AI Employee Interviews at Scale: How to Hear From Every Team in Days

The short answer

AI employee interviews at scale are adaptive, role-specific conversations that an AI agent holds with hundreds or thousands of employees at the same time, then synthesizes into process maps, pain points and ranked improvement opportunities. Horizon runs them for enterprises: Mercado Libre interviewed 2,000 employees across five countries in four days, reaching 100% of the functions in scope.

The method answers a question that surveys and workshops struggle with at enterprise size: how does work actually get done, team by team, including the exceptions and workarounds that never reach a dashboard. Each employee answers on their own time, the interview follows up on what they say, and the synthesis groups what many people described into a small number of findings leaders can act on.

Key takeaways

• AI employee interviews combine the depth of a one-to-one interview with the reach of a survey.
• The interview is designed per role from the organization's own context, and it asks follow-up questions based on each answer.
• The output is a set of process maps, quantified pain points and ranked opportunities, each traceable to what employees said.
• Trust decides the quality of the evidence: consent, how answers are reported and how the data is used need to be settled before launch.
• A first round on one process can run in a week.

Why do surveys and workshops miss how work really gets done?

Leaders who want to understand how work runs usually have three options, and each one breaks at a different point.

Surveys reach everyone and stay shallow. A fixed questionnaire cannot ask "why" when someone mentions a workaround, so the answer that would explain a delay stops at a rating. Response quality drops as surveys get longer, and open-text fields arrive in volumes nobody has time to read.

Workshops go deep with a few people. A good facilitator can surface rich detail in a room of eight. The sample is the problem: the people in the room describe the process from where they sit, senior voices shape the consensus, and the frontline staff who handle the exceptions are often missing. Mercado Libre's manual discovery model, before it changed approach, covered under 1% of the workforce and took more than 100 coordination hours per initiative.

Consultant interviews sit in between, and they cost time. One-to-one interviews produce the best evidence of the three, and a consulting team can run only so many per week. By the time the last interview is synthesized, the first one may already describe a process that has changed.

The common gap is coverage with depth. Leaders need to hear from enough people to see patterns across teams, regions and roles, with enough follow-up to understand why each pattern exists.

How do AI employee interviews work?

The workflow has four stages, and each one has a specific job.

1. Context in

The organization defines the question it needs answered and the scope: which process, which functions, which regions and which roles. It adds whatever context already exists, such as org charts, process documents, policies and known pain points. This is what keeps the interview from feeling generic.

2. An interview designed per role

From that context, the AI designs a different interview for each role. A procurement analyst gets asked about intake, approvals and supplier data. A finance manager gets asked about exceptions, sign-off rules and month-end pressure. The questions use the organization's own vocabulary for its systems and teams.

3. Adaptive follow-up

Each interview runs as a conversation. When an employee mentions a spreadsheet, a manual check or an informal approval, the agent asks what it is for, how often it happens and what would happen without it. Employees answer asynchronously, on their own schedule, so hundreds of conversations can run in parallel without blocking calendars.

4. Synthesis

The answers are grouped by process, system and cause. Several people describing the same delay in different words become one finding with a count behind it. Each finding links back to the conversations that support it, which is what lets leaders defend it in a steering meeting.

On Horizon, this runs through Discovery Cycles, which hold the interviews across functions and geographies at once. The Insights Dashboard ranks findings by impact and effort, the Process Library turns the conversations into structured process documentation, and the Initiatives Dashboard converts priorities into business cases with owners.

How do AI interviews compare with surveys, workshops and process mining?

Each method reads a different layer of the organization. The table shows where each one is strong and what it leaves out.

The methods combine well. Process mining shows where cases wait inside a system, and interviews explain why they wait. In the Mercado Libre engagement, interview transcripts were fused with data from SAP, ARIBA, PIC, BigQuery and Tableau to expose value gaps that neither source showed alone.

For a fuller map of the methods, see the business process discovery guide and the complete guide to organizational discovery.

How do you design interviews employees trust?

The quality of every finding depends on how candid people are, and people are candid when they understand three things.

Why they are being asked. Explain the business question in plain words before the interview starts, and say who will see the results. An employee who knows the goal is to remove a duplicated approval answers differently from one who suspects a headcount review.

How their answers are reported. Decide before launch whether findings are shown only in aggregate, whether quotes are attributed, and what the minimum group size is for any breakdown. Then say it in the invitation, and keep to it.

