How to implement AI in business so that it brings money, not just hype?

We all like to start a “new life” on Monday, with a birthday or, of course, with a new year. In business, this is often called a “strategy” – a kind of collection of fantastic stories about what the future of the company might look like.

And, of course, when planning for 2025, it is almost like a bad manners to present a strategy without mentioning artificial intelligence. Who wants to look like a dinosaur? Although, to be honest, I have seen companies that work without CRM, in Excel, and earn so much that they can afford to buy a real dinosaur skeleton for the office.

The reality is that AI has been tightly integrated into our lives for at least two years. And if you open social media, you will find a lot of young “visionaries, ambassadors, and evangelists” who will gladly tell you how to transform your business with just one click: cut costs by 90%, increase revenue by 300%. But there is one small problem: if you call them, a real person will answer, not AI. And for some reason, they did not receive the Margin Leaders award either. Because business is a bit more complicated than just connecting a new technology.

Why does business start implementing something?

Where does a strategy start? In our case, with the owner or manager. Not much has changed since the creation of websites, automation, or CRM systems. These are all the same technologies. I identify four main reasons that push businesses to implement innovations:

“Gypsy Quadrant” of technology adoption (let’s call it that):

1. “Like a neighbor”
   Probably the most popular driver of progress. Everyone does it, so we need to do it. This is a kind of “cargo cult”, blind copying without understanding the underlying reasons. Sometimes it’s because of FOMO (fear of missing out), and sometimes it’s because of the desire not to be embarrassed in front of friends in the bathhouse. It rarely ends in success.
2. Transformation itch
   This is when the owner is itching to implement something. Good afternoon, my name is Andrii, and yes, I am exactly like that 🙂 I love technology, and yes, business for me is sometimes like a testing ground for experiments. It is important to find a balance between a “brilliant visionary” and a “crazy businessman”. But such people are usually “in the know,” they have seen a lot, and although not all projects take off, some of them work.
3. “Putting out the fire”
   This scenario is clear to everyone. The technology is being implemented as an optimal “fire extinguisher”. In any case, this is already a task that needs to be solved. The main thing is to choose something that will really help.
4. Balanced strategy
   This is when you have analyzed, listened to different opinions, chosen a planning horizon, developed a strategy, and started acting on it. If you are like that, write to me, I want to be friends with you 🙂 But, unfortunately, this is a rarity in our fast-changing world.

The main thing is to honestly answer yourself in which quadrant you are now. This will help you understand how to proceed.

Why implement artificial intelligence at all?

Okay, let’s think about why all this is necessary. Let’s take a look at your company’s charter – it is unlikely that there is anything about artificial intelligence or “world peace”. But there is definitely a main business goal – profit. The rest are just tools that help to get it.

Any decision or project should end with the phrase: “And it will bring us profit”. If you have an “itch for change for the sake of change,” then it will bring satisfaction to the owner, but calling it an “investment” is a bit of a sugarcoating of dividends.

The best approach is to determine what tasks you have now and how their solution will affect your income. Forget about all these technologies, Python, or Excel for a moment. Imagine you are sitting in a serious strategic session where everything looks very official. Now put all that aside and think about the 7 Dwarfs. Yes, imagine that you have 7 dwarfs who can do magical things. What should they do in your field? You need imagination here – don’t limit yourself to reality.

But don’t dream that “gnomes will cancel VAT” or “customers will buy 3 times more expensive without objection”. These are letters to Santa Claus, not business tasks.

If you don’t have at least 10 specific tasks in your head, or better yet, on paper, there is no point in even starting to implement any innovations. If you’re a manager or owner, let’s be honest: can you do it all on your own all the time? Most likely not. You have a lot of other tasks: the route has been closed, the money is stuck – this needs to be solved here and now.

Just like in the gym, you can train hard for a month and then eat fast food for a year. The result, to put it mildly, will not be impressive. Therefore, as in any business, you need a team.

Artificial intelligence is not a single technology, but a whole set of knowledge. There are many tools, and your task is to match your needs with these tools. It’s like Tinder: you need to find the perfect match. To do this, you need a specialist – what I call an “architect.”

II models: choosing the best one

And now to the main point: what is a model and why ChatGPT is not everything?

interface that determines how and where you interact with the model; there is its cost, purpose, etc..

Let’s take the car as an example:

• The engine is a model. Cars with a 1 liter and 5 liter engine look the same, but drive very differently. A larger engine gives you more speed, but it also eats more. It’s the same with LLM models: more powerful ones are more expensive.
  
