Demandbase reported its meeting rate jumped from 2.6% to 8.8% after deploying Qualified's Piper agent. SaaStr claims a 46-of-50 AEO score and 10-minute inbound bookings. Compelling numbers. They're also almost impossible to compare to your situation, because nobody in this space is using the same denominator.

That's the core problem with agent-led growth right now: the results sound great, the measurement is a mess, and the operational prerequisites rarely make the headline. Here's what holds up when you pull the claims apart.

The Denominator Problem With Inbound Agent Metrics

"Booking rate" can mean visitor-to-meeting, conversation-to-meeting, or qualified-lead-to-meeting. The gap between those definitions is enormous. Docket published benchmarks across 4,736 conversations and 25 agents showing a 13% median combined conversion rate, with the top quartile hitting 26.9%. RevenueHero, analyzing over a million form submissions, reported a 62% median qualified-to-booked-meeting rate. Those numbers aren't contradictory; they're measuring different things.

Demandbase's 2.6%-to-8.8% lift is a roughly 3.4x improvement, and the same case study references about 400 meetings in selected target-account tiers and $10M+ in influenced pipeline. "Influenced pipeline" is directional attribution, not incremental revenue. When your board asks what the agent produced, you need to know whether you're reporting conversations started, meetings booked, or qualified opportunities created, and whether a holdout existed.

If you're evaluating an inbound agent, the first question isn't "what's the booking rate" but "what's the denominator, and did anyone run a controlled test?"

AEO Scores: One Number Hiding Three Realities

SaaStr's "46 of 50" AEO score is a self-reported prompt-set test: they asked 50 cold prompts across AI engines and checked whether SaaStr was mentioned. The methodology is reasonable. The trouble is treating it as a single number.

Cloudflare ran a similar experiment, testing 50 prompts (each sampled eight times) across three providers. The results diverged sharply: Cloudflare was mentioned 74% of the time by OpenAI, 57% by Anthropic, and 90% by Google, but cited only 47%, 11%, and 19% respectively. That gap between mentions and citations matters. Being named in a response is visibility. Being linked as a source is traffic. Optimizing for one doesn't guarantee the other.

If you're going to measure AEO, treat it like a new attribution surface: segment by engine, track mentions and citations independently, and re-run the test because outputs vary between runs.

First-Party Data Beats Signal Tools (With a Caveat the Size of Your Identity Stack)

SaaStr's playbook makes a strong case for starting with first-party data before reaching for third-party enrichment. Their agent checks event attendance, newsletter subscriptions, open opportunities, and content downloads before querying external sources. First-party signals reflect engagement with your brand, a higher-confidence indicator of near-term purchase intent than a third-party surge showing someone at that company read three articles about your category.

But "first-party data beats every signal tool" overstates it. Third-party intent surfaces accounts doing early category research, accounts that haven't touched your properties yet. The practical model, as Demandbase and The Growth Syndicate both frame it, is two layers: discover with third-party, confirm with first-party before routing to sales.

The bigger operational blocker is identity resolution. MarketScale reported that 44% of organizations say their AI data isn't ready, largely because fragmented buyer identity prevents them from connecting web behavior, product usage, and CRM records to the right account. First-party data only beats third-party signals if you can actually resolve who did what. Without identity resolution, your first-party data sits in three systems and none of them agree on who the account is.

What to Actually Run This Week

Setup: Pick one inbound flow (your highest-volume form or chat). Define the denominator you'll use: visitor-to-meeting, conversation-to-meeting, or qualified-lead-to-meeting. Lock that definition in writing before you touch anything else.

Hypothesis: If we add first-party context (past event attendance, content downloads, open opportunities) to the agent's pre-meeting research, then qualified-meeting rate will increase by 10-20% because reps enter calls with account history the prospect didn't volunteer.

Success metric: Qualified-meeting-to-opportunity conversion rate. Guardrail: Total meeting volume doesn't drop more than 15%. Stop-loss: If qualified-meeting rate drops below baseline after two weeks, revert and diagnose the qualification logic.

On AEO: Run a 50-prompt test across at least two AI providers. Record mentions and citations separately. Repeat in 30 days. The first run is your baseline, not your score.

SaaStr reports 10-minute bookings, but that's their stack with 30+ connected tools and 16 months of iteration. Your first version will be slower. The question is whether the quality of the first conversation improves enough to justify it.

Agent-led growth is real. The measurement infrastructure most teams have for it is not. Fix the denominator, segment the AEO score, and resolve identity before you trust the headline numbers.