What we solve ·Where do I put the budget?
Vanguard · Guide
Where does AI belong in human resources, and where should it not be touched?
They are pushing us to put AI into recruiting and I am the one who signs off on whatever comes out.
Your area's use cases ordered by return and by risk, with the ones you should not touch named and argued.
Prioritization for people functionsSend this page to whoever decides
This sounds like you if
- The market pushes to automate selection and the area is the one answering for the result.
- A vendor is already selling their tool inside your company.
- The system rejects candidates and the reason for each rejection is not recorded.
Use this today, without hiring anyone
How to order your candidates by risk before return, which is the reverse of how they will be presented to you. Four questions per candidate and one afternoon.
- Who does it decide about, and what does it take away if it is wrong? A candidate that affects a person's access to a job or a benefit goes to the top of the risk list even if its saving is modest.
- Can the affected person notice and challenge it? An error invisible to whoever suffers it weighs far more than a visible one, because nobody ever corrects it.
- What data goes in? If something goes in that you would not ask a candidate in an interview, it should not go into the system either.
- Could you reconstruct one specific decision six months from now? If the answer is no, it is not ready yet, however good it looks.
Demand this in writing from any vendor before signing. What gets recorded about each decision, for how long and who on your side reviews it before it has an effect on a person.
How we solve it
The method, not the promise.
- The area's processes get mapped by volume and by real pain, not by what the vendor wants to sell.
- Every candidate gets classified by consequence for the affected person. That order, not the return one, governs the rest.
- The ones that should not be touched get dropped, with the argument written. That is half the deliverable.
- The bias, privacy and traceability guide gets written for each approved candidate.
- Those requirements get turned into acceptance criteria to demand from the vendor in the contract.
What you receive
- The inventory of the area's processes with volume and pain point.
- The ranking by consequence and return, with written criteria.
- The reasoned list of exclusions, which in this area is what protects the most.
- The bias, privacy and traceability guide and the acceptance criteria for the vendor.
The proof that applies here
- Eight years under regulatory supervision without a single finding, with automated decisions reconstructed at the reviewer's request.
- Governance of automated decisions in an environment where every rejection had to be explainable to the affected person.
- A generational succession executed without interruption, which was a change of people before it was a change of technology.
Before you hire
The firm prioritizes and bounds the risk; it does not design selection, performance or termination criteria and does no employee surveillance in any version.
The way in is through the automated decision, which is the same in recruiting as in credit: who answers for what the machine decided about a person and with what evidence that decision holds up six months later. That question was answered for eight years to a banking supervisor. The employment and privacy layer is contributed by a specialist, and without one the firm does not accept the engagement.
If someone challenges a decision your system made tomorrow, what do you defend it with?
If the problem is a different one
The assessments people bring me talk about the company. I run operations.
Prioritization for operations
The regulator asks why that case was rejected and nobody can reconstruct it.
Automated decision governance
The forecast does not hold and we do not know why a deal falls until it has already fallen.
Prioritization for enterprise sales