Generative AI gets a lot of attention, and most of it is about what it might do someday. For a small or mid-sized business the better question is narrower: what job could it take off your team's plate this month? This article covers where it genuinely helps today, where it can go wrong, and how to begin without a big project.
Start with a job, not a tool
The most common mistake is buying or building "an AI" and then looking for something for it to do. The projects that work start the other way round: pick one repetitive task that eats time, and ask whether AI can do part of it.
Good first candidates share three traits. The task is repeated often, the input is text or documents, and a person can quickly check the result.
Four jobs it does well today
1. Answering questions from your own content. An assistant that reads your product details, policies and FAQs can answer routine questions from customers or staff at any hour. Because it draws on your content instead of guessing, answers stay on topic.
2. Reading and organising documents. Invoices, statements, forms and reports contain data that someone currently re-types. AI can pull out the fields, classify the document and hand the result to your accounting or ERP system for a person to confirm.
3. Drafting first versions. Replies to enquiries, product descriptions, summaries of long documents and report text can all start as an AI draft that a person edits. This saves the blank-page time, not the judgement.
4. Searching your data in plain language. Instead of building a report, someone asks "which products ran low last month?" and gets an answer drawn from your records, provided the AI is connected to your data properly.
What can go wrong
It helps to plan for these from the start rather than discover them after launch:
- Confident wrong answers. AI can state something false in a convincing way. Keep a person in the loop for anything that matters, and let the assistant say "I don't know" when it lacks the information.
- Private data. Think about what information is sent to a model and who can see it. Limit it to what the task needs, and keep access controls on who can use the tool.
- Cost creep. Usage-based pricing is cheap at small volume and can climb when a tool is used heavily. Set limits and watch usage in the first weeks.
- A tool nobody uses. If it doesn't fit how your team already works, it will be ignored. Build it into the process people already follow.
How to start small
- Pick one process where your team spends clear, repeated time.
- Write down how long it takes today, so you can tell afterwards whether it helped.
- Pilot with real examples, with a person checking every result at first.
- Expand only when it proves itself, then move on to the next job.
A focused pilot can often be built in a few weeks. A sprawling plan to "add AI everywhere" usually stalls.
Which model should you use?
There are several strong model families, including OpenAI, Anthropic's Claude and Google's Gemini. They differ in quality on different tasks, speed, price and data-handling options. The right choice depends on the job and your privacy needs, so it is better to pick per task than to commit to one vendor up front.
Where Rabtoise fits
We build GenAI solutions for businesses, including assistants trained on your content, document processing and automation, and AI features added to existing software such as an ERP or website. We start with one clear job and a fixed scope, so you know what you are getting before work begins.
If you have a task in mind, tell us about it. We'll say honestly whether AI is a good fit.
Talk to us about your project
Tell us how your business runs and we'll give you an honest recommendation, with a fixed-scope quote before any work starts.
Get in touch
