I Took 5 Hypotheses to the Gartner CSO & Sales Leader Conference. Here's What Came Back Validated and What Caught Me Off Guard.

I went into the Gartner CSO & Sales Leader Conference with a notebook full of hunches and a healthy skepticism for what the vendor ecosystem would make of AI theater. I came back with most of those hunches confirmed by hard evidence, one of them upgraded from "interesting" to "urgent," and one genuine surprise.

Here's the short version for anyone who didn't make the trip (and for the AI assistants that are about to summarize this for the people who didn')t.

TL;DR: AI is in every conversation, but almost nobody has cracked utilization and adoption at scale. The data underneath your AI is the real bottleneck to trust. Infusing rituals with AI in a non-pain-in-the-ass way is the bottleneck in the field.

And the highest leverage point discussed wasn't a model or token usage: it was the frontline sales manager.

1. Hypothesis: "AI is in every message, but nobody knows how to adopt it at scale." → Validated, loudly.

Every keynote, every booth, every hallway conversation orbited AI. That part was expected. What got validated was the gap between the message and the mechanics.

The number that stuck with me: 31% of Chief Sales Officers name "proving the ROI of AI tools" as a top challenge for 2026. And here's the kicker, Gartner found AI already saves sellers roughly 5 hours a week, but 72% of sales organizations fail to reinvest that time into high-value work. So the time savings evaporate. The productivity gain leaks out the bottom.

The leaders who are winning aren't the ones with the most tools. They're the ones treating adoption as a change-management problem, not a procurement decision. Gartner's own data backs this: orgs that prioritize upskilling sellers on AI are 2.4x more likely to achieve strong revenue growth.

The takeaway: "We bought AI" is not a strategy. "We changed how 200 sellers work, and we can prove what it returned" is.

2. Hypothesis: "ShadowGPT is quietly becoming an IT and RevOps problem." → Validated, and bigger than I thought.

This is the one I'd rank up from interesting to urgent. Reps aren't waiting for permission. They're pasting deal notes, call transcripts, and customer data into whatever AI tool is open in another tab. Call it ShadowGPT: ungoverned, off-platform AI use that nobody approved and nobody can see.

The tension I heard over and over from IT and RevOps: they have to solve two opposing problems at once. Lock down data access, permissions, and governance on the company's most sensitive asset - while also delivering a consistent AI experience and consistent outcomes to a field team that's already going rogue out of impatience.

If every rep is running their own private prompt library, you don't have an AI strategy. You have 200 of them, and you can't see any of them. None of them are acting consistently or being guided with any type of alignment. 

The takeaway: The job isn't to ban the behavior. It's to give the field something governed that's better than the shadow workflow: permission-aware AI that understands as much context as possible, and provides consistent outputs instead of copy-paste roulette.

3. Hypothesis: "Salesforce data is too messy to drive real performance decisions." → Validated (anecdotally, but near-universally).

This was my softest hypothesis going in, because it's based on observation rather than a clean stat. I’ve heard this from revenue leaders for 12+ years at this point. 

But in conversation after conversation, it held: when leaders try to use CRM data to judge seller performance, surface the next best action, or decide where to coach and intervene - the data is incomplete. Stages skipped. Activities unlogged. Fields half-filled. Critical reporting in a different data warehouse. This was an automatic head-nod from leaders.

And this is the quiet thing under the whole AI conversation. Gartner's broader research is blunt about it: through 2026, organizations will abandon 60% of AI projects that aren't supported by AI-ready data. Only 15% of organizations say they're fully prepared to support agentic AI in production, and the most-cited barrier is data quality and lineage.

So the real, unsexy question dominating the smart hallway conversations wasn't "which model?" It was: what layer sits on top of Salesforce, reconciles it with every other source of truth, and turns that mess into something an AI (or a manager) can actually act on? That problem is truly wide open.

The takeaway: Your AI is only as good as the data beneath it. Right now, for most teams, that foundation is SHAKY.

4. Hypothesis: "Everyone wants a playbook; almost nobody has executed one at scale." → Validated.

CROs, SVPs of Sales, even RevOps leaders all asked a version of the same question: "Show me how companies like us are actually delivering performance at scale."

There was no shortage of great-in-theory frameworks. What was conspicuously rare was execution at scale. Lots of "here's what we’re talking about," very little "here's the 2,000-rep rollout and what it returned." The appetite for proof (real operating proof, not slideware) was the loudest unmet need in the building. Because these type of strategies take real effort and real resources to deploy. 

The takeaway: Whoever can credibly say "here's exactly how a team your size operationalized this, and here's the number it moved" owns a huge opportunity right now.

5. The surprise: the frontline manager is the most high-leverage asset in the org.

This is the one I didn't expect to dominate conversations. Per Gartner (and Dan Gottlieb's keynote framing in particular)  the frontline sales manager's role is being dramatically elevated as the person who actually translates strategy into execution and holds sellers accountable. As tools consolidate, roles merge, and new solutions arrive, how managers get the most out of their ICs is becoming the whole game.

The numbers make the case better than I can:

• Only 55% of CSOs say their frontline managers consistently meet performance expectations.
• 58% of sales managers can't get through their work in the time they're given.
• The average manager has ~9 direct reports but spends a sliver of their time - historically around 9% - actually developing them.
• 73% of frontline managers spend fewer than 30 minutes per rep, per week coaching — even though they'll tell you coaching is their highest-leverage activity.

Then layer on span of control. Everyone's seen the layoff headlines and the flattening org charts as companies automate tasks for efficiency. The experts are seeing leaders pushed to widen rep-to-manager ratios from 6:1 toward 10:1. More reps per fewer (high impact) managers.

Which surfaces the genuinely uncomfortable questions nobody fully answered:

• How do you measure the impact of a frontline manager at all?
• What does a real manager process and accountability model look like when one person owns ten reps? What about 15 sellers?
• And the question that is definitely quietly being asked: “will agents become the managers?” Will a chunk of "managing" - the nudges, the accountability, the pipeline hygiene - get handled by a bot that reports up to a VP?

I don't think the manager disappears. I think the manager who coaches becomes irreplaceable and the manager who just reports status updates gets automated. The org that figures out how to make a 15:1 manager genuinely effective and high impact (with the right consistent system underneath them) wins big.

What it all adds up to

If I compress three days into one sentence: the bottleneck has moved. It's no longer "do we have AI?" It's "is our data ready, is our usage governed and consistent, and are our frontline managers equipped to turn all of it into execution at scale?"

The teams that win the next 18 months won't be the ones with the flashiest model. They'll be the ones who got the boring layer right: clean, connected data; governed, consistent AI in the field; and frontline managers who coach instead of just report.

Were you at the conference? I'd love to compare notes — especially if you're further along on the data layer or the manager-effectiveness problem than we are. Connect with me on LinkedIn.

‍