Dalton
Dalton
Dalton
Discover
Discover
Discover
Key Deliverables:
Key Deliverables:
→
→
→
Web application;
Web application;
→
→
→
Onboarding & website scan flow;
Onboarding & website scan flow;
→
→
→
Marketing website;
Marketing website;
→
→
→
Design system.
Design system.
Timeline:
2025 → Ongoing
Timeline:
2025 → Ongoing
Role:
Lead Product Designer
Role:
Lead Product Designer
Role:
Founding designer → Head of Design
(team of 3 designers)
Engagement:
End-to-end product ownership
Engagement:
End-to-end product ownership
Dalton is an AI CRO co-pilot that turns website optimization from a slow, developer-dependent process into a continuous, autonomous loop — scanning a site and its analytics, generating test-worthy ideas, deploying variants without code, and shifting traffic to winners on its own.
Dalton is an AI CRO co-pilot that turns website optimization from a slow, developer-dependent process into a continuous, autonomous loop — scanning a site and its analytics, generating test-worthy ideas, deploying variants without code, and shifting traffic to winners on its own.
Growth teams wanted the speed of AI experimentation without losing control of their brand or their site. Every design decision had to answer the same question a marketer asks before clicking "launch": can I trust what this is about to change on my live website? Dalton needed an interface confident enough to act autonomously, and transparent enough that a non-technical team would still hit approve.
Growth teams wanted the speed of AI experimentation without losing control of their brand or their site. Every design decision had to answer the same question a marketer asks before clicking "launch": can I trust what this is about to change on my live website? Dalton needed an interface confident enough to act autonomously, and transparent enough that a non-technical team would still hit approve.
<5min
<5min
First live experiment
First live experiment
100%
100%
Brands running on Dalton
20+
20+
Concurrent experiments
Define
Define
Define
Dalton enters CRO tooling built for a different kind of company. Optimizely and VWO were designed for organisations that already have a dedicated growth team, a design system, an engineering queue, and enough traffic to reach statistical significance in weeks. Everyone below that bar — most e-commerce brands, most lean SaaS teams — was priced and staffed out of testing entirely, running two to five experiments a year and defaulting to gut feeling everywhere else on the site.
Dalton enters CRO tooling built for a different kind of company. Optimizely and VWO were designed for organisations that already have a dedicated growth team, a design system, an engineering queue, and enough traffic to reach statistical significance in weeks. Everyone below that bar — most e-commerce brands, most lean SaaS teams — was priced and staffed out of testing entirely, running two to five experiments a year and defaulting to gut feeling everywhere else on the site.
AI page builders and generic "auto-optimise" tools promised to close that gap, but most treat a variant as content — swap a headline, hope for the best — with no measurement discipline behind it and no way to prove a change actually worked before it ships to every visitor.
AI page builders and generic "auto-optimise" tools promised to close that gap, but most treat a variant as content — swap a headline, hope for the best — with no measurement discipline behind it and no way to prove a change actually worked before it ships to every visitor.
24 growth marketer interviews
24 growth marketer interviews
We spoke with heads of growth, founder-led marketers, and CRO leads across DTC, e-commerce, and product-led SaaS — teams already running fast on paid acquisition but stuck moving slowly on-site. Walking in, we expected the core barrier to be technical: no developer, no time, no traffic.
We spoke with heads of growth, founder-led marketers, and CRO leads across DTC, e-commerce, and product-led SaaS — teams already running fast on paid acquisition but stuck moving slowly on-site. Walking in, we expected the core barrier to be technical: no developer, no time, no traffic.
The pattern that emerged was different. Marketers didn't lack ideas — most had a backlog of untested hypotheses already. What they lacked was the confidence to press launch. Every AI tool that promised autonomous optimisation triggered the same hesitation: what if it ships something off-brand while I'm not looking? Speed only mattered if it came with a moment where a human could see exactly what was about to change and say yes.
The pattern that emerged was different. Marketers didn't lack ideas — most had a backlog of untested hypotheses already. What they lacked was the confidence to press launch. Every AI tool that promised autonomous optimisation triggered the same hesitation: what if it ships something off-brand while I'm not looking? Speed only mattered if it came with a moment where a human could see exactly what was about to change and say yes.
This led us to ask: "How might we let a non-technical growth team ship dozens of experiments a month without ever losing control of what's actually live on their site?"