How the data is used and protected. Tell people where the conversations are stored, who can access them and how long they are kept. Horizon publishes its security and compliance posture in its Trust Center, and the employee consent guide covers consent design in detail.

The fear of job loss deserves a direct answer. When the purpose is to remove friction from the work, say so, and describe what happens to the time that gets freed. Programs that skip this conversation get polite, short answers that describe the official process.

When these conditions are in place, people engage. At La Segunda, all nine people on a team handling more than 700 active cases took part, rated the experience 8.2 out of 10, and 90% said they would take part again.

What do you get after the interviews?

A round of AI interviews produces four outputs that a transformation team can use directly:

• Process maps of how the work runs today, including the variants by team or region and the exception paths.
• Quantified pain points, each with the number of people who described it, the hours involved and the process step it affects.
• A ranked list of opportunities, scored by impact and effort, with the evidence behind each one.
• Business cases and owners for the initiatives the organization decides to fund.

At Mercado Libre, the company was carrying more than 600 initiatives, many of them overlapping. After interviewing 2,000 employees in four days, cross-country benchmarking surfaced duplications and best practices that were invisible from inside any single market. The portfolio resolved to 24 initiatives with ROI, effort and a roadmap attached, producing $2.3M in projected annual savings from 12,000 to 13,000 hours automated and 76 FTE equivalents avoided. The first implementations went live in under a week.

That is one engagement under specific conditions, and it shows what the outputs look like when interviews reach every function in scope.

How do you run your first round in one week?

A first round works best on one process that leaders already suspect is losing time. Five steps cover it:

1. Day 1: pick the process and the question. Name the business result you want to improve and the roles that touch the process.
2. Day 1: share the context. Org chart for the scope, existing process documents and the pain points people already talk about.
3. Day 2: invite the participants. Send the invitation with the purpose, how answers are reported and how data is protected.
4. Days 2 to 5: run the interviews. Employees answer asynchronously while the synthesis builds as conversations come in.
5. Day 5: review the findings. Walk through the process map and the ranked opportunities with the process owner, and pick the first initiative.

Hear from every person in one process within a week. Run Horizon's one-week challenge: AI interviews with the people who do the work, and a ranked list of what to fix first. Start the one-week challenge

Where Horizon fits

Horizon is an AI-powered organizational discovery platform, and AI employee interviews are its core. The platform designs role-specific interviews from each organization's context, runs them in parallel across functions and countries, and turns the conversations into process maps, ranked opportunities and initiatives with owners.

It fits best when the question spans several teams and the evidence lives with the people doing the work: a process that varies by region, an initiative portfolio that has grown faster than anyone can review, or a system migration that needs an accurate picture of the current state. For a single team that already agrees on how its process runs, a well-facilitated workshop may be enough.

Learn more about the platform on the product page.

Checklist before you launch

1. Can you state the business question in one sentence?
2. Have you named every role that touches the process, including approvers and downstream teams?
3. Have you shared existing documents and known pain points as context?
4. Have you decided how answers will be reported, and written it into the invitation?
5. Have you told participants how their data is stored and protected?
6. Is a process owner ready to review the findings and pick the first initiative?

Question 6 is the one that most often goes unanswered. Findings without an owner stay in a deck.

FAQ

How many employees can you interview with AI?

As many as the scope requires, since the interviews run in parallel. Horizon's published deployments range from a single team, as with the nine people at La Segunda, to 2,000 employees across five countries in four days at Mercado Libre.

Are AI employee interviews anonymous?

That depends on how the round is configured, and it should be decided before launch. Teams usually choose between aggregate-only reporting and attributed quotes with a minimum group size for any breakdown. The invitation should state which one applies. Horizon's security posture is published in its Trust Center.

How is an AI interview different from an engagement survey?

An engagement survey measures how people feel about their work with fixed questions. An AI interview asks how the work gets done, follows up on each answer and produces process maps and improvement opportunities. The two answer different questions and can run side by side.

How long does an AI interview round take?

A first round on one process can run in a week. Mercado Libre completed 2,000 interviews in four days, and synthesis builds while the conversations are still coming in.

Do employees need training to take part?

No. Participation is a conversation in plain language about one's own work, answered on the employee's schedule from a link.

Hearing from everyone changes the decision

Leaders make better decisions about work when the evidence comes from the people doing it, and AI interviews make that evidence available at the size of the organization. The interview reaches every role in scope, the follow-up captures the reasons, and the synthesis turns thousands of answers into a short list of things to fix.

See it. Fix it. Own it.