Stop Exporting to Spreadsheets: How AI Helps HR Leaders See the Whole Story in Their Data

The question usually arrives in an executive meeting. How many people left engineering this quarter, which teams scored lowest on the last survey, whether hiring is on track. The data exists, spread across an HRIS, an applicant tracking system, a survey tool, and a stack of case files, and someone will need a week to put it together.

By then the number is old. AI assistants that connect directly to those systems are changing the work of people analytics for an HR team, from building the report to asking the question.

On October 22, Rebecca Taylor of AllVoices spoke with Gregory Leyne, chief people officer at GitGuardian, a French cybersecurity company with offices in Paris and New York that finds leaked credentials such as API keys. Leyne has spent less than a year connecting Claude to his HR stack, and he shared his screen to show what he built, what it changed, and where it still gets things wrong.

Rebecca opened with a poll on whether attendees had connected an HR system to an AI tool. In the final count, 54% had not, 33% had, and 13% were not sure. One attendee answered no, but wanted to.

Why do HR teams still export data to spreadsheets?

Because the systems that hold people data were seldom built to report on it, and HR is rarely the team with the budget to fix that. Leyne said as much at the start of the session.

I’ve been very frustrated for years and years because our tools are usually not the best, because we are very often not the department with the biggest budget. - Gregory Leyne, chief people officer at GitGuardian

His HRIS splits employee records across five modules. To see one person in full, he had to open all five, and what he got was a snapshot of that moment, not a trend. It could not forecast turnover or retention, which are the numbers his CEO wants, usually right away.

Survey results told the same story. The survey tool exported a sheet of raw numbers, and the formulas, pivot tables, and slides for the executive committee fell to Leyne and one colleague.

From the end of the survey to the moment we deliver something to the executive committee with the workload HR has usually took me a month, a month and a half, which means my data was already really outdated. - Gregory Leyne

Rebecca, who started her career in HR recruiting, named the cost underneath the delay. The time spent stitching data together is time in which something goes unnoticed. When a report lands six weeks after a survey closes, Leyne said, any urgent problem in it is already late to fix.

What can an HR leader build with Claude without writing code?

A working dashboard, built by describing what you want in plain language and refining it one request at a time. Leyne started with no connections at all, only a spreadsheet of raw survey results.

He gave the file to Claude and asked for an HTML page, which he described as a mini website of the kind people built in the early days of the internet. The first version showed the survey overview. Then he asked for a heat map, then a heat map by team, by age group, by gender, by region, and by remote or hybrid status, all on anonymized data. One filter showed that his most senior employees were noticeably less happy than everyone else.

What once took a month now takes about 10 seconds after a survey closes. Leyne thinks the skill it calls on is one HR already has.

I think it’s even giving us HR an advantage, because I believe we are very good in communication. And so when you just have to talk to AI, it’s going to create things for you. - Gregory Leyne

His second build connected to the HRIS. The demo used a mock version with fake names and numbers, which he had Claude generate, but the structure matched his real one: headcount by region, turnover, gender split, remote status, and upcoming hires and departures with a forecast, filterable to engineering, to the US, or to contractors.

Each refinement was one sentence. Four people were joining next month and he could not remember who they were, so he asked for a pop-up that lists them when he clicks the bar. He wanted the list sorted by name or date, so he asked for sortable columns. French law is strict about probationary periods, so he added a view that keeps every one of those deadlines in front of him.

Rebecca pointed out that nobody starts at this point. The dashboard grew out of curiosity, one question after another, and Leyne agreed.

The more questions you ask, the more answer you get, and the more questions are coming to your mind, and you want to drill down in everything. - Gregory Leyne

Why is MCP safer than an API key for HR data?

Because an MCP connection grants scoped, agreed access, while an API key opens everything it can reach to whoever holds it. Leyne has run both.

The Model Context Protocol is an open standard for connecting AI applications to outside systems. Rebecca kept the technical detail out of the session on purpose. What matters to HR is that the AI connects to the systems where the data already lives, instead of the data being copied into the AI.

Leyne set two limits. His dashboards are not published anywhere; they live on his laptop, and sharing one means sending the file to a colleague. The second limit was on what the AI could change.

The first thing I’ve done is I have not authorized AI to write on my software. - Gregory Leyne

His HRIS has no MCP connection, so to build his people dashboard he taught himself automation in n8n, with Claude as his teacher. Tools such as Zapier and Make do the same job. Each automation pulls data through a webhook, authenticated with API keys stored on his laptop.

