Dalton
Dalton is an AI-driven conversion optimization co-pilot for marketing teams without developers. It scans your site, proposes experiments, generates and ships variants without touching your code, then shifts traffic towards the winners.
What is Dalton?
Dalton is an AI-driven conversion rate optimisation and A/B testing platform that its publisher describes as a CRO co-pilot. It is built by FirstMate BV, a Belgian company founded in 2024 in Ghent by Vincent Declercq and Ward Van Laer. Setup is a single JavaScript snippet placed in the page head, compared by the site to installing Google Analytics and announced as a one-minute job; the script is said to load in under 50ms with no impact on Core Web Vitals or SEO. The product works in four movements. It first scans the site, the brand and the connected analytics to find where visitors drop off, and looks at competitors' winning streaks for ideas. It then proposes experiments: a Brand Intelligence layer reads your existing copy to learn tone and visual identity, so variants come out on-brand without uploading a style guide. Third, it generates those variants, from headlines and calls to action to layouts, social proof and product descriptions, or builds whatever you describe in a plain-language prompt. Finally it optimises continuously: instead of waiting for statistical significance, a multi-armed bandit algorithm moves traffic towards the variants that convert and retires the ones that do not, with results announced in two to four weeks rather than three to four months. Two guardrails frame all of this. Nothing goes live without your approval, and every suggestion can be edited or rejected first. Changes are also applied as an overlay, so the original code is never modified and an experiment can be paused or stopped instantly. The commercial differentiator is the traffic floor. Dalton claims it works from 3,000 to 5,000 monthly visitors per page, which it presents as ten to twenty times lower than traditional testing suites. It runs on Shopify, WordPress, Webflow, Magento, Framer or custom builds, detects each page's language and generates variants in that language, and exposes an MCP server so the platform can be driven from an AI assistant such as Claude. The company raised a EUR 1M pre-seed round announced in September 2025, and displays a 4.6/5 rating over 8 G2 reviews.
What it does
- Scans your website, brand and analytics to surface high-impact experiment ideas
- Generates on-brand variants: headlines, calls to action, layouts, social proof and product descriptions
- Deploys changes with no code, as an overlay that never modifies your source code
- Shifts traffic towards winning variants automatically through multi-armed bandit allocation
- Stops underperforming variants on its own, without waiting for classical statistical significance
- Measures conversions and reports uplift in real time on a revenue-oriented dashboard
- Builds an experiment from a plain-language prompt when you already have your own idea
When to use Dalton / When not to
A quick filter to help you decide if Dalton is the right fit.
When to use Dalton
- Marketing and growth teams that want to run experiments without a developer or a dedicated CRO specialist
- E-commerce and SaaS sites with moderate traffic, from roughly 3,000 to 5,000 monthly visitors per page
- Teams running heavy paid acquisition that need landing pages to keep pace with their ad creative
- Operators of Shopify, WordPress, Webflow, Magento or Framer sites who cannot get changes into the source code
- Multilingual sites, where page language is detected automatically and variants are generated in that language
When not to use Dalton
- Small sites below the traffic floor Dalton sets itself, under 3,000 to 5,000 monthly visitors per page
- Teams already equipped with VWO or Optimizely: the documentation warns that two testing tools on the same pages cause conflicts
- Teams running in-house A/B tests on the same pages, where the docs advise choosing one testing method per page for accurate attribution
- Anyone needing back-end or server-side experiments: Dalton only overlays front-end rendering and never touches the original code
- Organisations that need a contractual guarantee that generated variants meet their legal, regulatory or brand requirements, which the terms explicitly disclaim
How to use Dalton
A typical end-to-end flow, from setup to results.
