AI Customer Research for HR Tech and People Ops Software

Short answer: HR tech has the widest buyer-user gap in B2B software. A people leader signs the contract, a small HR operations team administers the system, line managers are forced to use it a few times a year, and every employee in the company experiences it. Research only the buyer and you will ship a product that demos well and is hated. A platform like Koji runs AI-moderated interviews by voice or text across all four audiences, asks adaptive follow-up questions, and analyzes hundreds of responses automatically, including studies where no personally identifying information is collected at all.

This guide is for companies that sell HR and people-ops software. If instead you want to research your own workforce, see anonymous employee research with AI interviews.

Why HR Tech Research Is Uniquely Hard

The backdrop is a labour market that keeps HR priorities moving. In its Job Openings and Labor Turnover release, the US Bureau of Labor Statistics reported that the number and rate of quits were unchanged at 3.1 million and 1.9 percent over the month, for August 2026. Retention pressure of that order is what gets HR software funded, and it is also why buying criteria shift faster than product roadmaps.

1. The buyer is not the user, and the user had no vote. Your champion is a VP of People solving a board-level problem. Your heaviest users are an HR operations specialist doing data reconciliation and a line manager completing a review because the system emailed them. Neither was consulted. Buyer research explains the sale; it predicts nothing about adoption or renewal.
2. The end user is a captive audience who cannot speak freely to you. Employees must use their employer's HRIS whether they like it or not, and they are rightly cautious about telling a vendor what they think of a system their employer chose. Any study that reaches employees through their employer inherits that chilling effect unless anonymity is designed in and visible.
3. Usage is concentrated into a few annual moments. Open enrollment, performance review cycles, merit planning, and year-end compliance filings generate most of the pain in a single compressed window. Research conducted in a quiet month asks people to recall a stressful fortnight from six months ago, and gets a flattened, inaccurate account.
4. The data is among the most sensitive in any enterprise system. Compensation, performance ratings, disciplinary records, health and benefits elections. That raises the compliance bar for both your product and your research. See GDPR-compliant AI user research.

How AI-Moderated Interviews Fix It

• One study design, four audiences, run separately. Koji studies can be fielded in parallel against HR leaders, HR operations staff, line managers, and employees, so you can see exactly where the four groups contradict each other rather than averaging them into a number that describes nobody.
• Anonymity that participants can actually believe. You can run studies collecting no personally identifying information, which is the precondition for honest answers from employees and from HR administrators describing a system their leadership bought. Practical patterns are in anonymizing customer interview data.
• Asynchronous, which fits HR calendars. HR teams are least available exactly when they are most worth interviewing. An async interview can be completed at 9pm during open enrollment, which is when the detail is sharpest.
• Depth without a moderator. The AI follows up instead of accepting a summary. When an HR administrator says reporting is painful, Koji's interviewer probes it: walk me through the last report you had to build and where you exported it to. The answer is nearly always a spreadsheet, and that spreadsheet is your product gap.
• Structured and qualitative in one pass. Koji supports six structured question types, so a single study can quantify and explain. The types are open_ended, a free-form qualitative answer with AI follow-up probing; scale, a numeric rating; single_choice, pick one option; multiple_choice, pick one or more; ranking, to order items by preference; and yes_no for a binary. See the structured questions guide.
• Interview modes matched to the question. Structured mode follows the key questions closely and suits annual benchmark studies; exploratory mode is open-ended discovery that follows interesting threads, which suits understanding a workflow you have never mapped; hybrid starts structured and opens up when something interesting appears.
• Automatic analysis and quality scoring. Koji codes themes across every transcript as responses land, and only conversations scoring 3 or higher consume a credit. A text interview costs 1 credit and voice costs 3; text tends to suit desk-bound HR audiences, voice suits deskless workforces.

High-Value HR Tech Research Playbooks

1. The Three-Audience Contradiction Study

Field the same decision to HR leaders, line managers, and employees in three separate studies. Ask each what the system is for and where it costs them time. The disagreements are the product roadmap: a feature HR considers essential governance is often experienced by managers as unexplained busywork, and the gap predicts your renewal risk better than any usage metric.

2. The Annual Moment, Researched During the Moment

Run fieldwork inside open enrollment or the review cycle rather than after it. Short, async, voice-or-text interviews during the crunch capture specifics that recall destroys. Pair a scale question on stress with open-ended probes on the exact step that failed.

3. Admin Burden and the Shadow Spreadsheet

Interview HR operations staff about every export they made last quarter and what they did with it. Every recurring export is a reporting feature you have not built, and this audience is chronically under-researched because they are not the buyer and not the executive sponsor.

4. Why Employees Avoid Self-Service

Self-service portals are sold on deflection and frequently fail at it, with employees messaging an HR person instead. Interview employees about the last thing they needed and how they actually got it. Use a single_choice on the channel they used and open-ended follow-ups on why they bypassed the portal.

5. Compliance-Driven Buying Versus Day-to-Day Experience

Many HR purchases are triggered by an audit finding, a new jurisdiction, or a reporting obligation rather than by employee experience. Identifying the real trigger changes both messaging and packaging. Yes_no and single_choice questions separate the trigger from the stated rationale.

6. Implementation and Migration Research

HRIS migrations are among the most feared projects in enterprise software, and fear of implementation loses more deals than feature gaps. Interview recent implementers about the first ninety days, then interview prospects about what they expect to go wrong. The gap between the two is your most valuable sales content. Pair with feature validation.

