When you hire a new employee, do you give them every key to the office on their first day?
Do you hand over access to the company bank account?
Do you send them all your internal documents, customer data and passwords, then say:
“There you go. Help wherever you can.”
Probably not. 😄
First, you explain what their job actually is.
You show them which systems they need to use. You give them the information and access required for their role. They learn which decisions they can make independently, and when they need to ask a colleague or manager for help.
An AI assistant may be a digital team member, but in this respect, the principle is very similar.
Just because it is capable of doing many things does not mean we should give it responsibility for everything.
First: what is its job?
One of the easiest mistakes to make is giving an AI assistant a task that is far too broad.
“Help our customers.”
It sounds reasonable. But what does that actually mean?
Should it answer questions about services?
Help someone choose the right product?
Book appointments?
Collect the information needed for a callback?
Handle complaints?
Those are very different responsibilities.
So when we design an AI assistant, the first question should not be:
“What can AI do?”
It should be:
“What exactly is this assistant’s job in this business?”
Then we decide what it should work with
A new employee also needs to learn how the company works.
What services do we offer? How does appointment booking work? What are the conditions of a particular offer? How do we answer the questions that come up again and again?
An AI assistant also needs reliable, approved business information.
Not because it is “not intelligent enough”, but because it cannot be expected to guess how your particular business operates.
And if it does not know something with confidence, there needs to be a rule for that too.
Should it ask a clarifying question?
Should it say that it cannot confirm the answer?
Should it pass the enquiry to a member of the team?
That is part of the job as well.
Capability and permission are two different things
This is one of the most important distinctions.
An AI system may technically be capable of carrying out a particular action.
That does not automatically mean we should allow it to.
An appointment assistant, for example, may be allowed to check availability and perhaps make a booking if that is part of its role.
That does not mean it needs access to every customer record, financial information or another system that has nothing to do with the task.
It is much like giving a new employee the key they need to enter one room. You would not hand them the entire company keyring simply because they need access to one door.
A well-designed system gives the AI access only to what it genuinely needs to do its job.
But a job description is still not the same as training
We would not normally give a human employee a document listing their responsibilities and consider their training complete.
We show them how the business works.
We go through common situations. We explain how we communicate with customers. We show them what should happen when something does not follow the usual pattern.
And ideally, they also learn:
“If I reach this point, this is no longer my decision. This is where I ask for help.”
With an AI assistant, that has to be designed deliberately.
It is not enough to tell it what to do.
We also need to prepare it for how to do that job within this particular business.
That may include approved knowledge, communication rules, permissions, prohibited actions, unusual situations and a clear process for handing something over to a person.
Put simply:
A job description tells you WHAT your job is.
Training teaches you HOW to do it.
An AI team member needs both.
How do we approach this at NLS?
We do not start by choosing an AI model and then looking for something to automate with it.
We start by understanding the business.
What is happening now?
Where are customers, enquiries or time being lost?
Which tasks keep repeating?
Which of those tasks could sensibly be handled by an assistant?
What information and access would it need?
And at what point should a person take over?
Only then does it make sense to design the assistant’s role and prepare it for the job.
Because an AI assistant does not become a useful business team member simply because it can hold a conversation.
It needs to know what its job is. It needs to understand its boundaries. And it needs to know when it is not the right person — or, in this case, not the right non-person — for the task.
So before asking which AI solution your business should choose, there may be a simpler question worth asking first:
What exactly are we hiring it to do?