- Run the free website scan before signing up: no credit card is needed and the first findings appear in about 25 seconds
- Install the single JavaScript snippet in the head of your site, a one-minute job comparable to adding Google Analytics
- Let Dalton scan the site and read the CRO Report, which lists conversion opportunities within minutes
- Accept a suggested experiment in one click, or describe your own idea in a plain-language prompt
- Adjust the variant in the visual editor, or edit the HTML directly, previewing on both mobile and desktop
- Approve the experiment: nothing goes live until you explicitly validate it
- Set your conversion goals, from purchases and clicks to custom GA4 events
- Let the multi-armed bandit reallocate traffic and follow the uplift in the results dashboard
- Optionally connect GA4, included from the Pro plan, and import the supplied GTM container template
- Optionally connect the MCP server to Claude Desktop to drive Dalton in natural language; on multilingual sites, add each language URL separately
Pros & Cons
Pros
- Genuinely light installation: one script in the page head, and no developer dependency afterwards
- Low claimed traffic floor of 3,000 to 5,000 monthly visitors per page, where traditional suites ask ten to twenty times more
- The AI supplies both the ideas and the variants, so there is no hypothesis workshop and no design brief to run
- Explicit human guardrail: approval is mandatory before anything goes live, and every suggestion can be edited or rejected
- Non-destructive overlay: the original code is never modified and an experiment can be stopped instantly
- Multi-armed bandit allocation wastes less traffic on losing variants than a fixed 50/50 split
- Unusual documentary transparency for a company this size: a full DPA, five named sub-processors with storage location and transfer mechanism, EU processing claimed, deliberately thin collection with no name, no email and no IP kept in analytics, detailed technical docs and named case studies
Cons
- Very young supplier: company incorporated in September 2024, domain registered in February 2025, and only 8 public G2 reviews
- Monthly prices are never published: the pricing page defaults to annual billing and the Monthly toggle reveals no amount
- Steep 6.5x gap between Essential (99 USD, under 10,000 sessions) and Pro (648 USD, under 50,000 sessions), with GA4 integration reserved for Pro
- The traffic floor rules out small sites, and the tool conflicts with VWO, Optimizely or in-house A/B tests running on the same pages
- No security certification of its own, neither ISO 27001 nor SOC 2: the only certification shown on the site belongs to Cloudflare, a sub-processor
- No documented way to opt out of model training, while the terms grant a perpetual right over anonymised customer data
- No legal notice page and no contact page, commercial contact routed through a personal Calendly link, support limited to CET business hours, no mobile app or browser extension, and a Team plan cited in the docs that is absent from the pricing page
Pricing & Plans
Dalton follows a freemium model. A permanently free tier, Try Dalton, covers your first 5,000 sessions with no credit card, two experiments per page, AI prompts and a continuous stream of suggestions. The cheapest paid entry point is the Essential Plan at 99 USD per month, billed annually, for fewer than 10,000 monthly sessions; the Pro Plan is 648 USD per month, also billed annually, for fewer than 50,000 sessions; Enterprise is quoted on request. It must be stressed that no monthly-billing price is published anywhere: the pricing toggle sits on Annual -10% by default and monthly amounts are never displayed. Pricing is indexed on session tiers rather than actual usage within the plan limits. Prices exclude tax, invoices are payable within 30 days, self-service subscriptions are charged at the start of each period, and late payment carries 1.5% per month. Subscriptions renew automatically unless cancelled in writing 30 days ahead, price increases are notified 60 days in advance, term subscriptions remain payable in full, and fees paid are non-refundable. Sustained overage of a tier moves the account to the next one with 30 days' notice.
- first 5
- 000 sessions
- 2 experiments per page
- AI prompts and a continuous stream of suggestions
- fewer than 10
- 000 sessions per month
- 2 experiments per page
- AI prompts
- continuous suggestions
- everything in Essential
- fewer than 50
- 000 sessions per month
- GA4 integration and Slack support
- everything in Essential and Pro
- custom session volume
- priority Slack and email support
- No monthly-billing price is published for any paid tier
- and the MCP server documentation mentions a Team plan that does not appear on the pricing page
Data, GDPR & hosting
A consolidated view of how Dalton handles your data.
