AI in your business, starting from one specific job

Where to start

Starting from "let's use AI" usually ends in a demo nobody opens a month later. It's better to start from a job a person does today, that repeats every day and that nobody enjoys. For example:

• orders that arrive written in emails and get retyped into the business software;
• customer questions on the website, always answered with the same information;
• supplier PDFs to read line by line;
• the draft of a reply or a quote, checked before it goes out.

What it's good at, and what it isn't

A model like Claude or GPT reads text written by people well: it understands a badly written order, sums up a long email, finds the answer in a hundred-page manual. It can work in your business software if I give it the tools to.

It gets arithmetic wrong and makes things up confidently when it doesn't know. So calculations and exact rules stay in ordinary code, which costs less and always gives the same result. And the actions that matter, like sending a quote to a customer, go through a person who confirms them.

Where the data goes

With a cloud model, the text to read goes to the provider through its API, and with the main providers it isn't used for training. Everything else, your business software and database, stays where it is today.

If data can't leave the company, an open model runs on your own server. It costs more up front and makes more mistakes, but nothing leaves your network. I tell you which is better after a trial on your own documents.

How the work goes

I have already done AI projects, both for reading PDFs and with agents that work inside business software. We start like this:

1. We look at the job to take away, with real examples. The first analysis is free.
2. I run a trial on a batch of past cases and we count how many it gets right.
3. If the numbers hold up, I connect it to your business software, with a person confirming the doubtful cases.

One worked example: extracting data from PDFs with AI. For other kinds of automation, see Integrations and automation.

Frequently asked questions

What is an AI agent?

A program where a model like Claude or GPT decides the steps and uses tools I build for it: look up a customer in your system, read an order, write a draft. It can only do what those tools allow.

Will our data be used to train the AI?

Through the main providers' APIs, no: data sent via the API isn't used to train their models. If data must not leave the company, an open model runs on your own server.

Can we have a private AI, just for us?

Yes. An open model installed on your own server keeps everything in-house, but it needs a machine with a suitable graphics card, and smaller models make more mistakes. Before choosing, we try both on your own cases.

What does running a model cost?

Cloud models are paid by usage, based on the text read and written. Before starting I estimate the cost from your numbers, with a trial on real documents.

Read more

• The service: Integrations and automation
• More pages like this: Connections, automation and AI