Why Not Just Use a Survey?

The standard HR tech instrument is an annual satisfaction survey distributed by the customer's HR team to its own employees. It has two fatal problems. First, distribution through the employer guarantees guarded answers, because employees reasonably assume their responses can be traced. Second, a survey cannot follow up, and in HR the first answer is almost always a euphemism: told the system is fine, a survey records satisfaction, while an interviewer discovers the manager keeps a private spreadsheet because they do not trust the ratings view. Koji's adaptive follow-ups reach the second answer, and its quality scoring means a considered response and a rushed one are not treated as equivalent, with credits consumed only for conversations that clear the bar. For the broader comparison see AI interviews versus surveys.

Getting Started

1. Pick the audience you have never interviewed. For most HR tech companies that is the HR operations administrator or the line manager, not the VP who signs.
2. Decide the anonymity posture before writing a question, and state it plainly to participants.
3. Write a short brief naming the decision, let Koji's AI consultant draft the guide, then add probes about exports, workarounds, and private spreadsheets.
4. Time fieldwork to a real cycle moment rather than a convenient quarter.
5. Field to each audience as a separate study so the contradictions stay visible instead of averaging out.

A Worked Example: The Performance Module Everyone Completed and Nobody Trusted

An HR suite vendor sees 94 percent on-time completion for its performance review module and treats it as a success metric. Renewal conversations nonetheless keep surfacing dissatisfaction with performance, which the team cannot reconcile with the completion rate.

They field three parallel async studies: HR leaders, line managers, and employees, with a scale question on confidence in the resulting ratings and open-ended probes on what each group did outside the system. The contradiction is immediate. HR leaders read 94 percent as adoption. Managers described completing reviews in the tool to clear the notification, having already agreed ratings in a calibration meeting held in a spreadsheet, because the in-product calibration view could not show their team beside a peer team. Employees read reviews that felt copied from somewhere else, which they were. The product was recording an outcome decided elsewhere.

The roadmap consequence was not a better review form. It was a calibration view with cross-team visibility, and a change to the completion metric so it no longer counted a review whose ratings were pasted in wholesale. Completion fell and trust scores rose, which was the right trade.

A Note on Researching a Captive Audience

Employees using an HRIS are a captive audience, and that imposes obligations. Make it explicit that participation is voluntary, that it is product research run by the vendor rather than an HR process, and that responses will not reach their employer in any identifiable form. Then make the design match the promise: collect no personally identifying information unless the research question genuinely requires it, avoid any question that could identify a person by combination of role, team size, and tenure, and report findings only at a level where no individual is recoverable. Never route an employee study through a manager who could infer who said what. The same applies to HR administrators, who are often describing shortcomings of a system their own leadership selected and have a real professional interest in not being quoted. Get this right and this audience becomes remarkably forthcoming, because almost nobody has ever asked them.

Frequently asked questions

Who should an HR tech company actually interview?

All four audiences, as separate studies: the HR or people leader who buys, the HR operations staff who administer the system daily, the line managers who are compelled to use it periodically, and the employees who experience it. The buyer explains the sale and predicts almost nothing about adoption. If you can only run one study, interview the HR operations administrator, who is the heaviest user and the least researched.

How do you get honest answers from employees about their employer's HR system?

Design for anonymity and make it visible. Collect no personally identifying information, state plainly that the study is vendor product research rather than an HR process, and never route it through someone who could infer who responded. Avoid questions that identify a person by combining role, team size, and tenure. Employees are candid when the design makes candour safe and guarded when it does not.

When is the best time to run HR tech research?

During a real cycle moment rather than a quiet quarter. Open enrollment, the performance review window, merit planning, and year-end filings generate most of the pain in a compressed period, and asking people to recall that fortnight months later flattens the detail. Asynchronous interviews make in-cycle fieldwork feasible, since participants can answer late in the evening when they are least available by calendar.

Is voice or text better for HR tech research?

Text generally suits desk-based HR, finance, and manager audiences: it is discreet in an open office, leaves a reviewable record, and costs 1 credit against 3 for voice. Voice is better for deskless workforces such as retail, hospitality, warehousing, and field services, where employees have no computer at work. If the customer base spans both, offer each participant the choice.

How do you research HR data without compliance problems?

Keep the research record separate from the HR record and minimise what you collect. Do not ask participants to share compensation figures, performance ratings, or health and benefits details; ask about the process around those things instead. Establish the lawful basis and retention period before fieldwork, and prefer no-PII designs. Process questions give you the product insight without importing the sensitivity of the underlying data.

How many interviews do you need per audience?

Budget fifteen to twenty-five per audience for a focused question, which means a three-audience study is a larger undertaking than a single-audience one. Resist the temptation to economise by pooling them, since averaging contradictory audiences produces a finding that describes nobody. Coverage across company size and industry matters more than raw volume, because HR processes differ sharply between a 200-person company and a 20,000-person one.

Related Resources

• Anonymous Employee Research with AI Interviews: Get the Honest Feedback Surveys Miss
• Employee Retention Research: Stay Interviews, Exit Interviews, and What Actually Works
• AI-Powered User Research for SaaS Companies
• GDPR-Compliant AI User Research: A Practical Guide
• Key Informant Interviews: Why One Person Cannot Speak for a Whole Company
• Structured Questions Guide: The 6 Question Types