This led us to ask: "How might we let a non-technical growth team ship dozens of experiments a month without ever losing control of what's actually live on their site?"
The reframing shaped the product from the first screen. Of the three core modules — CRO Recommendations, No-Code Deployment, and Continuous Optimisation — CRO Recommendations was prioritised first in the experience, even though it's the least differentiated technically. It's the only interaction where Dalton asks for nothing in return: no code touched, no approval needed, no risk taken. Earning a credible list of test-worthy ideas before ever asking to touch the live site was the fastest way to earn enough trust to reach the module that actually mattered.
The reframing shaped the product from the first screen. Of the three core modules — CRO Recommendations, No-Code Deployment, and Continuous Optimisation — CRO Recommendations was prioritised first in the experience, even though it's the least differentiated technically. It's the only interaction where Dalton asks for nothing in return: no code touched, no approval needed, no risk taken. Earning a credible list of test-worthy ideas before ever asking to touch the live site was the fastest way to earn enough trust to reach the module that actually mattered.
Personas
Personas
Elena, the Head of Growth (33, Series B DTC brand):
Elena, the Head of Growth (33, Series B DTC brand):
→
→
Owns performance marketing and needs the site to iterate as fast as ad creative changes;
Owns performance marketing and needs the site to iterate as fast as ad creative changes;
→
→
Backlog of testing ideas has outpaced what a two-person team can design, build, and QA;
Backlog of testing ideas has outpaced what a two-person team can design, build, and QA;
→
→
Needs experiments live within hours, not sprint cycles, without breaking brand guidelines he doesn't have time to police manually.
Needs experiments live within hours, not sprint cycles, without breaking brand guidelines he doesn't have time to police manually.
Ray, the Founder & CMO (38, bootstrapped e-commerce brand):
Ray, the Founder & CMO (38, bootstrapped e-commerce brand):
→
→
→
Runs marketing solo, without a developer or designer anywhere on the team;
Runs marketing solo, without a developer or designer anywhere on the team;
→
→
→
Traffic is too low for traditional A/B tools to reach significance before the business needs an answer;
Traffic is too low for traditional A/B tools to reach significance before the business needs an answer;
→
→
→
Needs to trust that an AI-generated variant won't misrepresent the brand to a customer she's spent years building loyalty with.
Needs to trust that an AI-generated variant won't misrepresent the brand to a customer she's spent years building loyalty with.
Sofia, the Senior Digital & CRO Manager (41, mid-market retailer):
Sofia, the Senior Digital & CRO Manager (41, mid-market retailer):
→
→
→
Manages digital experience across multiple regional markets with lean local teams;
Manages digital experience across multiple regional markets with lean local teams;
→
→
→
Currently limited to two to five experiments a year by developer and design dependency;
Currently limited to two to five experiments a year by developer and design dependency;
→
→
→
Needs statistical rigor and automated traffic-shifting to defend testing ROI to leadership, without micromanaging every experiment personally.
Needs statistical rigor and automated traffic-shifting to defend testing ROI to leadership, without micromanaging every experiment personally.
Problem Statement
Problem Statement
Testing Requires a Team You Don't Have
Every test meant briefing a designer, queuing a developer, and waiting for QA — so most teams below enterprise scale ran it only a handful of times a year while the rest of marketing moved weekly.
Every test meant briefing a designer, queuing a developer, and waiting for QA — so most teams below enterprise scale ran it only a handful of times a year while the rest of marketing moved weekly.
Traffic Too Low for Statistical Confidence
Traffic Too Low for Statistical Confidence
Traffic Too Low for Statistical Confidence
Classic tools like VWO and Optimizely assume tens of thousands of monthly visitors per page. Most sites never clear that bar, so tests ran for months without reaching significance.
Classic tools like VWO and Optimizely assume tens of thousands of monthly visitors per page. Most sites never clear that bar, so tests ran for months without reaching significance.
AI Autonomy Without Accountability
AI Autonomy Without Accountability
AI Autonomy Without Accountability
Early AI optimisation tools promised to "just handle it" — with no visibility into what was about to change and no clear way to stop something quietly hurting conversion.
Early AI optimisation tools promised to "just handle it" — with no visibility into what was about to change and no clear way to stop something quietly hurting conversion.
