Florists prepare for Valentine's Day and Mother's Day with AI by starting eight weeks out: forecast orders from last year's sales, cut the range to a short pre-order menu, turn it into a stem order for the wholesaler, pre-write replies and cut-off messages, and plan delivery zones and routes. On the day, AI drafts replies while the team makes.
The reason to start that early is that the decisions with the most money in them, how many stems to buy and when to stop taking orders, are made weeks before the phones get busy. AI is most useful doing the arithmetic and writing in January or early spring, so that nobody is working out rose quantities at midnight in peak week.
How the two peaks differ
Treat them as two different projects. The same plan applied to both goes wrong in predictable ways.
| Valentine's Day | Mother's Day | |
|---|---|---|
| Date | Fixed: 14 February | Moves each year, and falls on different dates in different countries; check yours |
| What sells | Roses, above all red, from single stems to dozens | Mixed seasonal bouquets, pastels, plants and hampers |
| When orders arrive | Late: a large share in the last 72 hours | Earlier, with more pre-orders a week or two ahead |
| Deliveries | Crammed into one day, often to workplaces | Spread over the days before, often to homes |
| Wholesale pressure | Rose prices typically climb in the run-up | Wider flower mix gives more room to substitute |
That last row shapes your menu. At Valentine's, a design that depends on one variety of red rose is exposed to price and supply; at Mother's Day, a "florist's choice" option lets you use what's good that week.
Eight weeks out: forecast from last year's orders
Export last year's orders for the two weeks around the peak from your shop system or website, one row per order with date ordered, date delivered, product, price, delivery or collection, and delivery area. Remove customer names and addresses first. Then ask a chat assistant:
Attached: last year's orders from 1 to 15 February.
1. Count orders by product and by delivery date.
2. Show what share of all orders were placed in the last 3 days.
3. Give average order value, and the share delivered versus collected.
4. Forecast this year's orders by product, assuming overall growth of
10%, and show the working. Flag any product with fewer than 10
orders as too small to forecast.An excerpt of the output for an illustrative two-person florist (illustrative):
Total orders: 212. Placed 12-14 Feb: 58% of orders.
Deliveries on 14 Feb: 131 (62%). Collections: 41 (19%).
Average order value: $74.
Forecast at +10%: Dozen red roses 64 -> 70; Mixed luxe bouquet 48 -> 53;
Single rose wrap 37 -> 41; Hand-tied, florist's choice 29 -> 32...Mostly right, but it missed something a florist would catch: last year's 14 February fell on a weekday, so many deliveries went to workplaces; if this year's falls on a weekend, more go to homes, deliveries spread more thinly across your area, and some customers shift to collection. Tell the assistant which day the peak falls on this year and ask it to adjust the delivery and collection split. Treat the 10% growth figure as your assumption, not the AI's; base it on how the rest of your year is running.
Six weeks out: a short pre-order menu
Peak days are won by making fewer designs faster. Cut the range to five or six options with fixed recipes, so the team can make them on a production line and the stem order is predictable. A filled-in menu for Valentine's:
| Design | Price | Recipe |
|---|---|---|
| Single red rose, wrapped | $15 | 1 red rose, 2 eucalyptus, wrap, ribbon |
| Six red roses | $55 | 6 red roses, 4 eucalyptus, 3 ruscus |
| Dozen red roses | $95 | 12 red roses, 6 eucalyptus, 5 ruscus |
| Luxe mixed bouquet | $85 | 5 roses (mixed), 3 ranunculus, 3 tulips, 5 foliage stems |
| Florist's choice | $60 / $90 | Seasonal; built on the day from best stock |
A chat assistant is handy for writing the menu descriptions in your voice, but give it the recipe and tell it not to promise specific flowers in the florist's choice design. "Seasonal blooms chosen by our florists on the day" is honest; "roses, peonies and lisianthus" in February may not be.
A prompt that works: "Write a 20-word description for each design below, warm but plain. Use only the stems listed in each recipe. For florist's choice, name no flowers. No adjectives about quality you can't check." A typical first pass (illustrative) still needs editing:
Dozen red roses: Twelve premium long-stem red roses, hand-tied with
eucalyptus and ruscus, presented in our signature gift box.
Luxe mixed bouquet: A romantic mix of roses, ranunculus and tulips
with lush seasonal foliage, finished with a satin bow.Two things to fix. "Long-stem" and "premium" describe a grade you may not have bought (if your wholesaler sends 50cm stems, a customer expecting 70cm will notice), and "signature gift box" and "satin bow" are packaging the assistant made up. Delete anything that isn't in the recipe or on your packing bench. The luxe line is fine once "satin bow" becomes whatever ribbon you actually use.
