Devon is a recruiting manager at a 700-person consumer-products company, and he walked into the AI-tool approval meeting with one slide: the screening tool would save his team time. He walked out with a delayed decision. The CFO wanted ROI, the legal team wanted risk mitigation, the CEO wanted strategic alignment, and the DEI lead wanted to know the impact on diversity, and Devon had answers for none of them. He learned the hard way that a business case is not one argument; it is several arguments, each built on evidence and each aimed at a specific decision-maker. This lesson rebuilds his case the way it should have been built the first time, across efficiency, quality, fairness, and risk.

The Efficiency Case

Most cases start here because efficiency is the easiest dimension to quantify. Build it from a baseline: the hours recruiters spend screening per week, their fully loaded cost per hour, time-to-hire from open to start, and cost per hire, which is recruiting costs plus hiring manager time divided by hires per year. Then estimate AI impact, staying conservative, and compute annual savings. The structure is simple: hours saved per week times weeks times loaded hourly cost, minus the tool's annual cost. The discipline is in being honest about the assumptions rather than reaching for the most flattering number.

Two reference points help you set those assumptions without guessing. A typical AI screening deployment is described as cutting screening time by somewhere in the range of 30 to 40 percent and removing roughly 5 to 10 days from time-to-hire, and the tool itself typically runs between $500 and $2,000 a month, or $6,000 to $24,000 a year. Pick your point inside those ranges and say why you picked it.

You can express the resulting benefit three different ways, and which one lands depends on who is listening. The first is straight time savings: hours saved per week multiplied by roughly 50 working weeks multiplied by the loaded hourly rate. The second is a cost-per-hire improvement: old cost per hire minus new cost per hire, multiplied by hires per year. The third is a speed argument: old time-to-hire minus new time-to-hire, multiplied by the value of hiring faster. That last quantity is worth defining out loud, because executives will ask. Each week of speed buys you a week of earlier productivity from the new hire, plus the competitive advantage of reaching a candidate before another employer closes them.

Worked Example: Devon's Efficiency Math

Devon's baseline: 2 screeners at 40 hours a week, 80 hours total, at a fully loaded $50 per hour, is $4,000 a week of screening labor. He assumed the tool cuts screening time by 40 percent, freeing 32 hours a week, or $1,600 a week. Across 50 working weeks that is $80,000 a year in recovered recruiting time. He added a time-to-hire improvement of about 10 days, which pulls forward new-hire productivity, then subtracted the tool's cost, which he priced at the high end of $24,000 a year ($2,000 a month) to avoid sticker shock later. Net efficiency benefit landed around $60,000 to $74,000 a year. These are Devon's own pipeline figures, illustrative of the calculation rather than a published benchmark, but presenting the range, with the cost subtracted in plain view, is what made the number credible to his CFO.

Devon learned to tie the efficiency number to volume rather than leaving it abstract. His team handles roughly 45 open requisitions in a typical quarter, each generating about 120 applicants, so screeners process around 5,400 resumes a quarter by hand. At a current time-to-fill of 42 days and a fully loaded cost-per-hire near $4,500, even a modest compression matters: shaving 10 days off time-to-fill across 45 reqs is 450 req-days of reduced vacancy, and vacant roles carry their own drag on the teams waiting for them. When Devon expressed savings as both dollars and req-days, the CFO could see the operating impact, not just the line item. He also resisted the temptation to claim the tool would let him cut a headcount; recovered hours went toward sourcing and candidate experience instead, which kept the operations lead from reading the case as a layoff plan and turning into an opponent.

The other discipline Devon applied was sensitivity analysis. Rather than defend a single point estimate, he showed the number at three assumption levels: a pessimistic 25 percent time reduction, his base case of 40 percent, and an optimistic 55 percent. Even the pessimistic case cleared the tool's cost, which is the answer a skeptical CFO actually wants, because it proves the investment survives bad assumptions. Showing the floor, not just the headline, is what moved the conversation from whether to fund it to how fast to roll it out.