Ideas Outpacing Execution Capacity
Ideas Outpacing Execution Capacity
Ideas Outpacing Execution Capacity
Growth teams weren't short on hypotheses. The bottleneck was the distance between having an idea on Tuesday and having it live, measured, and resolved.
Growth teams weren't short on hypotheses. The bottleneck was the distance between having an idea on Tuesday and having it live, measured, and resolved.
How might we?
How might we?
→
→
→
HMW let a non-technical marketer launch a test in minutes without touching a line of code?
HMW let a non-technical marketer launch a test in minutes without touching a line of code?
→
→
→
HMW generate testing ideas that are actually worth running, not generic headline swaps?
HMW generate testing ideas that are actually worth running, not generic headline swaps?
→
→
→
HMW reach statistical confidence on the traffic a real mid-market site actually gets?
HMW reach statistical confidence on the traffic a real mid-market site actually gets?
→
→
→
HMW give a human a clear moment to say yes before anything changes on a live, production site?
HMW give a human a clear moment to say yes before anything changes on a live, production site?
→
→
→
HMW keep improving a page automatically, without turning into something the team stops checking on?
HMW keep improving a page automatically, without turning into something the team stops checking on?
Key metrics
Key metrics
To track success, we monitored:
To track success, we monitored:
→
→
→
Time to first live experiment — How fast a new team goes from site scan to a variant running in front of real traffic;
Time to first live experiment — How fast a new team goes from site scan to a variant running in front of real traffic;
→
→
→
Approval-to-launch rate — Share of AI-generated variants a marketer approves without requesting edits;
Approval-to-launch rate — Share of AI-generated variants a marketer approves without requesting edits;
→
→
→
Time to statistical significance — Median duration before an experiment produces a confident, actionable result;
Time to statistical significance — Median duration before an experiment produces a confident, actionable result;
→
→
→
Experiments run per team per month — The clearest signal of whether testing became routine or stayed occasional;
Experiments run per team per month — The clearest signal of whether testing became routine or stayed occasional;
→
→
→
Traffic auto-allocation accuracy — How reliably the system shifted visitors toward winning variants without human intervention.
Traffic auto-allocation accuracy — How reliably the system shifted visitors toward winning variants without human intervention.
Iterate
Iterate
Iterate
Our design process for Dalton was driven by the need to balance full AI autonomy — continuous experimentation, automatic traffic-shifting, self-generated hypotheses — with the confidence of a marketer who has never run a test in her life and is about to let an algorithm touch the site that pays her salary. To achieve this, we worked closely with the ML and growth engineering teams, testing every interaction against a single constraint: automation could never outrun the user's understanding of what it was about to do.
Our design process for Dalton was driven by the need to balance full AI autonomy — continuous experimentation, automatic traffic-shifting, self-generated hypotheses — with the confidence of a marketer who has never run a test in her life and is about to let an algorithm touch the site that pays her salary. To achieve this, we worked closely with the ML and growth engineering teams, testing every interaction against a single constraint: automation could never outrun the user's understanding of what it was about to do.
The challenge wasn't making AI powerful. It was making AI legible enough that someone without a CRO background would trust it with production.
The challenge wasn't making AI powerful. It was making AI legible enough that someone without a CRO background would trust it with production.
First Iteration:
First Iteration:
→
→
→
Started with a raw idea feed — Dalton scanned the site and analytics, then surfaced a scrolling list of AI-generated test hypotheses as text, ranked by predicted impact score;
Started with a raw idea feed — Dalton scanned the site and analytics, then surfaced a scrolling list of AI-generated test hypotheses as text, ranked by predicted impact score;
→
→
→
Approval was a single global action — marketers reviewed a written description of each idea and clicked approve or reject, without seeing what the actual page would look like after the change;
Approval was a single global action — marketers reviewed a written description of each idea and clicked approve or reject, without seeing what the actual page would look like after the change;
→
→
→
Launch and monitoring lived in separate views — once approved, a variant went live with no persistent surface showing how it was performing relative to control;
Launch and monitoring lived in separate views — once approved, a variant went live with no persistent surface showing how it was performing relative to control;
→
→
→
Validated that AI could generate a high volume of plausible-sounding hypotheses — but exposed a harder problem the moment marketers were asked to act on them.
Validated that AI could generate a high volume of plausible-sounding hypotheses — but exposed a harder problem the moment marketers were asked to act on them.
