Ten minutes to one: answering inquiries with ChatGPT

The single clearest AI win at Charcuterie by Bohemian Picnics is the inbox. Sara Rojas, who answers every customer inquiry for the shop and for the sister event business, used to spend about ten minutes on each reply. With ChatGPT she now spends a minute or less. The saving was large enough that she stopped paying an assistant she had hired the previous year specifically to help with email.

What makes this worth reporting is not the speed. It is that she still writes every reply herself, in the sense that matters to her — she reads what the client sent, she decides what to say, and she signs it. What changed is the part she describes as her own weakness.

The problem was overthinking, not typing

"I am an overthinker," she says. Left alone with a blank reply, she would rewrite the same paragraph several ways before sending it. That is where the ten minutes went. It was not the mechanics of composing an email. It was the loop of second-guessing a sentence that was already fine.

Her workflow now is short. The client's message goes into ChatGPT. It comes back with a draft. She tweaks it and sends. The loop is closed by the model producing something concrete to react to, which is a much faster decision than producing something from nothing.

A year of use is doing the real work

She has been using ChatGPT for roughly a year, and she is specific about what that time bought her. The model has learned the difference between her two businesses. Charcuterie by Bohemian Picnics is the shop off Central Park where customers build a board and rent a picnic basket. Bohemian Picnics is the event-planning company that came first. The two generate different kinds of inquiry and want different registers of reply.

"It kind of really embodies of like when I need charcuterie information and when I need Bohemian information," she says. "So it kind of replies in its way." Asked directly whether it had learned her and her businesses, she says it completely did.

That accumulation is the asset, and it is not transferable. An operator starting today gets a generic assistant. An operator a year in gets one that knows which of their businesses an email belongs to. Nothing about the product changed in between; the context did.

What it replaced

The previous year she had someone helping with email because the volume was consuming her day. She was clear that she did not want to hand the task away permanently. "I like answering the emails because I like being there from the beginning to the end and kind of getting all the information from the client." Inquiry handling is where she learns what a client actually wants, and outsourcing it cost her that.

ChatGPT let her take the work back rather than delegate it further. That is a different outcome from the one usually promised. The pitch for AI in a small business is normally headcount avoided. Here the owner reclaimed a task she had been forced to give up, which she values for reasons that have nothing to do with cost.

Where it stops working

Two limits are worth stating alongside the win, because she raised both unprompted.

• The tone drifts. She asks for friendly, gets friendly, and then at some point the model reverts to a generic register. "At some point it decides that this is too much. I'm just gonna do a generic one." She has to re-prompt, and she loses the version she liked.
• Volume still defeats it. Some mornings she wakes to around ninety-four emails. One-at-a-time drafting, even at a minute each, is still an hour and a half. What she wants is a pass over the whole inbox that she then reviews, rather than ninety-four separate conversations.

Both of those are covered separately. The point here is narrower and it holds: for a shop owner who answers her own inquiries, a general-purpose chatbot with a year of accumulated context removed a job she had been paying someone else to do, and gave her back a part of the work she wanted to keep.

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.