The Quality Case

Efficiency alone is not compelling; a faster process that hires worse people is a loss. So Devon shows quality holds or improves, using metrics his company already tracks: one- and two-year retention, hire performance against prior cohorts, promotion rates on expected timelines, and offer acceptance rates. The argument runs on retention math. If 85 percent of hires currently stay a year and replacing a hire costs $10,000, the annual turnover cost is 15 percent of his hires times $10,000. If better, more consistent screening lifts retention to 87 percent, the saving is the 2-point improvement times annual hires times $10,000. He is careful to source the improvement estimate from pilot data or vendor benchmarks rather than hope, and to present it as an estimate.

The general form of the quality argument is worth stating separately from the arithmetic, because it is what you actually say in the room. Establish the current level, the share of hires who stay a year and the share rated as high performers. Assert, with evidence, that quality will be maintained or improved. Then translate any improvement into money, since lower turnover means lower repeat recruiting cost and stronger performers mean more value per hire. Retention improvement is simply the cleanest of those translations to demonstrate.

The Fairness Case

Some stakeholders, the DEI lead above all, care most about fairness, and the case is that AI can improve equity rather than threaten it. The metrics are diversity of hires broken down by demographic group, time-to-decision by group so you can see whether some groups wait longer, offer acceptance by group, and disparate impact ratios showing whether groups advance at different rates. Devon frames it concretely: suppose women currently advance from screen to interview at 25 percent and men at 40 percent, a DIR of 25 / 40 = 0.625, below the EEOC four-fifths threshold of 0.80 and a genuine legal exposure, with a potential EEOC claim if the disparity continues. Blind screening that strips names and schools, plus standardized criteria, is projected to move the DIR toward 0.85, back within the compliance band. The business value is threefold: reduced legal risk, a stronger employer brand, and a larger effective talent pool. These DIR figures are illustrative of how he would model his own data, not an external statistic.

The Risk Case

Decision-makers, especially legal, think in risk. Devon shows both sides honestly. Risks AI reduces: unconscious bias through structured assessment, inconsistency through standardized criteria, speed risk when slow hiring loses candidates to competing offers, and scalability risk when a growing company cannot manually screen enough candidates. New risks AI creates: bias amplification if trained on biased data, opacity when candidates do not know why they were rejected, failure to recognize unconventional talent in candidates with non-traditional backgrounds, and regulatory exposure if the tool is not defensible. The case is the net: quantify current exposure, such as the cost of a past discrimination claim, then show that the new risks are mitigated by design through audit processes, bias testing, and human review, so AI lowers overall risk versus the status quo.

Devon made the regulatory exposure specific because vagueness reads as weakness to a lawyer. His company hires in New York City, which means any automated employment decision tool falls under NYC Local Law 144, requiring an independent bias audit within the prior year, public posting of the audit results, and advance notice to candidates. He folded those obligations directly into the case as line items with owners and dates rather than leaving them as open questions, so legal saw a managed compliance plan instead of a new liability. He also flagged ADA exposure: any video or game-based assessment must offer reasonable accommodation and must not screen out candidates with disabilities, which constrained which vendors he would even shortlist. By naming the statutes and showing the controls already mapped to each one, Devon converted legal from a blocker into a co-sponsor.

He closed the risk section with the asymmetry that decides most of these debates. The status quo is not risk-free; it is simply risk that nobody has measured. Inconsistent human screening already exposes the company to disparate-impact claims, and Devon could point to the unaudited DIR gap in the fairness section as live exposure that exists today, with or without AI. Framed that way, the question stops being "does AI add risk" and becomes "which risk profile is more defensible," and a documented, audited, human-reviewed AI process is easier to defend in front of the EEOC than an undocumented manual one.

