Sierra's $15B Valuation Is a Bet on a New Kind of SaaS Moat

B2B SaaS activation rates sit at a 37.5% industry median in 2026 — a number that has barely moved in three years of dedicated product-led growth investment, onboarding optimization, and time-to-value engineering. For product teams who have spent those three years running activation experiments, the stagnation is demoralizing. For product teams who look more carefully at what is driving the number, the situation is more interesting and more complicated: a growing share of those activations are not being performed by humans at all.

AI agents are completing onboarding steps on behalf of users. Integration scripts are triggering first-workflow events automatically. Automation rules are checking the boxes that define activation in the analytics dashboard before the actual human decision-maker has engaged with the product in any meaningful way. The benchmark is holding, but it is measuring something different than it was measuring three years ago.

The Number Everyone Cites and Nobody Trusts

The 37.5% figure comes from aggregated data across mid-market and enterprise SaaS products, weighted toward products with trial-to-paid conversion funnels. It has become a gravitational reference point — teams beat it or fall short of it and design roadmaps around closing the gap.

The problem is that the metric was designed for a world where activation meant a human user performing a defined action that indicated they understood the product's value. In that world, activation correlated strongly with retention because the causal chain was clear: user understands value → user integrates it into their workflow → user renews.

The tools measuring activation are still counting completions. They are not measuring who or what completed them. And in an environment where a substantial fraction of the B2B users reaching SaaS products in 2026 are using AI agents, workflow automation, or integration scripts to interact with those products on their behalf, the completion count is increasingly detached from the human comprehension that made the metric predictive in the first place.

How AI Agents Are Breaking the Activation Metric

The mechanism is not subtle. Consider a company evaluating a new analytics tool as part of a software consolidation exercise. A procurement AI agent — or a member of the operations team using an AI-assisted workflow — completes the product's defined onboarding sequence: creates a workspace, connects the primary data source, runs the initial report template, and triggers the first automated insight delivery. The product's activation funnel records a completed activation. The dashboard goes green.

Three weeks later, nobody in the company has logged into the product since the procurement agent completed the setup. The actual decision-makers who would need to become regular users never engaged with it. The product gets cancelled at the 60-day review.

This is what product teams are calling ghost activation: the account reaches the activation milestone through non-human action and the activation-retention correlation breaks. The product team sees a 37.5% activation rate and interprets it through the lens of the pre-AI model — these accounts understood our value — when the correct interpretation requires distinguishing which of those activations involved genuine human comprehension.

The Bimodal Split in the Data

The most revealing signal in 2026 SaaS activation data is not the median — it is the distribution. Activation rates are increasingly bimodal: a cluster of products in the 20–28% range and another cluster in the 55–70% range, with relatively few products in the middle. The industry median obscures this structure.

Products in the high-activation cluster share a specific characteristic: their defined activation milestone requires a behavior that cannot be easily completed by an automation or AI agent. They require a human-authored configuration decision, a peer collaboration action, a subjective judgment call, or an action that requires context that only the actual user possesses. These products are measuring human activation because their activation milestone filters out non-human completions by design.

Products in the low-activation cluster are concentrated in the category that PLG growth research from ProductGrowth.in identifies as "functional automation" — tools where the first-value event is operational in nature (running a report, sending a message, completing a workflow) rather than reflective or collaborative. These are the tools most susceptible to ghost activation because their defined aha moment looks identical whether performed by a human or an agent.

The practical implication for product teams is that the activation benchmark comparison is only meaningful if you are comparing products in the same behavioral category. A tool with a collaboration-based activation milestone competing against a tool with an API-triggered activation milestone should not be benchmarked against the same median.

Why Your Activation Tools Were Built for Humans

The incumbent activation and onboarding tool landscape — Appcues, Chameleon, Userpilot, Pendo, and their competitors — was built to solve a specific problem: human users landing in a product and not knowing what to do next. The entire category evolved from the premise that activation failure is a comprehension problem, and that in-product guidance, tooltips, checklists, and contextual prompts could reduce the comprehension gap enough to move more users to the aha moment.

That model worked when the activating actor was always a human who needed comprehension assistance. It starts breaking when a meaningful fraction of the activating actors are non-human agents that do not need comprehension assistance — they just need the API to be available.

Chameleon has moved furthest in 2026 toward agent-aware analytics, adding source attribution tagging that allows teams to flag automations as non-human. The other major incumbents are largely in a 12–18 month product cycle lag behind the problem.

The emerging category of AI-native activation platforms — several of which are in public beta as of mid-2026 — is built from the ground up to handle non-human actors. These platforms instrument the event stream with actor identity at the API layer, track human engagement signals separately from automation completions, and generate cohort analyses that distinguish human-activated and agent-activated accounts before they reach the retention stage.

The New Activation Metrics Stack

Redefining activation measurement for 2026 requires three instrumentation changes that most product analytics setups do not currently support.

Signal-source tagging. Every activation event needs to carry metadata identifying the initiating actor type: direct browser session (human), API call without session context (automation or agent), authenticated user via API (power user or developer), or known integration connector. This is a backend instrumentation problem, not a frontend problem. It requires the event tracking infrastructure to capture actor context at the moment of event emission, not just the event itself.

