/4/2 from our knitwear curve.
- Trigger set? Reorder 12 if 50% sells at full price in four weeks; markdown review at week eight.
It takes a few minutes per line and it changes the conversation with suppliers, because you're asking for the size split and a reorder option based on evidence. Once the stock lands, finding slow movers with AI and creating outfit and styling posts help the trend pieces earn their place on the rail.
Further reads
- Best AI Inventory Tools for Small Retailers Compared — Inventory tools that track sell-through for you.
- How Boutiques Can Answer Customer DMs Within an Hour Using AI — Customer DMs are a trend signal too; answer them faster.
- How to Forecast Next Quarter's Sales With AI Using Your History — Forecast next season's sales from your own history.
- Are AI-Generated Customer Personas Accurate? How to Check — Why an AI description of your customer needs checking.
- AI Model Shots for Boutiques: Try-On Images Without a Photoshoot — Photograph the new stock on models without a shoot.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Pinterest Trends and Pinterest Predicts pages; TikTok Creative Center trend discovery; Heuritech product pages; Perplexity pricing from the vendors' pages.