Today it seems like every business problem has an immediate solution
“Let’s add Artificial Intelligence.”
Customer service, administration, sales, document analysis, opportunity tracking, reports, Human Resources.
And yes: technically, we can automate more and more things.
But there is an important difference between something that can be automated and something that is actually worth automating with AI.
There are processes where adding Artificial Intelligence can save hundreds of hours.
And there are others where it only adds cost, maintenance, and the possibility of errors to something a simple rule could solve perfectly well.
So, before asking:
“What can we do with AI?”
it is better to ask a much more important question:
“Where in the process do we actually need AI?”
1. When a simple rule already solves the problem
Suppose we want to automate invoice approval.
The condition is:
If an invoice exceeds USD 5,000 → send it to the person responsible for approval.
We do not need Artificial Intelligence.
We have a perfectly defined rule.
The same can happen when we need to:
- send an alert when an invoice is due;
- assign a lead based on province;
- copy information between systems;
- create a task when a certain event occurs;
- update a spreadsheet;
- generate a report from structured information.
Using an AI model to solve these cases will probably make the solution more expensive and less predictable without adding much value.
Now let’s change the problem:
An email arrives from a customer and we need to determine whether they are making a complaint, requesting support, asking for a quote, or simply making an inquiry.
Now we need to interpret information.
And that is where Artificial Intelligence starts to make sense.
A simple rule to identify it
If we can clearly write “IF X happens → do Y,” we probably do not need AI.
If we need to understand text, context, images, documents, or conversations before knowing what to do, then it may be a good candidate.
2. When the process is still disorganized
Automating a poorly defined process usually does not solve the problem.
Many times, it simply speeds it up.
Imagine a company wants to automate discount approvals.
During discovery, we find that:
- each salesperson uses different criteria;
- some approvals happen over WhatsApp;
- certain customers have special conditions;
- there is no clear policy;
- several exceptions are known by only one person.
In that scenario, the first problem is not technological.
The process still needs to be organized.
If five people give different answers to the same situation when we ask “what should happen here?”, building a reliable automation will be very difficult.
First, we need to define:
inputs → rules → exceptions → owners → expected result.
Then we can evaluate which part makes sense to automate.
AI can help execute processes.
But it should not be used to hide processes that nobody has managed to define yet.
3. When automating costs more than doing the task manually
Not everything repetitive deserves automation.
Suppose there is an administrative task that:
- happens twice a year;
- takes twenty minutes;
- almost never generates errors.
We could probably automate it.
The question is:
is it worth it?
Because developing an automation also involves discovery, configuration, testing, monitoring, and maintenance.
Before automating a process, it is worth looking at least at:
- how often it happens;
- how much time it consumes;
- how many people are involved;
- how much it costs to do manually;
- how much impact a delay creates;
- how much it would cost to automate.
A low-volume, low-impact task is probably not the best place to start.
Technology should generate a return.
It should not become a project simply because we can technically do it.
4. When an error can have a consequence that is too significant
Suppose we have two systems with 95% accuracy.
The first uses AI to classify sales inquiries.
If it makes a mistake, a salesperson can correct the category.
The second uses AI to automatically approve an important financial transaction.
The accuracy is the same.
The risk is not.
That is why, when evaluating an automation, it is not enough to ask:
“How well does it work?”
We also need to ask:
“What happens when it makes a mistake?”
Special care is needed when an automation can:
- move significant amounts of money;
- block customers or users;
- approve or reject sensitive operations;
- execute actions that are difficult to reverse;
- create legal or contractual consequences;
- directly affect a person.
This does not mean AI cannot participate.
In fact, it can be extremely useful for analyzing information, detecting anomalies, prioritizing cases, or preparing recommendations.
But many times the best design is not:
AI decides → system executes.
Instead:
AI analyzes → proposes → human validates → system executes.
The greater the potential cost of an error, the greater the control over the automation should be.
5. When human interaction is part of the value
Not every human intervention is an inefficiency.
In some processes, talking to a person is precisely part of the service.
Think about:
- an important sales negotiation;
- a difficult conversation with a customer;
- a leadership situation;
- complex feedback;
- conflict resolution;
- decisions where there is a great deal of implicit context.
In all these cases, AI can help a lot.
It can summarize previous conversations.
Prepare information.
Detect important topics.
Recommend next steps.
Even draft a possible response.
But that does not mean we should fully automate the interaction.
Sometimes the best use of AI is not to replace the person.
It is to allow that person to arrive better prepared.
The mistake is not automating. It is using AI where it is not needed.
There is an important distinction that is often lost:
rules, automation, and Artificial Intelligence are not the same thing.
If we know exactly what needs to happen
We probably need a rule.
Example: If an invoice exceeds a certain amount → request approval.
If we need to connect systems and execute actions
We probably need an automation.
Example: When a customer completes a form → create the opportunity in the CRM → assign it to a salesperson → send a notification.
If we need to interpret information
That is where Artificial Intelligence may come in.
Example: Read the prospect’s message → identify what they need → detect urgency → summarize the case.
And, in practice, the best solutions often combine all three.
For example:
- a customer email arrives;
- AI interprets what they are requesting;
- a rule determines which area should handle it;
- an automation creates the case in the system;
- if it detects a critical situation, a person intervenes.
The key is not to use AI at every stage.
It is to place it exactly where it adds value.
So, which tasks SHOULD be automated?
The best opportunities usually appear when a task combines several of these characteristics:
- it happens frequently;
- it consumes many hours;
- it involves repetitive tasks;
- it uses digital information;
- it has relatively clear rules;
- it requires copying information between different systems;
- it generates manual errors;
- it creates bottlenecks;
- it delays other people.
And if the process also needs to interpret:
documents, emails, conversations, images, or unstructured information,
that is where adding Artificial Intelligence can create a much bigger leap.
Five questions before automating a process
Before developing any automation, we can make a fairly simple assessment.
1. How often does it happen?
The higher the volume, the greater the potential impact.
2. How much time or money does it currently consume?
Not every repetitive task represents a problem large enough to justify automation.
3. Is the process clearly defined?
If nobody can explain what should happen, there is still work to do first.
4. What happens when the system makes a mistake?
The cost of the error determines how much control we need.
5. Do we really need Artificial Intelligence?
Maybe yes.
Maybe automation is enough.
Maybe a rule is enough.
And making that distinction before starting can save a great deal of time and money.
What should you automate first in your company?
Once we understand which processes are not worth automating, an even more interesting question appears:
Where should we start?
That will be one of the central topics of our next webinar:
What should you automate first in your company?
We will show how to identify processes with automation potential, how to compare them, how to prioritize them, and how to detect when we need an integration, traditional automation, or an Artificial Intelligence-based solution.
Because adding more AI does not necessarily mean working better.
The goal should be much simpler: find where technology can create a real impact on the business.