The problem is that those keys are stored on my laptop. And that is called, because that’s my company business, that is called a secret. And that’s very dangerous, because if anyone accesses it, it can access all the data I have. - Gregory Leyne

Guarding secrets is GitGuardian’s business, and his own security team recently told him his laptop held a lot of keys that could be stolen. His applicant tracking system, by contrast, offers an MCP connection. The difference he described between the two comes down to four things.

Automation with API keys vs. an MCP connection

Built with automation and API keysConnected through MCP
Hours of setup, fixing broken steps by handA connection approved in about five clicks
Keys stored on a laptop, exposed to anyone who gets inA connection the AI tool and the software agree on, field by field
The key sends whatever it was set up to sendYou choose who can do what through the connection
Only the fields the automation pullsDetail down to the interview transcripts inside the software

GitGuardian’s security team has since set up a secure tool for sharing the dashboards beyond Leyne’s laptop. Rebecca’s advice was the same: whatever you build, build it in partnership with your security team.

Who sees what when AI is connected to employee relations data?

Only what each person’s role already allows, when the connection is built that way. That matters most for employee relations data, which Rebecca argued is the part of the HR picture most often left out.

Employee relations data is usually the missing piece, at least in our experience here at AllVoices. And some of the most important stories really only appear once it’s included. - Rebecca Taylor, head of community and content at AllVoices

Turnover and survey scores describe what happened. Case and complaint records often explain why, and without them the story an HR leader tells the executive team has a gap in the middle.

ER data is also highly confidential, which is why Rebecca described how AllVoices designed its own MCP connection. A user’s AllVoices account is tied to the same email as the company’s enterprise Claude, ChatGPT, or Copilot account, so the AI tool answers only within the access that user already has.

If you’re not in HR and you’re not supposed to know anything about an investigation and you say, has anyone ever complained about my manager? You’re not going to find out about that if that’s not information that you’re supposed to have. - Rebecca Taylor

Leyne named the same advantage from his side: the MCP connection lets him choose who can do what, which his API key automations never could. For the settings HR and IT should agree on before anything connects, an earlier AllVoices session covered the guardrails for AI and employee data.

Scattered records are the other problem. Parts of an investigation often begin in an email thread or a Slack conversation. Rebecca pointed out that with those tools connected, an assistant can find the relevant pieces and help move them into the right record, the same way a recruiter can ask it to attach an emailed resume to the right job.

What can AI find across HR systems that one report cannot?

Where work is stuck, which is usually in the gaps between systems and between people. Leyne’s clearest example came from the weeks before the session.

Hiring for US account executives had been slow for two months with nothing to show for it. Once his applicant tracking system had an MCP connection, he gave Claude a standing brief:

  1. Review every candidate at the business case and manager interview stages.
  2. Check the timeline, including how long the team took to give feedback.
  3. Flag any candidate who is stuck.
  4. Read the interview transcripts.
  5. Send a full report to Slack every two days.

When a candidate stalls or a question goes unanswered, he can now step in with the hiring manager or his own team right away. His old automations could not.

Only an MCP can give me this level of details and understanding of what’s inside the software. - Gregory Leyne

The same approach reaches past HR. When Leyne asked the RevOps team for sales data, a team member said it would have to wait a week for their manager to return. Leyne connected Claude to HubSpot and had his report anyway. Rebecca noted the quieter benefit: nobody has to justify to another department why HR wants to know.

Questions like these used to need an analyst. Rebecca’s example was asking how many top performers might be at risk of leaving because of where their stock options stand. Leyne’s own wish list starts with retention, knowing who might leave before they tell a manager. A broader look at how AI is changing HR data strategy makes the same case for moving from lagging reports to decisions made in time.

How do you catch AI mistakes in HR reporting?

By checking every number against what you already know, and treating the AI as a fast first draft rather than the source of truth. Leyne said he still catches errors often.

One dashboard showed no hires for the month when he knew of three. When he challenged it, Claude agreed it had stored the figure as a fixed value and never updated it. His rule is to stay vigilant, and he said it gets better the more you use it.

Rebecca’s rule is one HR already knows.

You can trust the answer that it gives, but it never hurts to verify. - Rebecca Taylor

In practice that means asking the AI tool how many cases have been open longer than 30 days, then opening the case management system to confirm the count. The answer is only as good as the data behind it, and HR data is people data, which is never perfectly clean.