Second Iteration:
Second Iteration:
Testing with growth marketers revealed a consistent pattern: nobody trusted a hypothesis they couldn't see. A written description like "test a more urgent CTA copy" produced hesitation, no matter how confident the predicted impact score looked — because the marketer had no way to picture the actual change against her actual brand, on her actual page, before it went in front of real customers.
Testing with growth marketers revealed a consistent pattern: nobody trusted a hypothesis they couldn't see. A written description like "test a more urgent CTA copy" produced hesitation, no matter how confident the predicted impact score looked — because the marketer had no way to picture the actual change against her actual brand, on her actual page, before it went in front of real customers.
→
→
→
Replaced text-only hypotheses with a live before/after preview — every AI-generated variant renders directly on a snapshot of the real page, side by side with the current control;
Replaced text-only hypotheses with a live before/after preview — every AI-generated variant renders directly on a snapshot of the real page, side by side with the current control;
→
→
→
Moved approval from a global action to a per-variant decision — marketers review, edit, or reject each experiment individually, with inline editing for copy and layout before anything ships;
Moved approval from a global action to a per-variant decision — marketers review, edit, or reject each experiment individually, with inline editing for copy and layout before anything ships;
→
→
→
Built a persistent control room view — live sessions, conversion rate by variant, and traffic allocation update in real time, so a launched experiment is never a black box once it's running;
Built a persistent control room view — live sessions, conversion rate by variant, and traffic allocation update in real time, so a launched experiment is never a black box once it's running;
→
→
→
Surfaced a confidence threshold visibly on every active test — marketers see exactly how far an experiment is from statistical significance, instead of wondering whether a result can be trusted yet;
Surfaced a confidence threshold visibly on every active test — marketers see exactly how far an experiment is from statistical significance, instead of wondering whether a result can be trusted yet;
→
→
→
Aligned the visual language of the review screen with the brand's own site — variants preview inside a frame that matches the marketer's actual fonts, colors, and layout conventions, closing the gap between "AI suggestion" and "my website."
Aligned the visual language of the review screen with the brand's own site — variants preview inside a frame that matches the marketer's actual fonts, colors, and layout conventions, closing the gap between "AI suggestion" and "my website."
These iterations reframed the product from AI that generates testing ideas to a system a marketer can watch make decisions in real time and stop at any moment. The shift was what turned Dalton from a novelty growth teams tried once into a tool they checked every morning — moving the product from idea generator into the daily operating rhythm of the team.
These iterations reframed the product from AI that generates testing ideas to a system a marketer can watch make decisions in real time and stop at any moment. The shift was what turned Dalton from a novelty growth teams tried once into a tool they checked every morning — moving the product from idea generator into the daily operating rhythm of the team.
Design
Design
Design
The Design phase translated an autonomous AI system into an interface a marketer with no design or CRO background could operate with confidence — someone approving her first experiment at 9am and checking uplift numbers before lunch. Every visual and interaction decision was tested against one question: would you actually trust what this is about to change on your live site?
The Design phase translated an autonomous AI system into an interface a marketer with no design or CRO background could operate with confidence — someone approving her first experiment at 9am and checking uplift numbers before lunch. Every visual and interaction decision was tested against one question: would you actually trust what this is about to change on your live site?
The scope was unusual for a different reason than most greenfield builds: Dalton already had a brand, but no design system extending it across a dense, data-heavy product surface. The visual identity had been built for a marketing site — confident, punchy, conversion-focused. Translating that same voice into a control room full of live sessions, statistical confidence bars, and side-by-side variant previews meant deciding what carries over from brand to product, and what a data-dense interface needs that a landing page never did.
The scope was unusual for a different reason than most greenfield builds: Dalton already had a brand, but no design system extending it across a dense, data-heavy product surface. The visual identity had been built for a marketing site — confident, punchy, conversion-focused. Translating that same voice into a control room full of live sessions, statistical confidence bars, and side-by-side variant previews meant deciding what carries over from brand to product, and what a data-dense interface needs that a landing page never did.
