SD Infotech helps mid-sized companies adopting AI recruiting tools decide, task by task, where AI can act on its own, where it should assist a recruiter, and where a person must stay accountable. Every recommendation rests on evidence, not vendor claims.
Keyword screening rejects people who can do the job and passes people who only list the words. Adding AI on top of an unclear requisition automates the same mistakes, faster.
qualified candidates found by a keyword screen in our screening pilot. The evidence-based screen found 10.
candidates rejected by the evidence-based screen without a person looking. The keyword screen rejected 14.
separate tasks and decisions in a typical recruiting process. Most AI decisions are made at the level of "screening" or "sourcing".
Pilot figures come from a controlled test on a constructed résumé set for one role. A live-requisition pilot is how we measure it in your organization.
Seven layers that take a hiring need from a vague request to a decision someone can defend. Layer 3 is the heart: what counts as proof that a person can do the work.
What must this hire change, and by when?
Outcomes and capabilities, not inherited JD lines.
Each claim is corroborated, unsupported, contradictory or unknown.
Designs how the decision is made. It does not make it.
Who can, should and must act on each task.
Scored per use case, never one company-wide score.
Outcome evidence that earns AI more autonomy over time.
Every task is scored on four dimensions. Autonomy is earned, never assumed.
AI autonomy = min(Readiness, Suitability, 4 − Risk)
Decisions that exclude a candidate stay human by default. A company that proves readiness with outcome data can move them one step, to AI recommends, and never straight to AI acts.
Your recruiting process broken into its real tasks and decisions, each scored and labelled Automate, Hybrid or Human.
Test a vendor's tool against the decisions it would influence. Find out what it may decide, what it must only recommend, and what evidence it needs.
Run intake and evidence screening alongside your current process on one open requisition. Measure qualified candidates found, missed and time to shortlist.
As your evidence and outcome data improve, tasks earn more AI autonomy. We re-score the map and update the rules.
Break your process into tasks, decisions and judgments.
Score readiness, suitability, risk and accountability per task.
Label each task and write the governance rules.
Hand over the map and a 90-day plan your team can run.
Owns hiring quality and speed, and the decision to adopt AI.
Redesigning the TA operating model and needs a defensible allocation of work.
Selecting and configuring recruiting AI and accountable for how it behaves.
One role, 22 résumés, two screens. The keyword screen found 3 of 11 qualified candidates; the evidence screen surfaced 10, with every rejection left to a person. The evidence screen also missed: it advanced one candidate with invented metrics, which is why interviews still verify evidence.
Constructed résumé set for one role. Method and data available on request.
Ten real job descriptions run against the résumés actually submitted. Half passed a literal keyword screen and four showed strong evidence for the role. None received a reply. Keyword matching explains only part of what filters candidates out.
Founder's own applications, analysed with consent.
Paps Muppa founded SD Infotech LLC to fix how companies decide who gets hired, and how much of that work AI should do. He has recruited since 2001: technical and executive search, leading recruiting teams, and in-house leadership hiring at JPMorgan Chase.
EHDA came from watching good candidates rejected by filters that never read their evidence, and from watching companies adopt AI without deciding what it was allowed to decide.
Tell us which AI tool or task you are weighing. We'll tell you how we'd score it and whether a diagnostic is worth it.