<- all tokdocs

Jev and PostHog Visitor Personas: Cheap to Classify, Hard to Email

By , founder of Mavric and Apso

View on TikTok ->

A TikTok from Angus Sewell McCann (@angusthenontechnical) shows how to install PostHog with Claude Code, cluster site visitors into personas, classify each visitor with TypeSafe's Jev model, and send a matching follow-up email. The classification half is cheap and works as shown. The identification half, which decides whether you have an email address at all, is where the funnel shrinks, and the video spends about nine seconds on it.

How the PostHog and Jev build works

The video walks through four steps:

  1. Install PostHog on an existing landing page with a Claude Code prompt that runs npx -y @posthog/wizard@latest.
  2. Wait about two weeks for data.
  3. Ask Claude to cluster visitors into archetypes.
  4. Add a page-leave handler that sends the session to Jev, gets a persona back, and fires a matching email.

The Jev call on screen is POST https://api.typesafe.ai/v1/systemone with "model": "jev-latest", returning "choice": "investor", "confidence": 0.94. The persona lands in a custom Visitor Persona property on a HubSpot contact, and two generated emails come from a fictional company called Tallyloop.

What is Jev, and is PostHog new?

The video opens "with the release of Jev and PostHog," but only one of them is new. PostHog launched on Hacker News in February 2020 as a Y Combinator W20 company and has since raised a $12M Series A and a $15M Series B.

Jev is the new one. TypeSafe AI put it into limited early access on 15 September 2026 alongside a $40M seed round led by DCVC, and opened it to everyone on 21 September. The video posted on 29 September.

The technical details match TypeSafe's docs. Jev is a System One model, which returns typed values with probability distributions instead of generated text. The models page lists one model ID, jev-1.13.0, aliased as jev-latest and jev-preview. The response in the video comes back stamped jev-1.13.0, which is what the alias does.

How many visitors does the clustering step need?

The archetype prompt's first line reads: "This is a demo for a video with synthetic data - do not write to any database, do not file issues/tasks/entities, do not reference my own skills or clients." So the result, "Your roughly 3,300 human visitors in the last 30 days sort into 8 types," is not from a real site. The eight archetypes (Decision Maker, Investor, Job Applicant, Junior Champion, Veteran Sales Guy, Happy Customer, Competitor, Researcher) are a plausible list that real data did not produce.

A landing page with a tenth of that traffic gives Claude a few hundred sessions, which is not enough to separate eight clusters. Size the wait by session count, a few thousand, rather than by two calendar weeks.

What is the real visitor identification rate?

The system only sends mail if it can match a session to an address. The video says the tools "will only find their email like 20% of the time," over a card listing RB2B, Vector, Snitcher, Dealfront and Warmly.

RB2B's support documentation reports contact-level identification of around 30 to 45 percent of unique US-based traffic for Pro users. Independent reviews of the category put realistic person-level rates closer to 5 to 20 percent. Both figures apply to US traffic only. RB2B does not resolve visitors outside the United States to a person, so international visitors come back at company level at best.

IP enrichment has a second limit. The video's own IPinfo lookup returns "Anthropic, PBC" with AS Type "Hosting," which is the network operator behind the address. Lookups often return cloud providers, VPN exits and corporate ISPs instead of the visitor's employer.

What does it cost?

The video says companies pay "30 grand" for this kind of system. Real pricing spans a wide range. Vendr's data puts the median 6sense annual contract at $62,440 and Demandbase's at $68,591. Dealfront's visitor identification starts around €99 per month, Snitcher at $49 per month, Vector's company-level tier at $399 per month, and Warmly prices its deanonymization module near $10,000 per year.

The build itself costs little. PostHog's free tier covers 1M events and 5,000 session recordings per month with no card required. Jev charges $0.042 per million input tokens and nothing for output. The real costs are the identification vendor and email sending.

Is it legal to email identified website visitors?

The video treats consent as a data problem. It says matching is easy for visitors who already opted in, and suggests identification tools for everyone else: "If not, you could use one of these identification tools." The law treats those as different cases, as Leadfeeder's GDPR explainer describes.

Company-level identification generally sits outside GDPR because it does not process personal data. Person-level identification does, and needs consent or a documented legitimate interest assessment. In the EU, B2B cold email can rely on legitimate interest, while B2C email requires prior consent under the ePrivacy Directive. In the US, CAN-SPAM allows unsolicited commercial email if you identify the sender, include a physical address and honor opt-outs. The demo emails carry a "Tallyloop Inc., Austin, Texas" footer with unsubscribe links, which covers that.

Mail to addresses you never collected can draw spam complaints that hurt your domain's sending reputation, so send it from a separate subdomain.

The safeguards in the code

The api/visitor-left.ts handler includes two checks. A suppression rule, if (contact?.persona_email_sent) return stop("already emailed"), plus a quietFor(visitor.id, "30m") debounce, keeps one session from triggering several sends. A confidence gate, if (persona.confidence < 0.6) return stop("not sure enough"), skips uncertain classifications. Because Jev returns a probability distribution, that threshold is a number you can tune against results. A wrong persona sends an investor pitch to a job applicant, so a higher starting threshold, closer to 0.85, is safer.

The creator discloses in the video description that he works with PostHog as a channel partner, and his site lists PostHog as a "Gold partner."

Bottom line

Classifying visitors with PostHog and Jev costs a fraction of a cent per call. Getting an email address works for a minority of US visitors and almost no international ones, and person-level identification brings GDPR obligations. Measure your identification rate on your own traffic before building the email step.

Sources

Published September 29, 2026. Based on a TikTok by @angusthenontechnical. Full video notes (markdown)