An LLM in a Paid Workflow: Tickets, Emails, and PDFs Become a Step a Human Can Audit

In a company, an LLM is not a chatbot you open when you are stuck. It is a model that reads unstructured language and emits something the rest of the stack can use: a draft, a JSON row, a route, a yes/no with a reason. If that output does not enter a workflow someone already pays for, you have a toy. The popular question is “what is an LLM in business.” The useful answer is that sentence - not a lecture on neural nets.

Cost tables and free tiers live in a separate article. Who sells the tokens is the provider article. Here: what “in business” means on a Tuesday, and what it does not.

1. What it is not

Most failed “AI projects” started with the wrong object.

• Not a search engine. It does not owe you a source. If you need a fact from your catalog, you retrieve then generate - or you invent prices.
• Not a junior hire. It has no memory of your company unless you give it one (files, RAG, tools). It will sound sure either way.
• Not a CRM, not an ERP, not a till. Those systems stay. The model sits beside them and writes or reads a field.

2. Three places it actually sits

Same model family. Different product. Mixing them is how you buy twenty seats and still paste into a chat.

• Coworker tool: a person in ChatGPT or Claude drafts a mail, a brief, a summary. Cheap to start. The company does not own the loop. Fine until the work has to happen without that person.
• Inside the product: a bot on the site, a help panel, a form that adapts. The customer never sees a model name. You own latency, cost, and the lie it might tell.
• In a pipeline: overnight, a batch of invoices or tickets becomes rows. No chat UI. This is where hours compress if the schema is strict and a human still sees the rejects.

3. When it is in the business, not next to it

A test that survives a board slide: name the step that already costs money. Tickets. Proposals. Invoice fields. First-line answers at 1 a.m. If you cannot name the step, you are buying seats for curiosity. Curiosity is allowed. It is not an LLM in business yet.

Conclusion: language in, a step out

An LLM in business is a model sitting on a real step: draft, extract, route, first answer - with retrieval if you need facts, and a person on the rejects. For who sells the tokens, read the provider piece. For what it costs, the comparison with free plans. If you want that step in a Next.js app or a bot, write via the form with the workflow that already hurts. We will not start with a slide named AI.