AI
This article is based on hands-on experience from real AI projects across different industries. It’s not a how-to guide, but insights we’ve gained by doing this work firsthand.
These days, everyone’s talking about AI. “We need something like that too.” “Our competitors are already using AI.” “I saw a demo, this would solve everything.” Sound familiar?
I’m not here to talk anyone out of AI – especially since working with it is literally my job But before diving head-first into transforming your processes with AI, there are a few things worth thinking through. I’ve collected some considerations and questions I wish someone had shared with me earlier. These aren’t rules, more like food for thought.
Why GenAI projects are “different”
We use traditional software because it does exactly the same thing, every time, with precision. But we expect GenAI to work for us, to interpret, summarise, rewrite, elaborate.
And here’s something important to clarify. All GenAI is AI, but not all AI is GenAI. GenAI is just one part of the AI universe, and even GenAI itself is an entire galaxy. Specific tasks are like solar systems within it. When someone says they want “AI,” they usually mean GenAI, but that’s not always the right answer. There’s plenty of AI that isn’t generative at all; it learns from large datasets and makes decisions. That’s what recommends your next post, what plans the route on your phone. It’s still AI, just for different purposes.
GenAI isn’t deterministic. It always responds a little differently, interprets things a little differently, weighs things a little differently. That’s not a bug, you just need to know where to use it.
Imagine if your pivot table or your monthly report showed slightly different results every time you ran it. Like a button that did something different each time. Same data, same button, different result. That’s a bug, not a feature. You’d say the software is broken. If you expected that kind of behavior from GenAI, then GenAI isn’t the solution for that task. But in its proper context? That’s a feature, not a bug.
First Things First, What’s the Problem?
Does this sound familiar? Has it come up at your company? “We need AI” or “Let’s implement AI.” The problem with this is that we’re avoiding the fact that we currently have a problem, hoping some magic box will fix it. But offline, in-house. I get it, but I have bad news.
What does a good starting point sound like? Being able to articulate exactly what hurts.
“We’re spending too much time on reports.” “We can’t find anything in our documents.” “HR keeps getting the same questions over and over.”
Trust me, I’ll save you weeks by saying this. If you can’t summarise the problem in a sentence or two yet, that’s fine, but then don’t start thinking about solutions. Sit down with your team, talk it through. Or bring in someone from outside to ask about the “we’ve always done it this way” and “we don’t talk about that” stuff. Sometimes the hardest part is simply seeing what the actual problem is.
Problem, meet your AI. AI, this is your problem.
When we say “AI solution,” many people picture a black box. Problem goes in, solution comes out. You introduce them to each other and everything gets fixed. “AI, meet the problem. Problem, meet the AI.” Done.
That’s not how it works.
With AI, every new conversation is like starting a chat with a stranger. AI is a tool, it does what you tell it. If you can’t articulate exactly what you want, AI won’t figure it out for you.
And if AI could just “solve our problems” on its own? Well, we might not actually want that. Think about what that would look like in reality, it would see where the problems are, tell us about them, and tell us what to do differently. Do you really want an AI telling you what you should be doing? Or doing things without you even knowing?
Go ahead and test it. Open a new chat and explain to the AI how your team works. What processes you have, how many people, what skills, what tasks, who you work with, who your clients are. Then ask what it thinks. Chances are it’ll tell you the team is well-structured and your processes are great. Because AI is fundamentally polite, and it works with what it’s given.
Because ultimately, AI doesn’t solve the problem. We do, possibly with AI’s help.
Do You Even Need AI?
This is worth asking before you start anything.
Sometimes a better process is enough. Sometimes a simple rule is enough, if this happens, we do that. Sometimes a properly built spreadsheet or a small automation does the trick.
AI is good where you need to interpret something. Text, context, nuances. Where you can’t predict every case in advance, but a human would “sense” the right answer. It’s excellent for brainstorming. But don’t use it for self-validation, it’s too polite for that and can do more harm than good. If you can describe what you want with rules, you probably don’t need to throw AI at it.
And here’s the thing, a finished solution is rarely just AI. Most working solutions are hybrid, there’s a human step, a rule-based part, some AI, and they alternate. Remember that non-deterministic nature we talked about earlier? That’s exactly why it’s worth using GenAI only where it’s truly necessary.
Imagine if your accounting gave you a different answer every time.
A simple example using just a prompt, you paste your email thread into the chat, type your prompt with context, check the response, make edits, paste it into your reply, send it.
With automation? The system reads the email (rule-based), runs the prompt, drafts a response, you read it, make edits, send it.
In both cases, you’re in the loop. Maybe that’s not how we pictured it, but that’s normal.
One Solution = One Task
A lot of people get this wrong.
ChatGPT does just one thing, you ask, it answers. A company chatbot does one thing, answers questions about internal policies. Copilot does one thing, helps with your documents. Each one is a single task.
There’s no such thing as “we’ll implement AI and everything will be solved.” There’s “we’ll address this specific problem with this specific solution.” Then comes the next one, and that needs something different again.
Fully automating entire positions is extremely complicated and, for now, nearly impossible. But automating a large portion of tasks, especially standardizable ones, is already doable. But then who’s going to oversee these tasks? A manager? An AI specialist? That’s an interesting question.
Part 2: Scaling your AI project the right way
Not sure whether AI is the right answer for your challenge? A short conversation can save weeks of experimentation.
Contact our team: https://attrecto.com/contact
Our AI Enablement program is designed to help teams clarify their challenges and explore where AI can truly add value.