Three Anti-Patterns

The efficiency-only case. Devon's original slide: $80,000 in time savings and nothing on quality, fairness, or risk. It happens because efficiency is easy to calculate and usually flattering while the other dimensions are harder to quantify, and the cost is that every other stakeholder's real concern goes unanswered and the decision stalls. Build a balanced case across all four dimensions.

Over-promise and under-deliver. A company promises AI recruiting will improve diversity, launches with great optimism, and six months later finds diversity metrics have actually worsened because of unforeseen bias in the tool. Enthusiasm invites over-promising and reality is more complicated, so the company ends up looking unreliable and trust in leadership erodes, which makes every future claim harder to land. Be honest about what you expect, transparent about the uncertainties, and explicit that you will monitor and course-correct.

Missing the key stakeholder. A great case wins most stakeholders but never addresses the CFO's cost concern, so the CFO approves only 30 percent of the requested budget. You focused on the concerns you cared about most and missed the one held by the person controlling the resources, and the result is approval without the means to implement properly. Identify the decision-makers first, then build arguments aimed at each one's specific concern.

Putting the Four Dimensions Together

The four arguments are stronger as one structure than as four slides shown in sequence, and Devon learned to sequence the room rather than the deck. He opens with the dimension that the most powerful skeptic cares about, because winning that person early shifts the gravity of the meeting. In his case the CFO held the budget, so he led with the efficiency floor from his sensitivity analysis, the number that survives pessimistic assumptions. Once the CFO was nodding, Devon moved to quality to show the savings were not bought by hiring worse people, then to fairness and risk, where legal and the DEI lead were waiting. Each dimension answered the previous objection before it was raised, so the argument felt like one case building on itself rather than four pitches competing for attention.

Devon also built a single summary view that put all four dimensions on one page: efficiency in net annual dollars and recovered req-days, quality in retention points and turnover cost avoided, fairness in the DIR movement from 0.625 toward 0.85, and risk as a short ledger of exposures reduced against new exposures mitigated. The point of the summary is not to flatten the detail but to let a busy executive see the whole shape of the case at a glance, then drill into whichever dimension they own. He paired it with a phased ask: a 90-day pilot on two job families with predefined success metrics, an audit checkpoint, and a go or no-go gate before any wider rollout. A staged commitment is far easier to approve than a blanket one, and the pilot data replaces his estimates with his own evidence for the next conversation.

Practice

Build these against your own organization's numbers, because a business case assembled from generic figures convinces nobody who controls a budget.

• Calculate your efficiency case. Work out the time savings AI screening would produce in your organization, what the tool would cost, and what the resulting return on investment looks like once the cost is subtracted in plain view.
• Build the quality case. Decide how you would maintain or improve quality of hire while improving efficiency, and which of your existing quality metrics you would commit to reporting.
• Construct the fairness case. Name the fairness problems your current recruiting actually has, then set out specifically how AI would address them and how you would know whether it did.
• Design the risk mitigation case. List the risks AI reduces for you and the new risks it introduces, then argue the net position rather than pretending one side of the ledger does not exist.
• Tailor to your stakeholders. Identify the three people who most determine whether AI recruiting happens in your organization, and write a customized argument for each one built on what that person actually cares about.

Reflection

Answer these before your next approval meeting, not during it.

• Which of the four arguments, efficiency, quality, fairness, or risk, is genuinely strongest in your organization right now?
• Which argument would most convince your CFO, and do you have the numbers to make it?
• Which argument would most convince your legal team?
• Which argument would most convince your DEI leader?
• What is the one concern you would need to address head-on to earn genuine buy-in from leadership rather than reluctant approval?

Glossary

• Business case. A documented argument for an investment, covering costs, benefits, risks, and a recommended course of action.
• ROI (return on investment). The financial return from an investment, calculated as benefit minus cost, divided by cost.
• Disparate impact. A hiring practice that appears neutral but disproportionately affects protected groups.