Five weeks out: turn the menu into a stem order
This is pure arithmetic, and it is where AI saves the most stress: forecast orders for each design multiplied by its recipe, plus a buffer for breakage, blown heads and the walk-ins you didn't forecast. Taking the forecast above and a 15% buffer:
| Stem | From forecast | With 15% buffer | Order (in bunches of 20) |
|---|---|---|---|
| Red roses | 70 dozen x 12 + 41 singles + 20 six-stem x 6 = 1,001 | 1,151 | 58 bunches |
| Eucalyptus | 70 x 6 + 41 x 2 + 20 x 4 = 582 | 669 | 34 bunches |
Ask the assistant to produce this table for every stem and show each line's calculation, then check the two most expensive lines by hand. In testing, the most common error is a design missing from one line, such as the six-stem roses left out of the eucalyptus total, which is why showing the working matters. Place provisional orders with your wholesaler early and confirm the final quantities once you see pre-orders at three weeks. If the peak spend strains your cash, a short forecast using your sales history helps you see it coming.
Reading pre-orders at three weeks takes one comparison and has one trap. Ask the assistant how many orders last year had been placed by the same date, then compare. Suppose last year's export shows 30 orders placed by 24 January, 14% of the final 212, and this year you have 45 by the same date. The naive reading is 50% growth and about 318 orders. But if you opened pre-orders a week earlier this year, much of that gap is the same customers ordering sooner. A safer rule is to trust the pace only when it holds at two checkpoints (three weeks and ten days out), and even then move the forecast by half the difference: from 233 to about 275, not to 318. Stems you over-order at peak prices are money in the bin by 16 February.
Four weeks out: open pre-orders and nudge once
Pre-orders are the single best defence against peak-day chaos, because every one moved earlier is a design you can plan and a delivery you can route in advance. Open them four weeks out for Valentine's and five for Mother's Day, with a reason to order early that isn't a discount: a guaranteed delivery slot, or first pick of the luxe designs before they sell out.
The announcement is a good job for a chat assistant, with your menu, prices and cut-off pasted in. The difference between a first draft and the version worth sending is usually specificity. A typical first draft opens with "Love is in the air! Make this Valentine's unforgettable with our stunning blooms." The edited version opens with the facts customers act on: "Valentine's pre-orders are open. Delivery slots for the 14th usually sell out by the 11th, so book now to be sure of a morning or afternoon delivery." Send one reminder to your customer list about ten days before the cut-off, and no more. Two emails about the same peak are useful; five are noise.
Three weeks out: replies, cut-offs and card messages
Most peak-week messages are the same ten questions. Write the answers now, calmly, with a chat assistant, and save them as quick replies in whatever inbox you use. The three that matter most:
CUT-OFF (post from 8 Feb, pin on the website)
Valentine's delivery orders close at 5pm on 12 February or when we're
fully booked, whichever comes first. Collection orders are open until
noon on the 14th. We'll post here the moment delivery slots sell out.
SOLD OUT (reply template)
Thank you for thinking of us. Delivery for the 14th is now full, but
we still have collection slots until noon, or we can deliver on the
15th with a card that says so. Would either work?
SUBSTITUTION (on the order form and confirmation email)
If a flower in your design isn't at its best on the day, we'll replace
it with one of equal or higher value in the same colour.Card messages need a strict rule of their own. If you use AI to tidy order notes, it must never change the words of a card message. A realistic slip: an assistant "correcting" an unusual spelling of a name, or turning a private joke into standard English, and a customer's message printed wrong on the most personal gift of their year. Paste card messages into production sheets exactly as written; let AI flag only what might be a typo, for a person to check with the customer. For the order handling around this, using AI for florist orders, reviews and social posts covers the everyday version.
Delivery week: zones, slots and routes
Split your delivery area into zones, give each zone a morning or afternoon slot, and cap slots per zone at what one driver can do. Then let a route planner sequence each run. Routific, for example, prices by orders per month rather than by vehicle: up to 100 orders a month is free, and 101 to 1,000 is $150 flat for that month. Every plan includes route optimisation, driver apps, real-time tracking, customer email notifications and proof of delivery; text-message notifications are an add-on. A florist doing 131 deliveries in February and 40 in a normal month would pay only in the peak months.
