Subscription or custom AI: which makes sense?

The choice between Subscribe to AI Tools and build something to measure it is placed as an aut-aut, and it is not. They are two different levels: one concerns how people work, the other how processes work.

The expensive mistake is not choosing badly between the two, but skip the first to go directly to the second. You end up building solutions for imagined problems, instead of those that emerge using tools in everyday work.

The Differences That Really Matter

The most undervalued line is the penultimate. A subscription can be discontinued tomorrow, but if you've built a way of working on it, changing providers means remaking habits and training. A tailored system is yours, but it requires someone to maintain it.

The signal that says when to pass

It's not the budget and it's not the size of the company. It's much more concrete: when people start doing repeated copy-paste between the AI tool and your systems.

It always happens the same way. Someone discovers that the assistant writes the answers well, but to do so he must first paste the customer data taken from the management software, then copy the answer and paste it into the email. It works, saves time, but each time there are two minutes of manual work.

Those two minutes multiplied by the number of times a month is the exact measure of what an integration is worth. If it is thirty a month, the calculation does not justify a project. If it is three hundred, yes.

When the subscription is enough and advanced

It is worth being explicit about something that those who sell projects rarely say: for a significant share of Italian SMEs the left column covers all real needs. If your case stands there, spending thousands of euros does not produce advantage.

The Balance Point on Volumes

There is a scenario in which tailor-made is also convenient without integrations: when people are many. A subscription costs per user, every month, forever. An internal system costs once and then has an exercise that does not scale with the number of users.

The bill is elementary: it multiplies the monthly fee per person by the number of people and by twelve. Compare with development cost plus annual exercise. On small teams he almost always wins the subscription; on large teams the draw can come within a few years.

However, be careful not to be fooled by economic comparison alone: a commercial subscription tool receives continuous improvements that an internal system does not have. The comparison is correct only at the same function, and rarely is.

Practical examples

Consulting firm, 8 people. Subscriptions for everyone, no plans. People use AI to write and analyze, the necessary data brings them who works. Cost: a few hundred euros a year. Trying to build something here would be wasteful.

E-commerce, 20 people. Subscriptions for marketing and customer care, plus a tailored system that responds to customers by connecting to management software. The two levels coexist: the subscription serves people's work, the system serves the process. It is the most common configuration among companies that get results.

Manufacturing company, 60 people. Here the volume bill begins to count, and in addition there is a confidentiality constraint on technical drawings that makes the use of public tools problematic. Case to be assessed for an internal solution, possibly with Local AI Models.

The Risk of Addiction, Explained Well

The topic of supplier dependency is dismissed too quickly, and deserves precision because it is different in the two scenarios.

With the subscription depend on the business choices of a company you don't control: it can raise prices, change usage limits, remove a function on which you've built a way of working, or be acquired and change direction. It is not a theoretical risk: in subscription software it happens regularly. The practical mitigation is not avoiding subscriptions — it would be absurd — but not building critical processes over specific functions of a single product, and making sure the data is exportable.

With the custom made Addiction shifts: it doesn't depend on a product but on who built the system. If the code is not yours, if no one has documented it, if the credentials of the services are in the name of the provider, changing partners becomes almost impossible. It's a more insidious addiction because it doesn't have a list price: it only emerges when you try to change.

• Ask in writing who owns the code and data
• Claim documentation readable by those who did not write it
• The credentials of the used services are sent to the company
• Verify that data is exportable in a standard format

There are four requests that are easily obtained before signing and with difficulty after. Those who work well have nothing against granting them; those who stiffen themselves are signaling something. The theme is taken up in DIY AI or agency.

A frequent perspective error

Many companies face this choice by wondering which of the two is "the AI solution". It's the question that leads astray, because it assumes that there's a time when you adopt AI and then you're fine.

In the reality of companies achieving results, the trajectory is different: subscription tools remain forever, because they serve people's jobs, and tailor-made systems are added one at a time as processes that are worth the investment emerge. Neither replaces the other, and those who expect the custom project to make subscriptions unnecessary are disappointed.

This perspective also changes the way you plan your spending: not a one-off investment but a small continuing entry plus timely investments. The complete reasoning is in AI budget for SMEs and in the general framework of how much AI costs in the company.

Mistakes to Avoid

• Skipping the subscription phase. You end up building for imagined problems.
• Customising what an instrument already does well. Expenditure without advantage.
• Staying at the subscription when the copy-paste has become systematic. You're paying in hours what an integration would solve.
• Not considering dependence on the supplier. Prices and functions change.
• Treat them as alternatives. In practice they live together.

How to apply it in the company

1. Activate subscriptions for those who can take advantage of it and let two or three months pass.
2. Ask people Where they waste time in the transition between the instrument and your systems.
3. It counts how many times a month that passage happens.
4. Choose a single process, the one with the highest account, and assesses the tailored for that.
5. Keep subscriptions: they serve at a different level.

Step 2 is the one that gives the best answers, because working people know exactly where the friction lies. The cost framework is in how much AI costs in the company, and the choice over who realizes in DIY AI or agency.

One last practical note: this assessment must be re-made every year. Volumes change, subscription tools add features that used to require development, and processes that didn't justify a project last year can justify it now. A decision made two years ago is not necessarily the right one yet.

Conclusion

Subscription and tailor-made are not alternatives: they are two levels. The first improves people's work and costs little; the second changes how a process works and costs more.

The sequence that wastes the least is always the same: subscriptions to figure out where there is value, and tailored only where the daily copy-paste shows that the value is there but you're paying for it in hours.

If you want to go from testing with ChatGPT to an AI system integrated into business processes, Giallo Studio custom designed agents and automations.