Never watch another session replay again
One ~21KB script, zero personal data. It does the watching so you don't have to.
<script async src="https://harvv.com/px/YOUR_KEY/pixel.js"></script>
We caught Klaviyo slowing a store down before anyone reported it.
Median 2.7s, 9.8s at the 95th percentile, on pages the store didn't control. Named the URL, the latency, and the shoppers hit.
No more heatmaps, dashboards, or hours of video.
Replay records video and asks you to watch it. Heatmaps hand you a colour wash and ask you to interpret it. Both bill monthly for work nobody has time to do.
An MCP that just tells you what you want to know
Connect Harvv to the assistant you already use and ask in your own words: why aren't people clicking add to cart? which page got slower this week? You get the case, the numbers, and the fix prompt back in the chat.
5 minutes and you're done
next.js, wordpress, shopify, framer, plain html
Install one line of code
Your unique site key. First events appear within seconds.
Connect your MCP
harvv.com/mcp
Paste the connector URL into Claude, ChatGPT, Gemini, or any other LLM tool.
Done
Cases start arriving on their own. Ask for one any time.
See what's broken on your site. Free, in 60 seconds.
No pixel, no signup. Drop in a URL and get a real teardown: dead clicks, tap targets, speed, and AI-search readiness, ranked by what costs you the most.
Real things we've found in live client data
Free visitors added to cart at 8.3%. The visitors this store paid for: 2.9%. Paid traffic was 3× less likely to reach the cart than traffic that found the store on its own, and the gap held across every product page we measured.
What we told them to do: Rebuild the ad audiences around the segments that match organic buyers, and stop paying for lookalikes that never touch the cart.
651K sessions · 14 days14% of paid-ad sessions never rendered a real person. No scroll, no pointer, no paint of the page a human would see. Full ad price paid, zero shoppers delivered, and completely invisible in the ad manager.
What we told them to do: Exclude the placements sending the bot traffic and take the click logs to the ad platform for a refund.
93K ad clicks · 30 days84% of shoppers who used on-site search never opened a single result. These are the highest-intent people on the site, typing what they want to buy, and the results page sent them away with nothing. Zero-result queries and bad ranking split the blame about evenly.
What we told them to do: Fix the top zero-result queries first, then pin the ten most-searched products to the top of results.
16,418 searches · 7 daysOnly 9.8% of engaged visitors reached the end of the page. 62% never got past the first quarter. Everything below that line, including the reviews and the guarantee, played to an almost empty room.
What we told them to do: Move the proof and the primary call to action into the first quarter of the page, then re-measure.
748K sessions · 30 days8,857 shoppers hovered a size, then froze. 87% never added to cart. The size selector rendered at 31×31px on mobile, under the 44px minimum, and there was no size guide within reach of the selector.
What we told them to do: Enlarge the tap targets to 48px and put a one-tap size guide next to the selector.
apparel store · 7 days1 in 18 shoppers had the store open in several tabs at once, flipping between it and a competitor. That is comparison shopping you can finally see while it is happening, not infer from a lost sale a week later.
What we told them to do: Show the price promise and the shipping cutoff while the shopper still has your tab open.
apparel store · 7 daysIdentify new billable work, generate the proposal, and present the fix in one click.
Harvv surfaces the issues costing your client conversions, ranked with the dollar impact.
A branded, white-label audit report under your logo, with a small “powered by Harvv.”
Send the link, embed it, or hand over the PDF. The fix prompt is attached. You scope, you bill.
We caught Klaviyo slowing a store down before anyone reported it.
Third-party scripts are the slowdowns you can't see. Harvv measures every request the page makes, not just the clicks.
On July 22, the pixel flagged that Klaviyo's settings API was answering in a median of 2.7s, and as slow as 9.8s for the worst 5%, on a client's pages. 78 shopper sessions sat waiting on a third party the store didn't control. Analytics would show a dip and no reason; a replay would show a person waiting and no cause. Harvv named the exact URL, latency, and shoppers hit. It cleared on July 30.
Every site on the pixel sharpens the baseline for the rest.
Hundreds of millions of events cancel the noise, so a finding on your site is measured against real ground truth, not a hunch.
Collecting 0 PII since day one
Privacy by architecture, not by certificate. Harvv captures no personal data, which is what makes the rest true.
No names, emails, phones, addresses, or IPs. The behavioral events table has no IP column, so an IP can never be joined to a visitor's behavior.
We respect each site's own consent banner, honor Global Privacy Control, and log every consent choice for Article 7 record-keeping.
Minimal personal data, and we honor GPC. Connecting IPs are used in memory for rate-limiting and bot detection, never stored raw.
App and PostgreSQL on Railway (US), encrypted backups in AWS S3 (US). Raw events pruned after 7 days; aggregates kept for trends.
Free until it pays for itself.
annual pricing shown · 2 months freeNever watch another session replay again