Color Scheme & Visual Identity:
Color Scheme & Visual Identity:
→
→
→
The product's primary surface splits deliberately — a dark, focused navigation rail for control and configuration, paired with a light, card-based content area for scanning experiment status at a glance. The split mirrors how marketers actually use the tool: quick orientation in the sidebar, sustained attention in the workspace;
The product's primary surface splits deliberately — a dark, focused navigation rail for control and configuration, paired with a light, card-based content area for scanning experiment status at a glance. The split mirrors how marketers actually use the tool: quick orientation in the sidebar, sustained attention in the workspace;
→
→
→
Dalton's violet was carried over from the marketing brand and narrowed to a single job in the product: it marks anything AI-touched — a suggested experiment, a generated variant, an insight the system surfaced on its own. Everything human-authored stays neutral, so the eye always knows who made a given decision;
→
→
→
Uplift and status indicators use a restrained green/red system reserved exclusively for performance data — never repurposed for UI state or brand emphasis, so a marketer scanning a page of experiment cards can trust the color coding without reading every number;
Uplift and status indicators use a restrained green/red system reserved exclusively for performance data — never repurposed for UI state or brand emphasis, so a marketer scanning a page of experiment cards can trust the color coding without reading every number;
→
→
→
Typography and spacing were tuned for a data-dense audience — tighter default line-heights and a compact card grid than the marketing site uses, prioritising the ability to scan a dozen experiments in one view over the more spacious, editorial rhythm of Dalton's public-facing pages.
Typography and spacing were tuned for a data-dense audience — tighter default line-heights and a compact card grid than the marketing site uses, prioritising the ability to scan a dozen experiments in one view over the more spacious, editorial rhythm of Dalton's public-facing pages.
User Experience & Engagement:
User Experience & Engagement:
→
→
→
Every AI-generated hypothesis renders as a live, on-brand preview before a marketer is ever asked to approve it — no written descriptions to interpret, no imagining what a "more urgent CTA" actually looks like against the real page;
Every AI-generated hypothesis renders as a live, on-brand preview before a marketer is ever asked to approve it — no written descriptions to interpret, no imagining what a "more urgent CTA" actually looks like against the real page;
→
→
→
Approval happens per variant, not in bulk — a marketer reviews, edits, or rejects each experiment individually, with inline editing available before anything reaches a single visitor;
→
→
→
Statistical confidence is surfaced visibly on every running test, not buried in a report — marketers see exactly how close an experiment is to significance, replacing the anxious question "can I trust this yet?" with a number they can read at a glance;
Statistical confidence is surfaced visibly on every running test, not buried in a report — marketers see exactly how close an experiment is to significance, replacing the anxious question "can I trust this yet?" with a number they can read at a glance;
→
→
→
A live control room view keeps every launched experiment observable — real-time sessions, conversion rate by variant, and automatic traffic allocation update continuously, so nothing running goes unmonitored between the moment of approval and the moment of resolution.
A live control room view keeps every launched experiment observable — real-time sessions, conversion rate by variant, and automatic traffic allocation update continuously, so nothing running goes unmonitored between the moment of approval and the moment of resolution.
Adoption & Distribution Strategy:
Adoption & Distribution Strategy:
→
→
→
Onboarding was designed around a single first-value milestone — install one script, and the first AI-generated test-worthy insight about your own site appears in under a minute, giving every new user proof of value before they've configured anything;
Onboarding was designed around a single first-value milestone — install one script, and the first AI-generated test-worthy insight about your own site appears in under a minute, giving every new user proof of value before they've configured anything;
→
→
→
The marketing site itself was treated as a preview of the product's confidence — the same restraint, the same clarity, the same "here's exactly what changes and why" transparency that defines the in-product experience;
→
→
→
Experiment cards were designed as a reusable pattern from day one, appearing identically across the dashboard, the onboarding scan results, and the marketing site's product screenshots — so a prospect evaluating Dalton from an ad sees the same interface they'll actually use;
Experiment cards were designed as a reusable pattern from day one, appearing identically across the dashboard, the onboarding scan results, and the marketing site's product screenshots — so a prospect evaluating Dalton from an ad sees the same interface they'll actually use;
→
→
→
Platform-agnostic integration was made visible early in the funnel — logos for Shopify, Webflow, WordPress, and a dozen other stacks appear before signup, removing the single biggest hesitation for a non-technical marketer: will this even work with what I already have?
Platform-agnostic integration was made visible early in the funnel — logos for Shopify, Webflow, WordPress, and a dozen other stacks appear before signup, removing the single biggest hesitation for a non-technical marketer: will this even work with what I already have?
