AI as an accelerator: where it actually helps in HR work
Less manifesto, more practice: the tasks where AI genuinely saves me time, the ones it makes worse, and the criterion I use to decide.
There is a big difference between using AI to produce volume and using AI to save time on what actually eats your day. The second one changes the work.
Research from ABES in partnership with IDC indicates that 95.2% of Brazilian companies already plan to bring AI into some function. The number is high enough to make the real question relevant: bring it in where, exactly.
The criterion: AI accelerates the draft, not the decision
I use a simple cut. AI fits well in tasks with these three characteristics at the same time:
- The output is a draft, not a commitment.
- I have the repertoire to judge whether it came out good or bad.
- The cost of being wrong is low and reversible.
When one of the three fails, I do not use it. A decision about a person is not a draft, the cost is not reversible, and the responsibility is not delegable.
Where I genuinely gained time
First version of a communication plan. A change plan has many pieces carrying the same content in different registers: the leadership announcement, the champion material, the team FAQ. Writing the first version of all those pieces from one core message is exactly the kind of work where I save an afternoon.
Synthesizing open survey responses. When a climate survey has a thousand open comments, a full read never really happens. Grouping themes and surfacing representative quotes per group is useful, as long as I read the original quotes before concluding anything.
Preparing for a leadership meeting. Feed in the project context and ask for the ten most uncomfortable questions that director might ask. Half are useless. The other half keep me from being caught off guard.
Translating analysis into business language. I know what the data says. Making it fit into three sentences an operations manager understands mid-meeting is a different job, and AI is good at it.
Where it got in the way
Organizational impact assessment. I have tested it and the result is always plausible and generic. Real analysis depends on knowing that one specific coordinator is the person holding the relationship with three carriers. That is in no document.
Text that goes out under my name without review. People can tell. And when they can tell, the cost is credibility, which is the most expensive asset to rebuild.
Anything involving sensitive personal data. Names, performance reviews, health or termination information do not go into a tool I do not control. That is not excessive caution, it is the minimum.
What I would tell someone starting out
Do not start with the tool. Start by listing what consumed your time last week and mark what was a draft. It is a much shorter list than the discourse suggests, and it is exactly where the gain is.
And train the team properly. Granting access is not enough: people need to know how to write a good request and, above all, how to critically assess the answer. Without that, the tool becomes a generator of bad text that looks good.
Source: adoption figures cited by ABES in partnership with IDC.