Related Lessons

A business case draws on work done elsewhere in the program, and these lessons supply the inputs it depends on.

• Stakeholder Mapping: Who Needs to Be Involved? is the direct antidote to the missing-key-stakeholder anti-pattern, since knowing who actually decides is what tells you which argument to lead with.
• Metrics and Monitoring: Tracking Efficiency, Quality, and Fairness provides the baseline numbers this case is built from, and it is also what lets you replace projections with evidence after the pilot.
• Vendor and Tool Selection -- Evaluating AI Solutions turns the compliance constraints in the risk section into shortlist criteria, so a vendor that cannot support an audit never reaches the business case at all.
• Legal and Compliance Partnerships: Ensuring AI Use Is Defensible is how the risk argument gets its credibility, because a compliance plan co-owned with counsel reads very differently from one written at counsel.
• Piloting and Iteration: Testing Workflows and Gathering Feedback is the mechanism behind the phased ask, and it is where the go or no-go gate that makes a staged commitment approvable actually gets designed.

Closing

A strong business case is not persuasion; it is clarity. It shows stakeholders what you are proposing, what benefits you expect, what risks you have identified, and how you intend to manage them, and the good ones address several stakeholder concerns at once rather than assuming everyone shares yours. Build it carefully, because it is the foundation on which your organization decides whether to commit to AI recruiting at all.

Frequently Asked Questions

How do I answer the CFO who says the savings are just soft, recovered hours, not real dollars? Devon converts recovered hours into a decision: either the hours backfill work the team is currently not getting to, such as sourcing and candidate experience, or they let the team absorb growth without adding headcount. He quantifies the avoided next hire or the work that moves from neglected to done, and he shows the efficiency floor from sensitivity analysis so the CFO sees the number holds even under pessimistic assumptions. Soft savings become credible when tied to a specific tradeoff the CFO recognizes.

What if the bias audit finds the tool actually performs worse on fairness than our current process? That is exactly why the pilot and the audit checkpoint exist before any wider rollout. Devon designed the case so a failing DIR is a stop signal, not a sunk cost; the go or no-go gate means the company can walk away having spent only the pilot budget. He would rather find the problem in a controlled 90-day test on two job families than discover it after full deployment, and being able to walk away is itself part of what makes the case defensible to legal.

Does NYC Local Law 144 mean we cannot use these tools at all? No. It means any automated employment decision tool needs an independent bias audit within the prior year, the results posted publicly, and candidates notified in advance. Devon treats these as build requirements with named owners and dates, not as reasons to abandon the project. A vendor that cannot support an audit or accommodation under ADA simply does not make his shortlist, which turns compliance into a selection criterion rather than an afterthought.

How conservative should my projections actually be? Conservative enough that the case still wins on the pessimistic case. Devon presents projections as estimates with visible assumptions, sources improvement figures from pilot data or vendor benchmarks rather than hope, and shows a range instead of a single flattering point. The credibility cost of one over-promise that misses is higher than the upside of a bigger headline number, so he anchors stakeholders on the floor and lets the actual results exceed it.

Key Takeaways

• A complete case spans four dimensions. Efficiency, quality, fairness, and risk; an efficiency-only case answers one stakeholder and stalls in front of the rest.
• Tailor the argument to the audience. The CFO wants ROI, legal wants risk mitigation, the DEI lead wants the diversity impact; one argument cannot carry all three.
• Build on evidence, not hope. Use pilot data, vendor benchmarks, or research, and present projections as estimates with their assumptions visible.
• Frame fairness with the four-fifths rule. A DIR below 0.80 is both a fairness problem and legal exposure; showing the projected move toward 0.85 quantifies the value.
• Show the net on risk. Name the risks AI reduces and the new risks it creates, and demonstrate the new ones are mitigated by audit, testing, and human review.
• Find the decision-makers first. Address their specific concerns rather than assuming everyone cares about efficiency, or you get approval without the resources to deliver.