An event proposal still takes three days, with AI

The clearest AI failure at Charcuterie by Bohemian Picnics is the event proposal. For a wedding or a large event, Sara Rojas still spends two to three days building one, and ChatGPT has not reduced that. It drafts the copy. It cannot source the items or assemble the document, and those are the two things the days are actually spent on.

What a proposal contains

Clients commissioning anything elaborate want to see how it will look: the backdrops, the table settings, the flowers, and any add-on the client has asked for. When they ask for something she does not stock — she gives a sofa as the example — the work begins.

• Go through multiple rental suppliers looking for the item
• Find the flower arrangements to match
• Move into Canva and lay the whole thing out
• Make it look good enough to send to someone spending money
• Wait for the other vendors to come back with their prices before any of it can be finalised

She calls the Canva stage what it is. "It's a lot of manual, and then I want to make it look pretty." Galina Fendikevich's summary during the interview — that it is manual — she agrees with immediately.

Where exactly AI stops

Her own account of the boundary is precise, and it is the most useful sentence in this part of the interview:

"ChatGPT does help me with kind of like streamlining of what I should write in the proposal, but it doesn't really put things together."

Copy is solved. Retrieval is not — it cannot look across her rental suppliers and find a sofa at a price. Assembly is not — it cannot produce the laid-out document. So the model contributes to the fastest part of the job and nothing to the slowest.

This is the same operator whose inbox went from ten minutes to one and whose flower costing halved. The difference is not competence with the tool. It is that inquiries and costing are reasoning over information she already has, and a proposal requires fetching prices from outside and producing a designed artefact.

The two or three days, itemised

It is worth separating the components, because they have different fixes.

Vendor latency. Part of the elapsed time is not her working at all — it is waiting for suppliers to send prices. No drafting tool touches this. It shortens only if the pricing is queryable rather than requested by email.

Sourcing. Searching several rental catalogues for one item is retrieval across sources that do not share a format. This is the part she would most obviously hand to software.

Layout. Getting it into Canva and making it presentable is the part she describes as hours. Canva itself is not the problem; moving content into it by hand is.

What she asked for instead

Her description of the missing product is a straightforward composition of three tools she already uses: ChatGPT for the language, Canva for the layout, Pinterest for the visual sourcing. Galina restates it back to her as exactly that, and she agrees — "yes, and putting everything together in one."

Pushed further, she adds the detail that matters to a builder: she wants it to stream the proposal as it reads the inquiry. Not a blank template to fill, but a document that starts assembling itself from the email that triggered it.

She estimates the saving at the full two or three days, minus whatever the vendors take to reply. For a business where the owner is also the finance department and the client contact, that is not an efficiency. It is the difference between how many events a year she can quote for.

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.