The design evolution for Dalton reflects a deliberate stance on AI autonomy: the product doesn't ask marketers to trust a black box — it shows its work, at every step, in a visual language dense enough for a growth team living in analytics dashboards and calm enough for someone who has never run an experiment before. Visual system, product surface, and adoption strategy were built around the same idea — automation earns trust by staying visible, never by disappearing.
The design evolution for Dalton reflects a deliberate stance on AI autonomy: the product doesn't ask marketers to trust a black box — it shows its work, at every step, in a visual language dense enough for a growth team living in analytics dashboards and calm enough for someone who has never run an experiment before. Visual system, product surface, and adoption strategy were built around the same idea — automation earns trust by staying visible, never by disappearing.
Measure & Test
Measure & Test
Measure & Test
Data Analysis & Testing Approach:
Data Analysis & Testing Approach:
→
→
→
Instrumented the product with PostHog and structured event logging from day one — every scan, generated insight, variant edit, approval, and traffic reallocation was tracked end-to-end, giving the team a continuous signal on where marketers hesitated and where they moved fast;
Instrumented the product with PostHog and structured event logging from day one — every scan, generated insight, variant edit, approval, and traffic reallocation was tracked end-to-end, giving the team a continuous signal on where marketers hesitated and where they moved fast;
→
→
→
Found that variants shown as a live, on-brand preview were approved without edits 2.4x more often than the earlier text-only hypothesis format — confirming the central thesis that marketers trust what they can see, not what they're told;
→
→
→
Session replays revealed the moment of highest hesitation wasn't idea generation — it was the approval screen. Marketers spent more time studying a single variant before launch than reviewing an entire week of insights, which shifted the team's design priority away from the dashboard and onto that one screen;
Session replays revealed the moment of highest hesitation wasn't idea generation — it was the approval screen. Marketers spent more time studying a single variant before launch than reviewing an entire week of insights, which shifted the team's design priority away from the dashboard and onto that one screen;
→
→
→
Implemented a measurement loop where every product release was paired with a corresponding approval-rate check — UI changes that made the review screen harder to parse were caught before they shipped, even when the underlying AI logic hadn't changed at all.
Implemented a measurement loop where every product release was paired with a corresponding approval-rate check — UI changes that made the review screen harder to parse were caught before they shipped, even when the underlying AI logic hadn't changed at all.
Enhancements Following GDPR & Compliance Readiness:
Enhancements Following GDPR & Compliance Readiness:
→
→
→
As Dalton's script began running on production sites across the EU and beyond, the product surface was upgraded to make privacy and data compliance a visible signal of trust rather than a page buried in the footer:
As Dalton's script began running on production sites across the EU and beyond, the product surface was upgraded to make privacy and data compliance a visible signal of trust rather than a page buried in the footer:
As Dalton's script began running on production sites across the EU and beyond, the product surface was upgraded to make privacy and data compliance a visible signal of trust rather than a page buried in the footer:
•
•
•
Consent-aware variant delivery, so experiments respect a visitor's existing cookie preferences automatically;
Consent-aware variant delivery, so experiments respect a visitor's existing cookie preferences automatically;
Consent-aware variant delivery, so experiments respect a visitor's existing cookie preferences automatically;
•
•
•
Data residency and hosting region exposed directly in workspace settings, not just in a signed agreement;
Data residency and hosting region exposed directly in workspace settings, not just in a signed agreement;
Data residency and hosting region exposed directly in workspace settings, not just in a signed agreement;
•
•
•
A full change log surfaced inside every workspace, giving a customer's own compliance team an audit trail without needing to ask Dalton's;
A full change log surfaced inside every workspace, giving a customer's own compliance team an audit trail without needing to ask Dalton's;
A full change log surfaced inside every workspace, giving a customer's own compliance team an audit trail without needing to ask Dalton's;
•
•
•
Instant pause and rollback controls made visible at the top level of every experiment, not buried two clicks deep.
Instant pause and rollback controls made visible at the top level of every experiment, not buried two clicks deep.
Instant pause and rollback controls made visible at the top level of every experiment, not buried two clicks deep.
→
→
→
These signals shortened enterprise sales cycles — compliance questions that previously meant a round-trip through email and a signed DPA were increasingly answered inside the product itself.
These signals shortened enterprise sales cycles — compliance questions that previously meant a round-trip through email and a signed DPA were increasingly answered inside the product itself.