GDPR overview
GDPR implementation is concrete and documented. Dalton is established in Belgium, so no Article 27 representative is required and none is named. The privacy policy cites consent, contractual necessity and legitimate interest as legal bases, lists the rights of access, rectification, erasure, restriction, objection, withdrawal of consent and portability, and promises a reply within 30 days at team@getdalton.com. A full Article 28 DPA is published and incorporates the sub-processor page by reference; transfers outside the EU rely on Standard Contractual Clauses, with 14 calendar days' notice to object to a new sub-processor. A US state-law addendum makes Dalton a CCPA/CPRA service provider that neither sells nor shares data. Cookies are set only after analytics consent. The privacy policy is dated March 2026, the terms, DPA and sub-processor list July 2026. The documentation states plainly: Is Dalton GDPR/CCPA compliant? Yes.
Who owns the data?
Under the terms, the customer owns all Customer Data and all Customer IP, and Customer Data is treated as the customer's confidential information; Dalton acquires no rights over Customer IP. Dalton keeps every right in the Dalton Technology, including the outputs of its machine-learning models. Its licence to process customer data is limited to delivering the service and expires with the subscription, after which data is returned or deleted under the DPA. For visitor data collected on the customer's website, Dalton acts as a processor under Article 28 GDPR, working on the customer's documented instructions.
Reuse rights
Clause 6.2(a) grants Dalton a licence to process Customer Data solely to deliver the service, and that licence expires with the subscription. Clause 6.2(b) goes further: the customer authorises Dalton to irreversibly anonymise or aggregate the data, and grants a perpetual, worldwide, royalty-free right to use the result to operate, benchmark and improve its products, including training machine-learning models, producing statistical insights and reports, and running Dalton's own marketing. Clause 6.2(c) confines that perpetual right strictly to anonymised or aggregated data and never to identifiable or pseudonymous data; Dalton also undertakes not to sell Customer Data and not to use it for third-party advertising. Visitor data is itself pseudonymous: a random session identifier, no name, no email address, and no IP address kept in the analytics. On the customer's side, ownership is unaffected: the customer keeps its data and needs no permission to reuse or export it. No option to exclude your data from model training is documented anywhere on the site.
Data retention & training
Hosting summary
The documentation states that all Dalton data is processed and stored exclusively within the European Union, with eu-central as the primary data zone. Encryption is TLS/SSL in transit and AES-256 at rest. Five sub-processors are named together with their storage location: Cloudflare (global edge network, routing and transient buffering, no persistent storage of personal data at the edge), Vercel (algorithm and API, European servers only), Redpanda (event streaming, AWS eu-central-1 in Frankfurt), Tinybird (visitor event storage and querying, Google Cloud europe-west2 in London, United Kingdom) and Supabase (authentication, experiment data and results, eu-central-1). Transfer mechanisms are the EU-US Data Privacy Framework plus Standard Contractual Clauses for Cloudflare and Vercel, SCCs for Redpanda and Supabase, and the UK adequacy decision plus SCCs for Tinybird. One tension deserves attention: the claim of storage exclusively within the EU is contradicted by Tinybird's UK location. Finally, the web server IP, an Amazon anycast node geolocated in the United States, says nothing about where the platform hosts customer data.
Things to keep in mind
Risks and trade-offs to weigh before adopting Dalton.
- Stale legal address: the terms and the footer give Sint-Pietersnieuwstraat 11, while the Belgian company register and VIES have listed Korianderstraat 2 box G, 9000 Ghent since 19/06/2026. The company name itself appears in four spellings across the site and the registers: Firstmate BV, FirstMate BV, FirstMate and BV FirstMate.
- The only certification displayed on the site, an active EU-US Data Privacy Framework certification, belongs to Cloudflare, a sub-processor, and not to Dalton. Dalton itself holds no ISO 27001, SOC 2 or Cyber Essentials certification of its own.
