Back in 2024, Gartner was already anticipating a rapid shift in traffic entry points: an increasing share of traffic from traditional search engines was expected to be absorbed by conversational agents and generative search. Six months later, Similarweb was already measuring a significant increase in referral traffic from ChatGPT to e-commerce websites. Yet while acquisition surfaces are being redefined at speed, Statsig’s State of Experimentation report reminds us that the median mature organization still caps out at only a few dozen A/B tests per year, and that only a minority of companies sustain a continuous weekly experimentation cadence. The gap between market velocity and optimization velocity has therefore never been wider.
The challenge is a straightforward one: how do you manage user experience in a granular, data-driven way while keeping pace with the speed imposed by the market?
Historically, conversion rate optimization (CRO) has been a highly labor-intensive process. A consultant’s production capacity generally tops out at around thirty tests per year, depending on complexity. This ceiling is not a talent problem; it is primarily a structural one.
When you test thirty hypotheses per year and a portion inevitably fails, your annual learning output shrinks to only a handful of exploitable signals. Over the course of an e-commerce journey containing, according to the Baymard Institute, dozens of documented friction points throughout the standard purchase funnel, that ratio becomes mechanically insufficient.
For CRO consultants, workloads tend to crystallize around four strategic pillars which, without intelligent automation, become bottlenecks:
Beyond testing itself, the challenge becomes building an experimentation architecture capable of leveraging both historical and current data to anticipate future growth levers. This means moving away from fragmented, artisanal CRO toward industrialized CRO augmented by data and AI.
The market quickly applied AI to both ends of the experimentation chain.
Contentsquare, through its Sense layer, targeted the diagnostic bottleneck: transforming heatmaps, session replays, and behavioral signals into readable hypotheses. At fifty-five, we observed up to a +60% improvement in insight extraction time during are, an outcome consistent with McKinsey benchmarks, which identify analytical functions among the areas most impacted by generative AI.
Kameleoon, meanwhile, initially focused on execution through its Personalization Builder eXperience (PBX): generating variations and modifying the DOM without requiring already overloaded engineering teams. And here, some nuance is needed because the product has evolved significantly.
The PBX 2.0 Interlude At launch, PBX was built around a deliberate assumption: rather than layering AI on top of a visual editor, the interface itself would be designed around the agent. With PBX 2.0, Kameleoon moved from a single agent to five specialized agents, each dedicated to a stage of the experimentation lifecycle: Ideate (identifying high-potential experiments), Build (generating variants), Configure (conversational setup), Analyze (structuring insights), Ship (deploying winning variations).
Kameleoon now positions this suite as a team member for CRO teams, capable of reasoning, executing, and helping orchestrate experimentation programs. This evolution confirms that agentic systems applied to CRO are becoming the defining direction of the industry; Forrester had already identified this dual movement (AI for insights, AI for execution) as the major transformation shaping the category. When a vendor of Kameleoon’s scale moves toward a multi-agent architecture spanning the entire experimentation cycle, the market signal becomes difficult to ignore.
Still, a limitation remains — which is not a criticism of the quality of these tools, but a question of architecture. Sense, PBX 2.0, and other agentic components are powerful, but each remains anchored within its own perimeter: Sense reasons within Contentsquare’s behavioral data; PBX orchestrates experimentation inside Kameleoon’s ecosystem. Used independently, these high-performing suites recreate precisely what CRO was trying to escape, i.e. intelligent islands that share neither memory nor common reference frameworks. The consultant therefore becomes the manual connective tissue between ecosystems once again, and the ceiling merely shifts rather than disappears.
This is precisely the point that deserves attention from CMOs and CTOs: while the productivity of individual links has improved, the end-to-end chain itself has not yet been fundamentally redesigned.
Agentic vendor suites are powerful, but sustainable value lies in eliminating data silos. This is the rationale behind the fifty-five approach: not replacing these tools, but orchestrating them within an AI-augmented CRO operating model.
In practice, our approach involves twelve specialized agents (the GrowthSense suite) capable of orchestrating the full cycle, from scoping to scaling, and automating up to three quarters of repetitive tasks within the experimentation process. Where a vendor suite accelerates execution within its own platform boundaries, these agents operate above the existing stack.
The strength of this approach relies on three principles:
The ultimate objective is simple: to cut production time per test in half. By removing operational execution from teams’ workloads, brands can refocus on strategic arbitration. The distinction is clear: PBX 2.0 makes Kameleoon faster; an agnostic orchestration layer makes the entire CRO program faster, more rigorous, and capable of compounding learning regardless of the underlying toolset.
But in this new paradigm, what becomes of teams? McKinsey’s analysis of generative AI’s economic potential is more nuanced than it is often remembered: while a majority of current marketing tasks could be automated, value creation concentrates in the remaining fraction — specifically judgment, prioritization, and strategic framing.
Applied to CRO, this means that the consultants who will succeed are those capable of reading annual experimentation portfolios the way investors read asset portfolios, through theses, trade-offs, and disciplined allocation. The role moves upward along the value chain, becoming more strategic than operational.
This is also the honest promise worth making, contrary to much of the surrounding narrative: AI does not replace teams and does not automatically discover the “best” optimizations. It reduces uncertainty, frees time, and compounds learning. Real experimentation remains the ultimate arbiter.
Brands will not lose the optimization race because they lack hypotheses, but because they cannot test them fast enough for the resulting learning to remain relevant. And in a market where an increasing share of user searches will flow through generative interfaces, speed becomes the top priority.
In this context, far from a premium option, Industrialized CRO is becoming the new baseline. Competitive advantage will therefore belong to organizations capable of orchestrating agents on top of unified data foundations, preserving historical learning, and maintaining disciplined decision-making downstream.
Discover all the latest news, articles, webinar replays and fifty-five events in our monthly newsletter, Tea O'Clock.
.svg)