These signals shortened enterprise sales cycles — compliance questions that previously meant a round-trip through email and a signed DPA were increasingly answered inside the product itself.
User Research & Iterative Refinements:
User Research & Iterative Refinements:
→
→
→
Early prototypes leaned toward full automation — a short grace period after which unreviewed variants launched by default, minimising friction at the cost of visibility;
Early prototypes leaned toward full automation — a short grace period after which unreviewed variants launched by default, minimising friction at the cost of visibility;
→
→
→
Interviews with growth marketers revealed the opposite need. Marketers didn't want less control — they wanted fastercontrol. "I don't want the AI to need less of me. I want to get through my review in ten seconds instead of ten minutes,"one Head of Growth put it;
→
→
→
Microcopy was rewritten to state consequence explicitly rather than softly — "Launch to 100% of visitors" instead of a generic "Publish," every approval screen naming the exact audience a change would reach before it reached them;
→
→
→
A/B testing the onboarding flow itself — appropriately — showed that restructuring the first-review screen increased completion of a marketer's first approved experiment by 19%, with no measurable drop in completion among returning users.
Expanded Integrations & Industry Credibility:
Expanded Integrations & Industry Credibility:
→
→
→
Platform logos (Shopify, Webflow, WordPress, Wix, BigCommerce, and a dozen others) were surfaced as a first-class visual proof point during onboarding — answering the first hesitation any non-technical marketer has: will this even work with what I already run?
Platform logos (Shopify, Webflow, WordPress, Wix, BigCommerce, and a dozen others) were surfaced as a first-class visual proof point during onboarding — answering the first hesitation any non-technical marketer has: will this even work with what I already run?
→
→
→
Third-party validation — G2 reviews, named case studies with real uplift numbers — was pulled directly into the marketing site and sales materials, rather than kept in a separate reviews page most prospects never visit;
Third-party validation — G2 reviews, named case studies with real uplift numbers — was pulled directly into the marketing site and sales materials, rather than kept in a separate reviews page most prospects never visit;
→
→
→
Named client outcomes (Reima, Sunday, Aikido, Dexxter) became the central credibility anchor across the site — in performance marketing, where every growth lead already knows every other growth lead, one recognisable logo does more convincing than another feature list;
Named client outcomes (Reima, Sunday, Aikido, Dexxter) became the central credibility anchor across the site — in performance marketing, where every growth lead already knows every other growth lead, one recognisable logo does more convincing than another feature list;
→
→
→
The onboarding scan itself was designed as a shareable artifact — a marketer's first set of AI-generated insights could be screenshotted and sent straight to a founder or CMO, turning a personal trial into internal buy-in before any budget conversation happened.
The onboarding scan itself was designed as a shareable artifact — a marketer's first set of AI-generated insights could be screenshotted and sent straight to a founder or CMO, turning a personal trial into internal buy-in before any budget conversation happened.
The Measure & Test phase shifted Dalton from an AI tool that suggests experiments to the system a growth team trusts to touch their live site without supervision. The metrics that mattered most were never session length or engagement — they were approval-to-launch rate, time to first trusted insight, and experiments sustained per team per month. Every iteration was tested against those three.
The Measure & Test phase shifted Dalton from an AI tool that suggests experiments to the system a growth team trusts to touch their live site without supervision. The metrics that mattered most were never session length or engagement — they were approval-to-launch rate, time to first trusted insight, and experiments sustained per team per month. Every iteration was tested against those three.
Impact
Impact
Impact
Dalton wasn't built from zero. Customers were already running experiments on production sites, revenue depended on trust that couldn't be interrupted, and the brand already had a public voice before any of this design work began. The job wasn't to invent a product — it was to raise how much marketers trusted an AI system that was already touching their live websites, without ever giving them a reason to hesitate at the approval screen.
Dalton wasn't built from zero. Customers were already running experiments on production sites, revenue depended on trust that couldn't be interrupted, and the brand already had a public voice before any of this design work began. The job wasn't to invent a product — it was to raise how much marketers trusted an AI system that was already touching their live websites, without ever giving them a reason to hesitate at the approval screen.
What that work actually changed sits across four different kinds of evidence.
What that work actually changed sits across four different kinds of evidence.
