Thousands of flowers, no photographs, no live stock

The data an event planner needs to price flowers does not exist in a usable form. Sara Rojas's wholesaler lists thousands of entries on a website that shows what can be sourced rather than what is on the floor, many without a photograph, at prices that move with the season and spike around holidays. She has assembled a workaround in ChatGPT. The underlying feed is the missing product.

What the current systems give her

Her description of the wholesaler's site is the core of it. "They have like a website where you go into their system and kind of see everything that they have — but it doesn't mean that it is in stock at the moment. It's just that's something they can do."

A catalogue of capability, in other words, not inventory. Which means the site cannot answer the only question that matters when she is quoting: can I get this, this week, at this price.

Three further gaps follow from her account:

• No images. "Because I don't know the name, like it's just the name. There's sometimes no picture to it. So then I have to Google it and see how it is. So it's time consuming." She came into the business without a floristry background; a name alone is not information.
• Lead times are invisible. Tropical flowers are available, but she needs to know a week or two in advance for them to be shipped, and they cost considerably more. Neither fact is surfaced at the point of choosing.
• Scale defeats memory. Asked how anyone is supposed to remember a thousand flowers, her answer is that there is no way. A florist of twenty years might; she is not one.

The pricing volatility

This is where the missing data costs money rather than time. Flower prices move by season and jump around holidays. She names Valentine's Day, when everything is hiked up, and describes the mistake plainly:

"Before I used to still quote them with the prices that weren't on Valentine's Day, and I'm like, wait, this whole like two weeks all the flowers are really expensive. I should have quoted higher."

She knows to check now. But knowing to check is a habit built out of losses, and it is exactly the kind of thing software should be telling her instead.

The stakes are structural, not incidental. She identifies flowers as the biggest recurring investment in event planning: equipment costs at the start and pays off across enough events, while flowers are bought again for every single job.

What she is doing instead

Her workaround is to supply the knowledge herself. She screenshots an arrangement she likes and sends it to ChatGPT with the colour theme and budget, and gets back stem counts, what is in season in her area, and substitutions that fit the money. She says it halves the time.

That works because the reasoning is easy. It is unreliable because the model has no connection to her wholesaler's floor. It is guessing at seasonality in the abstract while the actual stock sits behind a login two miles away.

The product, stated plainly

Galina Fendikevich summarises it on the record and Sara agrees: the inventory is available across these wholesalers, it just needs to reach her in a form where she can see both price and availability. Sara's addition is the third field — "pricing, yes, and availability, and how" — plus, repeatedly, a picture.

So the requirement set is short:

• Live availability, not catalogue capability
• Current price, with holiday and seasonal movement visible
• Lead time where a stem has to be shipped
• A photograph against every entry
• Substitution by look and colour within a budget

The last item is the only one an AI model is needed for, and it is the one she has already solved herself. The other four are integration work. That is the whole finding: the intelligence is cheap and available, and the data it needs to be useful is locked in wholesaler systems that were never built to be queried.

Where this comes from

S2E2: The secret AI strategy this small NYC business uses to save thousands — the full interview with Sara Rojas and Luisa Ramirez. Listen or watch: YouTube, Spotify or Apple Podcasts.