What's really happening with AI adoption: Insights from Bullhorn Engage 2026 RecToks

Plot twist: You don't have an AI problem. You have a data problem

by Matt Brawn, EVP, Regional Director Americas at Kyloe Partners


On the first morning of Bullhorn Engage Boston 2026, we did something simple. We got 25 recruitment leaders into a room, asked them how AI was really going, and then got out of the way. 

What followed was the most honest conversation we'd had in ages. Nobody was selling anything or chasing headlines. Just people being real about what's working, what's flopping, and where they're still stuck.


The plot twist? Almost nobody in that room had an AI problem. They had a data problem - they just hadn't called it that yet.

What we heard at RecTok Boston 2026

This wasn’t a room of AI sceptics, quite the opposite. Most attendees were already actively using AI to help with different processes such as sourcing, screening, and outreach. But several openly admitted they moved too fast: “We made one mistake.” They had to pause their AI rollout entirely and go back to focus on a dedicated data quality project with Kyloe before seeing real value. That pattern came up again and again:
  1. AI adoption is high
  2. Confidence in outputs is mixed
  3. Data quality is the limiting factor

And the gap is widely recognised. As The Global Recruiter said, while 90% of organisations say data is their cornerstone, 76% still don’t have a clear data strategy in place.

 


When your data breaks your AI

The most revealing conversations came when we dug into what’s happening inside Bullhorn environments.

One multi-state operator shared they’re seeing around 30% duplicate records every week from sourcing tools like Indeed, Vivian, and LinkedIn. In some cases, duplicates are even created intentionally for contractor workflows, adding another layer of complexity that breaks automation and enrichment.

Others highlighted how stale data quietly undermines AI performance. A healthcare-focused team shared they apply a 36-month recency rule to nursing records, because anything older introduces noise rather than value.

The takeaway is hard to ignore: AI is only as good as the data it’s working with, and most databases aren’t structured to support it.

Why tools (not intentions) matter

Everyone in the room knew data quality mattered. But very few had a scalable way to manage it. Manual processes, one-off clean-ups, and “we’ll fix it later” approaches don’t hold up when you’re dealing with:
  1. High-volume data ingestion from multiple sources
  2. Duplicate-heavy workflows
  3. Inconsistent field usage across teams
  4. Ongoing enrichment needs

What good looks like

There were also strong examples of what successful AI adoption looks like when the right foundations are in place.

One team using AI screening reported a 36% candidate completion rate, 88 hours saved in six weeks, and a 4.48 out of 5 candidate satisfaction score.

Another shared how they have built trust with candidates by clearly disclosing when AI is involved, even giving their screening agent a name and introducing a human handoff at the right moment.

But even here, success was not just about the AI itself. It depended on clean data, clear workflows, and thoughtful implementation.

The next challenge

Beyond data, another theme stood out: AI governance is still catching up.

Most organizations admitted they have not fully audited their AI transparency practices. Compliance approaches vary by geography, and many teams are still working out what good looks like in practice.

What resonated most was risk, and the possibility of class action exposure, especially in sectors like healthcare where scrutiny is increasing.

This is quickly becoming the next frontier: not just making AI work, but making it accountable.

Where this is heading

The roundtable RecTok Boston 2026 confirmed what many recruitment teams are now experiencing first-hand: AI adoption is not a challenge anymore. Readiness is. And readiness comes down to three things:

  • Clean, structured, actively managed data.
  • The ability to control and target enrichment, rather than simply switching it on.
  • Clear, defensible approaches to AI transparency.

This is not a future problem. It is already shaping who is getting value from AI and who is still stuck troubleshooting it. The teams pulling ahead are not the ones with the most advanced tools. They are the ones who fixed their data first.


Want to know what your data actually looks like and how to fix it?

 
Get in touch