Comprehension-based event design. The activation milestone definition needs to be stress-tested against the question: could an AI agent complete this without a human understanding anything? If yes, the milestone is measuring the wrong thing. Comprehension-based activation milestones require behaviors that reflect human judgment — a user-authored description of a use case, a configuration decision requiring contextual knowledge, an annotation explaining why a particular option was chosen. These behaviors cannot be spoofed by an automation without the contextual knowledge they are designed to reveal.

The 72-hour engagement signal. The most practical interim metric for teams that cannot rebuild their activation instrumentation immediately is to track whether a human user actively engages with the product within 72 hours of an activation event. A human login, a human-initiated action, a human-generated event within that window indicates that the activation involved or followed by human comprehension. No human engagement within 72 hours of an activation event is the strongest readily measurable proxy for ghost activation.

The Agent-Aware Activation Playbook

The following five-step playbook converts a standard PLG activation motion into one that handles AI-agent traffic without losing the signal quality that makes activation a useful retention predictor.

1. Audit the activation baseline. Run a 90-day lookback on your activation events. For each activation, determine whether it was initiated by a direct browser session with human behavior signals or by an API call, integration trigger, or known automation. If your current analytics stack does not support this distinction, use proxy signals: session presence, click events within the same session, time-on-page during the activation workflow. The goal is an estimate of your current ghost activation rate — the percentage of activations with no concurrent human engagement signal.

2. Instrument source-actor tagging. Add actor-type metadata to every activation event in your product analytics stack. This is the foundational technical change. Products using Segment, Mixpanel, or Amplitude can add actor_type as a custom property on activation events with modest engineering investment. Products without a centralized analytics layer will need to instrument directly in their event capture layer before this becomes feasible.

3. Re-baseline the human activation rate. Once source-actor tagging is live, calculate your human activation rate separately from your total activation rate. This number will be lower than your current reported rate. That is correct. It is the number your product roadmap should optimize against, because it is the number that correlates with retention in the way the original activation-retention model predicted.

4. Redesign the aha moment for the AI-agent era. If AI agents can easily complete your current activation milestone, redesign it to require a behavior that reflects human comprehension. The PLG activation ceiling research suggests that products with subjective, judgment-based activation milestones outperform products with purely functional milestones by 15–20 percentage points in 6-month retention, independently of activation rate. The activation milestone is a product design decision with retention consequences.

5. Build the 72-hour engagement check. Create a cohort analysis — automated if possible — that identifies activations where no human engagement was detected within 72 hours. Route those accounts to a human-touch intervention sequence: a direct outreach email, a calendar link for a guided session, or a high-touch CSM engagement if the account size justifies it. The saas retention cliff research shows that re-engagement probability drops significantly after the 14-day mark; the 72-hour window is the point at which intervention is still cost-effective.

The 90-Day Implementation Checklist

For product and growth teams starting this work, a realistic 90-day implementation timeline looks like this:

Days 1–30: Audit current activation data. Identify the proxy signals available to estimate ghost activation rate without new instrumentation. Brief the engineering team on the source-actor tagging requirement. Define the target architecture for actor-type metadata in the event stream.

Days 31–60: Ship source-actor tagging to the event capture layer. Begin collecting clean separation between human-initiated and automation-initiated events. Run parallel cohort analyses using the old and new activation definitions to quantify the ghost activation gap.

Days 61–90: Re-baseline the human activation rate. Evaluate the current activation milestone against the AI-agent stress test. Design and spec the revised activation milestone if the current one fails the stress test. Build the 72-hour engagement check in the analytics pipeline and connect it to the intervention sequence.

What AI-Era Activation Actually Looks Like

Deloitte's 2026 enterprise software adoption research found that 34% of enterprise SaaS products now see AI agent activity within the first 30 days of account creation — up from less than 5% in 2024. For developer tools, that number is above 60%. For API-first products, it approaches 80%.

The activation function has not disappeared. Human activation still correlates strongly with retention, and improving genuine human activation rates still drives the same compounding revenue outcomes it always has. What has changed is that the metric has become noisy in ways that obscure the signal, and the tools designed to improve activation are measuring noise as signal.

Activation optimization programs that do not account for AI agent traffic will optimize for ghost activation — they will run experiments that improve automation completion rates without improving human comprehension rates, and they will attribute retention improvements that do not materialize because the retention model breaks when activation is not human-led.

The products that will lead the next wave of PLG growth are the ones whose product teams understood in 2026 that the benchmark broke and built activation architectures that measure human value discovery rather than event completion. They will look at the same 37.5% median figure as everyone else and see a different problem: not a comprehension gap, but a measurement gap that is hiding a comprehension gap beneath a layer of automation noise.

Takeaway: The 37.5% B2B SaaS activation median is increasingly unreliable as a comparative benchmark because a growing fraction of activations are being completed by AI agents and automations rather than by human users. Product teams that continue measuring activation the traditional way are likely overstating their retention outlook in accounts with high AI agent prevalence and optimizing onboarding experiments against a metric that no longer predicts what it was designed to predict. The practical path forward is a three-layer instrumentation upgrade — source-actor tagging, comprehension-based event design, and 72-hour human engagement tracking — deployed over a 90-day implementation cycle. Products that make this transition will see their reported activation rate decline (because ghost activations will be correctly excluded) and their activation-to-retention correlation strengthen (because the metric will again be measuring what it was always supposed to measure).