• The keystone is the interface. For example, you can choose several models on the ChatGPT website, but the interface will remain the same. Interfaces can be different: website, Telegram bot, voice assistant, API console.
  
• Tuning is the adaptation of models for specific tasks. For example, just like in a garage, engines are “flashed” for maximum speed, AI models can be “tuned” for narrow tasks.
  
• Bundle – additional functions. For example, autopilot or lane keeping. In AI, it can be Excel analysis, voice input, or other amenities added by developers.

There are a lot of models now, for text, images, video, voice. And, of course, everyone is afraid of making a mistake: if you take one, it should be the “best” one. But no, there is no such thing as the best. There is only optimal for your tasks. That is why we return to the first point: determine what exactly you need.

There is no such thing as too much data

Nowadays, everyone is trying to create cool voice bots. There is already an OpenAI Realtime API, ElevenLabs also offers its solutions. Yes, they are still a bit crude, there are delays. But you should agree that by the end of 2025, we will have an AI assistant that will be able to communicate in any language without any problems.

But what’s next? You have a perfect bot ready to replace your call center, and there’s only one thing left to do: Upload your data about products, customers, company, etc. Why? Because AGI (general artificial intelligence) can’t go into your CRM, find the necessary dialogs, browse websites, or knowledge base on its own. You have to do it.

This stage is not the sexiest in the process of technology implementation. It’s routine, boring, but strategically crucial.

Prepare your data today to use the technology tomorrow.

The more quality data you collect now, the faster you can adapt the technology to it. One-size-fits-all models don’t know how to interact with your customers, which products you have the best, or how you usually solve problems. This all needs to be carefully collected in a structured way for the model to understand your business.

Remember: models have many roles, but they are not psychics. They will not be able to look into the heads of you or your employees.

For fine-tuning or any work with your databases, you need normal, organized data.

OpenAI recently showed that it will soon be possible to train o1 models that can “think”. What does this mean? In a simplified form, you give the model a query and the desired answer, collect 100-500-1000 such examples, and the model is customized to your tasks. The problem is that with this approach, the model is more likely to “adjust” the answers than to learn to think like you.

And in many processes – legal, technical, analytical – it is the logic and the train of thought that is important. And now, very soon, it will be possible to train the model so that it thinks like you.

What is needed for this? Data. Specific cases. Logical chains. Start collecting them now.

How much money do you need to be happy?

Or here’s another one: “Please tell me a free service that will analyze all competitors, write scripts, generate characters, make videos, edit them, and put everything online.” This request consistently wins the audience antipathy award, because I hear it several times a week.

This reminds me of the search for SMM managers: “You will bring us to the top, but there will be no money for advertising, because even a fool can handle money.” Yes, money is not the only part of success, but without it, nothing will work.

So, if you’ve read this far, you’re probably serious. Let’s figure out how much money you need to get your team on the same level.

How much money will you need to get started?

1. Text models:
   An architect needs access to several models, for example, ChatGPT, Claude, or Gemini. They have different functions, and it is important to compare them. Two paid models will cost about $50 per month.
2. Visual content:
   You need to create images, videos, music. This can be done through an aggregator, such as FAL (where you pay only for the work done), or through services: MidJourney, Runway, ElevenLabs, HeyGen. It’s about $50 more per month, depending on your needs.
3. Reposit for new products:
   A minimum of $100 per month should be free. As soon as a new network or service appears, you need to test it: buy tokens, try it out, and decide whether it is worth paying further.

Yes, there is another problem – some services do not accept our cards. But you are smart! If you can find a way to bring goods from Paris so that they are “supposedly from Bishkek” but via Istanbul, you will find a solution here as well.

Conclusions

1. Technology is just a tool.
   It’s cool, but will you be better off without it? Not a fact. Will you be better with it? Maybe, maybe not. It depends on how you use it. The tool itself doesn’t change anything. For some people, Excel is a chaos with numbers, and for others, it’s perfect accounting on 50 sheets with sub-tables.
2. Risk tolerance.
   All these costs are venture capital investments. Be prepared for the fact that in six months the results may be zero. In this case, the project will have to be rethought, corrected, and moved on. No one can guarantee success.
3. Competition at a new level.
   Even if super-universal technologies appear that solve all problems by themselves, you will have two problems:</span
   • If it’s too expensive, you won’t be able to afford it.
   • If it’s cheap, your competitors will implement it too.

And then a new round of competition will begin: everyone will have the best sites, the best offers, the best managers. How to fight for the attention of the consumer? It will be a tough fight at an even higher level.

It all comes down to one thing – the right process gives the right result. And it starts with clearly defined tasks.