Some of what matters is not in any system. Leyne’s team transcribes its conversations, so he can ask what was discussed in the last meeting. He was candid that AI still cannot read emotion, a choking voice, or the look in someone’s eyes, and that a decision improves when the two are combined.

And so if you complete both together, you can start to have a way better understanding of your situation and draw more context and take better decisions. - Gregory Leyne

What should HR do with the time AI gives back?

Spend it with people, which Leyne argued is the work HR exists to do. When he moved first, his team worried about what AI would do to their jobs. His answer was that it takes away the work nobody misses.

AI is going to remove painful, unuseful, non-value added job. And you will be able to stand up and go see and talk to people, which is our job. - Gregory Leyne

His team now spends less time on presentations, data, and typing, and he sends them out to talk to colleagues. He sees the need. AI, he said, is creating mental health strain for developers, salespeople, and marketers, and supporting them is work only people can do.

Rebecca recalled a conversation with Brandon, then head of people at Zapier, who asked his leadership team and employees what they were opening up time to do. At Zapier the answer included external courses and more time for deep work.

In HR, it’s very easy to stay busy, but this way it kind of reframes the concept of what busy is. - Rebecca Taylor

The next step for Leyne is giving AI write access in 2027, so that his HR operations team can change a job title or set up a new hire’s onboarding from a spreadsheet by asking, instead of clicking through ten screens. Rebecca flagged the condition. A write action, such as reassigning an investigator from one case to another, is only right when someone who knows the context has confirmed it.

Start with the question you wish you could ask your data

Rebecca’s advice to attendees who had not connected anything yet was to skip the grand plan. Write down two or three questions you wish you could ask your data in plain language, about your people, your role, or your department.

Think to yourself, what are some of the questions that I wish that I could easily ask about something about my people or about my role or my department? Because that’s the best starting point for you to figure out how to start to connect some of this stuff. - Rebecca Taylor

In her own HR years, those questions were about promotions and problem areas: what else to know before a promotion decision, and where complaints, understaffing, or poor performance cluster. For an employee relations team, many of those answers sit in case data, and AllVoices turns employee relations cases into early warning data by manager, location, and issue type.

Leyne closed with a story from an executive committee meeting. The sales leader was pushing back on the CEO, insisting something could not be done, and Leyne walked through how to connect the systems to do it. The CEO’s response was that this was HR talking.

If you go into it, you will win a lot of credibility by mastering those tools. - Gregory Leyne

HR has spent years waiting on reports. The leaders who learn to ask the question themselves stop waiting.

Frequently asked questions

What is MCP in HR?

A standard way for an AI assistant to connect to the software where your HR data already lives. Instead of exporting files into a chatbot, you approve a connection, and the assistant reads from that system within the access the connection allows. Its documentation calls it an open-source standard for connecting AI applications to external systems.

How long does it take to connect an HR tool to Claude?

A few minutes, once IT approves it. Leyne confirmed it is about a five-click process: find the platform in your AI tool’s list of available connections and authorize it. Many companies require an administrator to approve each new connection first, so ask your IT or security team before you start.

Can AI predict which employees will quit?

Not from what this session showed. Leyne named early warning on retention as the question he most wants AI to answer, so HR can act before an employee resigns. His dashboards today forecast the hires and departures already recorded in his HRIS, which is a step toward that, not the prediction itself.

Is it safe to paste employee data into ChatGPT?

Rarely, and Rebecca called it a workaround that trades one problem for a worse one. Pasting records into a chatbot moves sensitive data out of the system that controls access to it. A connection that reads the data in place, within each user’s permissions, keeps that control, though your security team should still approve it.

Should AI tools have write access to an HRIS?

Not at first. Leyne has kept his connections read-only so far, and plans to allow writes in 2027 for routine changes such as job titles and onboarding records. Any write action still needs a person who knows the context to confirm it is correct before it counts.

Can employees use AI to see complaints about their manager?

Not when the connection respects permissions. In the AllVoices MCP connection, the account is tied to the same email as the company’s AI tool, so the assistant answers only within that user’s role. An employee outside HR who asks whether anyone has complained about their manager learns nothing they are not cleared to see.

Stay up to date on Employee Relations news

Sign up to our newsletter

Thank you! We look forward to meeting you soon
Oops! Something went wrong while submitting the form. Please try again or use the email below to get support.
Join our newsletter for updates. Read our Terms