Set the caps from your own drivers' pace, not a guess. Time a few normal-week runs, then fill in a plan like this one (illustrative figures for a shop with two drivers):
| Zone | Distance from shop | Drops per hour | Slot | Cap per run |
|---|---|---|---|---|
| A: centre | Up to 10 minutes' drive | 8 | Morning, 8am-noon | 28 |
| B: inner suburbs | 10-20 minutes | 6 | Morning, 8am-noon | 21 |
| C: outer suburbs | 20-30 minutes | 5 | Afternoon, 1pm-5pm | 17 |
| D: villages | 30+ minutes | 3 | Afternoon, 1pm-5pm | 10 |
The caps allow half an hour per run for loading and the first drive out. Add them up and the problem shows itself: two drivers, four runs, 76 drops, against a forecast of 131 deliveries on the 14th. The gap of 55 has to go somewhere, and there are only three places: a third driver, deliveries on the 13th for customers who accept them, or more collection slots. Knowing that in January is the whole point; discovering it on the morning of the 14th means late roses and refunds.
Proof of delivery, a photo at the door, is worth its weight at Valentine's, when "it never arrived" messages are most common and most emotional. For more on sequencing runs with a small fleet, see how a small courier firm plans routes with AI.
On the day, and the week after
On the day, one person owns the inbox and phone, working from the saved replies, while everyone else makes. AI drafting is useful for the unusual messages, a late change of address or a delivery complaint, but a person sends everything.
The complaint that arrives mid-afternoon shows why. A customer writes: "It's 3pm and my partner still hasn't got her flowers. I paid for a MORNING delivery." Pasted into a chat assistant with no other context, the draft comes back as something like "We're so sorry for the inconvenience! Your order is on its way and will be with you shortly" (illustrative). That sentence is a guess, and if the order was actually delivered at 11:40 to a reception desk, it makes things worse. Check the driver app first, then give the assistant the facts to work with. The reply worth sending reads: "I've checked with our driver: your roses were delivered at 11:40 and signed for at the front desk of [building]. Here is the delivery photo. If they haven't reached her desk, reception should have them." Only if the order really is late do you apologise, give a time you can keep, and offer something concrete.
Phone orders need a fixed script, because the caller is often in a hurry and the details are easy to get wrong. Keep this checklist by the phone and fill it in the same order every time: design and price; recipient's full name; delivery address with a landmark or buzzer number; zone and slot; card message, written word for word and read back to the caller; sender's name and phone number; payment taken. If you take phone orders into your system later, an AI assistant can turn a quick voice memo of the call into this format, but the card message is typed from the read-back, never from the transcript.
A one-person studio florist working from home needs a smaller, stricter version of everything above. A sole florist can't make and answer the phone at the same time, so the plan becomes: pre-orders only, no same-day orders, a cap on total orders set by how many designs they can make in two long days, and collection slots where possible. The AI jobs are the same (forecast, stem maths, replies written in advance) but the most valuable one is the sold-out message, sent automatically the moment the cap is reached, so the florist isn't turning people away one by one while trying to work.
The week after, capture what happened while it's fresh, because next year's forecast depends on it. Paste your notes and this year's order export into a chat assistant and ask for a one-page debrief: orders by design against forecast, stems ordered against used, when slots sold out, and the five most common customer questions. Filed with the order export, that page is next year's starting point. Check its conclusions before you file it, though. A debrief that says "Red roses: 1,160 ordered, 1,020 sold in designs, 12% waste; cut the buffer to 5%" (illustrative) has probably missed the stems that went into florist's choice bouquets and walk-in wraps, which your system records as products, not stems. Count what was left in the buckets on the 15th instead; that is the real waste figure, and it may well show the 15% buffer was about right. It is also the moment to ask for reviews, from every peak customer rather than only the ones you expect to be pleased, since Google's rules bar picking and choosing, and automated review requests make that a single scheduled message rather than a job nobody gets round to.
Mother's Day follows the same countdown with two changes: base the forecast on the weekend before the date, since more orders are placed earlier and delivered on several days, and plan more collection slots, because many customers pick up a bouquet on the way to a family lunch.
Further reads
- Can AI Mock Up a Flower Arrangement Before You Make It? — Showing customers a design before you make it.
- How to Set Up AI Abandoned Cart Emails for a Small Online Store — Recovering pre-orders people started but didn't finish online.
- Should a Small Business Let AI Answer Customer Messages? — Deciding which peak-week messages AI may answer on its own.
- How to Build a 13-Week Cash Flow Forecast With AI Help — Planning cash for the big wholesale spend before each peak.
- Can AI Run Your Social Media on Autopilot? What Breaks — What breaks if you let AI run your peak-season posts alone.
- A Simple AI Risk Register for Small Businesses (With Template) — A one-table AI risk register with a scoring scale, a template to copy, twelve filled-in rows from a florist and the triggers for updating it.
- How to Write Wedding Flower Proposals With AI — A florist's method for AI-assisted wedding proposals: consultation brief, recipe costing, prompts for the words, and the seasonality check AI can't do.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Routific pricing page (per-order pricing, features included on all plans).