The cost logic
The question I hear most often from managing directors is what AI in HR costs, and the honest answer is that the tool is the smallest part. In companies with 50 to 500 employees, the model licence is usually already in the IT budget, because Microsoft 365 with Copilot, ChatGPT Enterprise, or Claude for Work was bought for the whole company, not for HR. What costs is the work of turning a licence into a routine: deciding which routines, writing the templates, agreeing the rules with the works council, training the team, and reviewing the results.
That work is mostly one-off. The saving is recurring. A job ad routine is set up once and used forty times a year. A reference letter template is written once and used every time someone leaves. This asymmetry is the whole business case, and it is also the reason the case only works if the routine is actually used after the project ends. A routine that is set up and forgotten is a pure cost. That is why this page talks about owners and reviews as much as about setup.
The four cost components
- LicencesOften zero additional cost, because the enterprise tool already exists. Where a licence is new, it is a per-user monthly fee for the HR team, small next to the other components. A specialised HR AI product is a separate decision and not assumed here.
- Setup and training, one-offPer routine: the template, the prompts, the checklist, a pilot, and a half-day training for the team. In my projects, a few working days per routine, delivered as a scoped project on a day-rate basis.
- Works council agreement, one-offWhere the routine falls under co-determination: two to four months of elapsed time, several sessions, HR preparation, sometimes a lawyer's review. Not needed for pure text drafting without personal data.
- Ongoing reviewA few hours a month: sampling outputs, updating the knowledge base, reporting to the works council where agreed. This is the component most business cases forget and the one that keeps the routine trustworthy.
Manual routine against AI routine
The comparison that matters is per routine and in hours. The bars below show the pattern I see across the five routines on this site, as relative effort, not as amounts. The exact figures for your team come out of the time calculator and the conversation that follows it.
The honest row is the third one. An AI routine has a setup cost the manual routine does not have, and for a routine used twice a year it never pays back. The routines that pay back are the ones with volume: job ads in a company hiring thirty people a year, reference letters where twenty people leave, employee questions in any company above a hundred people. The first step of any project is therefore to pick the routines with volume and leave the rest alone.
A worked example, in hours
Take an HR team of three people in a company with 180 employees, five open roles a month, and six reference letters a month. Using the assumptions from the time calculator, that team ties up roughly six hours a week per open role, two and a half hours per reference letter, and three hours a week per HR headcount in standard questions. Those figures come from my own roles, not from a survey, and they are adjustable. Added up, the three routines bind a substantial part of one full-time position every week.
In my experience, about 60 percent of that can be freed with clean routines: the drafting, the sorting, the standard answers. The remaining 40 percent is judgement, conversations, and review, and it stays. Against the freed hours stands a one-off setup of a few days per routine, possibly a works council agreement, and a monthly review. Over twelve months, the freed hours exceed the setup by a wide margin for a team of this size. Over three months, they do not. The decision is therefore about horizon and volume, and both are yours to enter. The calculator does the first step, the call does the second.
What a project with me costs
I work on a day-rate basis, with a services contract that fixes scope, duration, deliverables, and confidentiality. An AI project for an HR team is scoped by the number of routines and by whether a works council agreement is part of it. The day rate depends on that scope and is named on the first call, so that you can compare it against the hours the routines currently tie up. There are no licence resales, no tool commissions, and no retainer after the handover. For a single routine that can be set up in an afternoon, a half-day project sprint is the smaller format. What the rest of the AI offer looks like is under AI for HR teams.
AI routines and works council agreements introduced in HR teams, HR admin cut from 30 to 5 hours a week in one role. Day-rate logic from measured routines. More about me
Frequently asked questions
Why is there no price on this page?
Because the project cost depends on the number of routines, whether a works council agreement is needed, and which tools are already in place. A flat figure would be wrong in one direction or the other. On the call I name the day-rate logic after ten minutes, once I know those three things.
Do we need to buy a new tool?
Usually not. Companies with 50 to 500 employees that run Microsoft 365 already have Copilot available, and many have ChatGPT Enterprise or Claude for Work. The routines on this site are built in those. A specialised HR AI product is a separate decision with a separate business case, and rarely the first step.
Is the works council agreement a recurring cost?
No. It is a one-off negotiation effort of two to four months, with HR time, a few sessions with the works council, and sometimes a lawyer's review. Afterwards there is a review clause, typically annually, which is a short meeting, not a new negotiation.
How quickly does the time saving show?
For text routines such as job ads and reference letters, in the first month after training. For a bot, after the pilot, so about three months in. For screening, only after the agreement and the pilot, so five to six months. The saving is per week and recurs, which is why the total case works out over a year.
Is AI in HR cheaper than hiring another HR person?
For recurring standard work, in my experience yes, because the freed hours recur every week and the setup cost is paid once. For judgement work, no, and it should not be compared. The honest comparison is the hours the routine ties up today against the hours it will tie up after, plus the one-off setup.
What if the team does not use the routines after the project?
That is the main risk and the reason the project includes training, a filed set of prompts and templates, and a named owner per routine. A routine that lives in one person's head is not a routine. The handover at the end of the project is how the saving survives me.