The metric that changed the sales conversation
The metric that changed the sales conversation
Approval-to-launch rate — the share of AI-generated variants a marketer ships without requesting an edit — became the single number the team optimised design around. It isn't a vanity metric. It's the difference between Dalton being a tool that suggests ideas and Dalton being a tool a growth team actually runs on. Every screen in the review flow was redesigned around one job: closing the gap between an AI suggestion and something a marketer would have written herself.
Approval-to-launch rate — the share of AI-generated variants a marketer ships without requesting an edit — became the single number the team optimised design around. It isn't a vanity metric. It's the difference between Dalton being a tool that suggests ideas and Dalton being a tool a growth team actually runs on. Every screen in the review flow was redesigned around one job: closing the gap between an AI suggestion and something a marketer would have written herself.
The behaviour marketers described unprompted
The behaviour marketers described unprompted
Founders and growth leads running lean teams started describing Dalton the same way in interviews and testimonials, without being asked to: "like having a CRO specialist overnight, and you just pick the winners in the morning." That framing wasn't written into any copy — it emerged because the product made the AI's overnight work legible enough for a human to review in minutes instead of hours. The design didn't invent the value proposition. It made a value proposition marketers were already living into something they could put into words.
Founders and growth leads running lean teams started describing Dalton the same way in interviews and testimonials, without being asked to: "like having a CRO specialist overnight, and you just pick the winners in the morning." That framing wasn't written into any copy — it emerged because the product made the AI's overnight work legible enough for a human to review in minutes instead of hours. The design didn't invent the value proposition. It made a value proposition marketers were already living into something they could put into words.
The signal from teams with no CRO background at all
The signal from teams with no CRO background at all
The clearest proof the interface worked wasn't from experienced testing teams — it was from founders and solo marketers who had never run an A/B test before Dalton and had no design or engineering background to fall back on. Multiple early customers reported finding a winning variant, like a tagline or hero headline, within their first week — a result that traditionally required a dedicated CRO hire and months of setup. For a product built explicitly for teams without that headcount, that's the outcome the entire design existed to produce.
The clearest proof the interface worked wasn't from experienced testing teams — it was from founders and solo marketers who had never run an A/B test before Dalton and had no design or engineering background to fall back on. Multiple early customers reported finding a winning variant, like a tagline or hero headline, within their first week — a result that traditionally required a dedicated CRO hire and months of setup. For a product built explicitly for teams without that headcount, that's the outcome the entire design existed to produce.
Built for teams that scale past their first experiment
Built for teams that scale past their first experiment
The real test wasn't the first approved variant — it was the tenth. Teams that trust the review screen once tend to keep coming back; teams that don't, quietly stop checking. The design system, the review flow, and the control-room view were all built around that second kind of usage: not a one-time trial, but a habit a growth team builds into its week. Whether that habit holds at scale, across hundreds of workspaces running dozens of experiments each, is the outcome still being measured.
The real test wasn't the first approved variant — it was the tenth. Teams that trust the review screen once tend to keep coming back; teams that don't, quietly stop checking. The design system, the review flow, and the control-room view were all built around that second kind of usage: not a one-time trial, but a habit a growth team builds into its week. Whether that habit holds at scale, across hundreds of workspaces running dozens of experiments each, is the outcome still being measured.
CRO tooling has historically split into two camps: powerful platforms that assume a team you don't have, and AI shortcuts that assume trust you haven't earned yet. The opportunity for Dalton was to sit between them — powerful enough to replace a specialist, transparent enough that replacing them never felt like a risk. The design work was built to hold that middle position, not drift toward either extreme.
CRO tooling has historically split into two camps: powerful platforms that assume a team you don't have, and AI shortcuts that assume trust you haven't earned yet. The opportunity for Dalton was to sit between them — powerful enough to replace a specialist, transparent enough that replacing them never felt like a risk. The design work was built to hold that middle position, not drift toward either extreme.
Copyright © 2026 Monochromatic Ltd. All rights reserved.
Workhub, 77 Lower Camden Street, Dublin, D02 XE80, Ireland
Copyright © 2026 Monochromatic Ltd. All rights reserved.
Workhub, 77 Lower Camden Street, Dublin, D02 XE80, Ireland
Copyright © 2026 Monochromatic Ltd. All rights reserved.
Workhub, 77 Lower Camden Street, Dublin, D02 XE80, Ireland