- Contradictory hosting claims: the documentation states that data is processed and stored exclusively within the European Union, while the sub-processor table places Tinybird in the United Kingdom, on Google Cloud europe-west2 in London. The site's server IP, 216.150.1.1, is also an Amazon anycast node geolocated in the United States, which describes the marketing site's delivery and not where customer data lives.
- No monthly-billing price is published: the pricing toggle sits on Annual -10% by default, and the 99 USD figure is a monthly rate billed annually. The MCP documentation also advertises a Team plan that does not exist on the pricing page.
- Training with no exit: the terms grant a perpetual, worldwide right to train machine-learning models on irreversibly anonymised or aggregated customer data, and no opt-out option is documented anywhere on the site. The publisher also disclaims any warranty that generated variants meet your legal, regulatory or brand requirements.
- Second-hand domain and thin verifiable trust signals: the first Wayback capture, dated 15/07/2013, is a parked GoDaddy page, with a gap from 2016 to 2025 and re-registration on 12/02/2025. The 4.6/5 rating over 8 G2 reviews shown on the site could not be verified at source, the Instagram account appears only in the structured data and never in visible content, and homepage testimonials are duplicated verbatim, two of them signed with a first name only.
- No legal notice page and no contact page exist, and the only commercial channel is a third-party Calendly link in a founder's name. There is a human risk too: delegating copy and layout decisions to a model that rewrites your pages continuously can quietly erode a team's own understanding of why its customers convert.
Setup & Integrations
Technical difficulty
Very low for the core setup. Installation is a single JavaScript snippet in the page head, compared by the publisher to Google Analytics: one minute on the site, five in the documentation. The first scan returns findings in about 25 seconds, and no development skill is needed afterwards, since the visual editor is no-code with direct HTML access available. Optional steps are more technical: connecting GA4, importing the GTM container template and replacing the measurement ID by hand, declaring custom dimensions, allowing domains in your CSP, and adding each language URL separately on multilingual sites.
Deployment
Integrations
Behind Dalton
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Dalton.
Frequently asked questions
Do I have to approve changes before they go live?
Will Dalton slow my website down?
How quickly do results appear?
How much traffic do I need?
Do I need a developer?
Can Dalton break my site?
How is it different from VWO or Optimizely?
What can I actually test?
Is there a free plan?
Where is my data stored?
Should you pick Dalton?
Dalton sets out to make continuous experimentation available to teams that have neither a developer on call nor a CRO budget, and on that promise it is convincing. The real differentiator is not AI generation, which competitors are adding too, but the combination of a low traffic floor, 3,000 to 5,000 monthly visitors per page, with multi-armed bandit allocation that finds winners without waiting for classical significance. Add a one-minute install, a mandatory human approval step and an overlay that never rewrites your source code, and the cost of trying it is genuinely low. Maturity is the counterweight. FirstMate BV was incorporated in September 2024, the domain was registered in February 2025, the company has raised a EUR 1M pre-seed round, and the public review base is eight G2 ratings. That is a young supplier, and buyers should size their commitment accordingly. Documentary hygiene, on the other hand, sits above average for that stage: a full Article 28 DPA, five named sub-processors with their storage location and transfer mechanism, EU processing claimed, and a deliberately thin data collection with no names, no email addresses and no IP addresses kept in the analytics. Four reservations remain. No monthly-billing price is published anywhere. The jump from Essential to Pro is a factor of 6.5, with GA4 integration reserved for the higher tier. Dalton holds no security certification of its own. And the terms grant a perpetual right to train models on anonymised data, with no documented way to opt out. Recommended for marketing and e-commerce teams with moderate traffic who want to ship many experiments without engineering time, provided they check the annual commitment and accept a young vendor.
- Choosing a selection results in a full page refresh.
- Opens in a new window.