Hiring · the deployment we can describe

Top-performer hiring at one Fortune 1000 insurance carrier

Saad Bin Shafiq, Founder, Nodes·Last reviewed Nov 16, 2026·Read the paper

Nodes is in production at several companies. The one we can describe in public is a Fortune 1000 insurance carrier, live since January 2025, and hiring is one of its three workflows. This page is the hiring hub: what the deployment does, the research behind it and what a hiring team should test before relying on any of it.

Source: "Decision Traces," Saad Bin Shafiq, Nodes, 2026. Research cohort of 10,765 agents hired 2022 to 2025 at a Fortune 1000 insurance carrier. Read it on arXiv.

What the deployment does

The current insurance application evaluates resume and assessment evidence against configured hiring criteria. It has evaluated 900,000+ people using location-specific top-performer profiles across 200+ locations nationwide, and the carrier has hired 10,000+ insurance agents from that pool. Those hires are separate from the 10,765-person research cohort below.

Recruiters and managers make the final call. Nodes never automatically rejects a candidate. Each new application needs its own scope, evidence and validation.

The same carrier also runs two other Nodes workflows: licensing progress, with approved reminders based on each person's latest activity, and educational support, an assistant that uses each person's context. The hiring results below do not describe those workflows. See all three at the carrier.

Test hiring rules against later results

Most screening measures how much a candidate resembles a job description. The more useful question is which permitted signals went with the outcome the business defined, such as a first production milestone, ramp time or another approved measure. That test needs four things:

  • Connected records: permitted ATS, HRIS, assessment and CRM records, linked under the company's rules.
  • A defined outcome: a documented milestone the business already trusts.
  • Historical validation: a rule or model tested against completed cases before it is used live.
  • A named owner: a person who can approve, edit or decline what is proposed.

Past hiring decisions are context, never ground truth. A manager's approval shows what someone chose, which is different from whether it worked.

What the research showed

  • In the reference carrier record, none of 3,597 testable resume keywords predicted the first production milestone after Bonferroni correction, and 30 were anti-predictive. This does not establish the same finding for every role. Details.
  • An insurance-experience filter would have rejected 2,863 producing agents associated with $17.7M in annual production. This retrospective counterfactual does not establish incremental revenue caused by Nodes. Details.
  • A fitted relationship in that carrier associated each day faster to the first production milestone with $54.35 more in annual production per producing person. It is a planning estimate rather than daily cash or a causal guarantee. Details.
  • Resume keywords reached an AUC of 0.558. In a small n=229 sample, behavioral assessment alone reached 0.647 and fused evidence reached 0.735. These are ranking results in a small research sample, and AUC does not measure accuracy. Details.

The research library keeps the population, method and limits for each finding.

Where a hiring team should start

Start narrow enough to test. A practical first question is who advances from screening for one role family, measured against a post-hire outcome the business already trusts. Before any access, agree:

  • which current signals go with the outcome
  • which filters reject people who later succeed
  • where the evidence is too weak to support a recommendation
  • which differences between groups need review
  • who may approve, change, delay or decline

Scheduling and other routine coordination can continue within the approved workflow permissions without a new human click for every internal step. Actions requiring approval remain gated by the company's policy. Quarterly production reports remain evidence of the results they describe. New role families, selection policies and workflow scope need their own validation. The owner expands, holds or narrows delegated work after reviewing agreed task quality, exceptions and total AI and human cost.

Group outcomes need review, with limits

A group-outcome review can surface differences that need investigation. It cannot prove that a process is unbiased. Your people, legal and data-governance teams decide how protected data is handled, which groups are compared, the thresholds and the corrective steps. The Decision Trace keeps the evidence, recommendation, human input, action and later outcome, including failed and inconclusive results.

How the categories compare

ApproachRecognizable examplesTypical inputQuestion to ask
Behavioral and skills assessmentsPredictive Index, SHLpersonality, skills or cognitive measuresIs the measure validated against this role's outcome?
Video interview and screeningHireVueinterview and screening evidenceHow is the output linked to later performance?
Talent intelligence platformsEightfoldskills and role-matching dataCan the customer test the match against its own outcomes?
ATS and HRIS modulesWorkdayworkflow and employee recordsCan evidence be reconciled across systems and traced to an action?
Nodesthis deploymentATS, HRIS, assessment and production outcomesDo the company's own outcomes support the rule?

These examples identify categories, not a ranking or a claim about every product configuration.

Security and deployment

Nodes supports Nodes Cloud, a single-tenant customer VPC, and customer-managed on-premises deployment. Private configurations can enforce zero customer production-data egress where the approved model and operational paths preserve it. Nodes Cloud has a separate Nodes-managed boundary. Nodes is SOC 2 Type II attested. Review deployment controls.

Frequently asked questions

Does this replace the ATS?

No. The ATS stays the system of record. Nodes reads permitted candidate, performance and assessment records and records what was decided.

Who makes the employment decision?

A named recruiter or manager. Nodes can recommend, explain, abstain and prepare the next step.

How accurate is it?

Accuracy depends on the data, outcome and validation method. In the published study, fused evidence reached an AUC of 0.735 on a small n=229 research sample, compared with 0.558 for keyword screening. AUC measures ranking. It does not give an accuracy percentage.

Does the carrier result transfer to my company?

No result transfers automatically. Another company or role is a fit for testing when it has a repeated hiring decision, a measurable post-hire outcome and enough completed cases.

Can Nodes help with more than hiring?

Yes. Hiring is one workflow at one carrier. Most companies start with a question about money lost between teams. See how the pilot works.