SIGRID — AN AGENT THAT FIXES YOUR WEBSITE AND PROVES THE LIFT
You’re paying for traffic your page wastes.
One snippet on your site. Sigrid reads your analytics, runs A/B tests you approve, and reports only the lift she can prove. Launching summer 2026 — the waitlist goes first.
Webflow · Framer · Next.js · Shopify · WordPress — one snippet, nothing to replatform
Premium products for modern living
Free returns, always. Checkout in 60 seconds.
Verified +0.0% revenue per visitor — against your untouched 10% holdout. Predicted +4–9.
✓ n=24,610 · 95% CI 2.4–12.9
TRAFFIC THROUGH A SIGRID EXPERIMENT — SIMULATION · HOVER A LANE
TEST 14 · measuring — split fixed until the verdict
■ VARIANT 45% ■ ORIGINAL 45% ▨ HOLDOUT 10% — untouched
FROM SIGRID’S HYPOTHESIS LIBRARY — SAMPLES
YOU’VE PROBABLY SAID ONE OF THESE
01
“CPC keeps climbing. Conversions don’t.”
cost / conversions
02
“I know I should be testing. It’s been ‘next sprint’ for a year.”
four quarters of “next sprint” · zero tests run
03
“The last tool claimed 30% lift. Revenue didn’t move.”
the claim / your revenue
01 · HOW IT WORKS
Fourteen days. One thread.
Sigrid runs the full loop on your site — and the whole thing happens where you already work. One cycle, end to end, scrubbed by your scroll. (Illustrative — Sigrid is pre-launch.)
READ
Sigrid · Day 1 · 08:02
I pulled 30 days of GA4, Clarity and Stripe, and read your five competitors’ pricing pages. You don’t have an average visitor — you have three types, and your plan table loses 61% of the mobile ones. Starting there.
REASON
Sigrid · Day 1 · 08:04
Three hypotheses, ranked by expected lift × confidence ÷ effort. Top: collapse the plan table to two options, one highlighted. Predicted +9 to 14%, low risk. Approve?
✓ Approved — you, 08:31
RENDER
Sigrid · Day 2 · 07:15
Variant live to 45% of traffic. Your 10% holdout stays untouched — they never see any variant. Brand drift: 0.04 of 0.30 cap.
READ AGAIN
Sigrid · Day 6 · 09:00
Nothing to report. n=9,180 of ~21,000 needed. This is what honest measurement looks like — most of it is waiting.
READ AGAIN
Sigrid · Day 9 · 09:12
Trending +9.8%. Not enough visitors yet to prove it. I don’t call winners early. Holding.
READ AGAIN
Sigrid · Day 14 · 08:40
Verified: +11.4% trial signups (95% CI: 4.2–18.9) against the holdout. Promoting to all traffic except the holdout.
REMEMBER
Sigrid · Day 14 · 08:41
Noted for next time: your visitors respond to clarity, not urgency. Re-weighting future hypotheses for this site.
launch news first · €79 audits reserved through the same list
02 · THE BASELINE — A CONVERSION AUDIT
Before the first test, the homework.
Sigrid doesn’t test blind. Day 0 builds a working model of your business — who lands on your site, what the market around you claims, what your brand sounds like — so the first hypothesis is already an informed one. Every subscription starts here; the €79 audit is this baseline, standalone.
STEP 1 · CONNECT
Plug in your tools
Read-only access to the tools you already run — two minutes. The snippet fills the gaps they can’t see.
STEP 2 · READ
30 days of behavior
Funnels, drop-offs, device split, revenue per session. Where money leaks — measured, not guessed.
e.g. 68% mobile checkout drop-off · found, not assumed
STEP 3 · RECOGNIZE
Your actual visitors
Not one average visitor — segments with different jobs. Each gets its own variants: personalization is Day 1, not an upsell.
41% ad price-checkers · 33% comparison shoppers · 26% returning
STEP 4 · STUDY THE MARKET
Competitors, public pages only
What your competitors claim, charge and promise — and the gaps you can take. Nothing behind a login, ever.
e.g. 4 of 6 show shipping cost on the product page — you don’t
STEP 5 · FINGERPRINT THE BRAND
Voice, tone, image style
Your copy and imagery become a brand corpus — what the drift caps protect. Variants that don’t sound like you never ship.
caps: voice 0.07 · image 0.10 · layout 0.20 · CTA 0.10
STEP 6 · BASELINE THE MATH
What your traffic can prove
The power math: which lifts are provable on your traffic, and how fast — known before you pay, not discovered after.
floor: ~15k sessions/mo · below it she’ll tell you no
The deliverable: the ranked hypothesis queue ↓ — every score traced to these six sources.
She reads from everywhere you measure.
Five sources at launch; the rest ship in the order customers vote. All read-only, aggregates not individuals — testing requires no PII.
READS — LIVE AT LAUNCH
READS — QUEUED · YOUR VOTE SETS THE ORDER
Tallies are public — vote on the roadmap →
SHIPS VARIANTS VIA
SPEAKS IN
MCP-FIRST
Sigrid is a Model Context Protocol server. The chat apps are conveniences; the protocol is the contract.
03 · THE SIGNATURE
How Sigrid counts
Every A/B testing tool promises lift. Almost none will show you how the number was made. Sigrid’s numbers are checkable — that’s the product.
The math, concretely: at 60k sessions/mo and a 2% baseline, a 20% lift verifies in about three weeks. A subtler 10% lift takes ~11. Below ~15k sessions/mo, proof stops fitting in useful time — that’s the floor, and she’ll tell you so.
RULE 1
A permanent holdout
10% of your visitors never see a variant — ever. Every lift number is measured against them — a permanent control group — not against last month, not against a model.
RULE 2
The honesty gate
Below statistical power, Sigrid says “not yet” — never “winner.” However good the trend looks.
RULE 3
A public ledger
Verified lift across customers gets published live — including the inconclusive tests. The dashboard launches with the product.
SIGRID · AUDIT LOG · WHAT AN HONEST RESULT LOOKS LIKE
TEST 11 — headline rewrite14 days
samplen=8,412
lift+2.1% (95% CI: −3.0 to 7.4)
verdictINCONCLUSIVE
actionrolled back · not counted
“Losing tests are data. I don’t hide them.” — Sigrid
This is the reporting you’d get. Join the waitlist →
04 · SAFETY
She won’t break your site.
Or your brand. Every variant runs a five-layer gauntlet before a single visitor sees it — and brand drift is a number with a cap, not a promise.
Brand drift, measured per dimension
Cumulative, per page-week cap 0.30
Scale: 0 = identical to your page today · 1 = unrecognizable. Scored per dimension by the review models — 0.07 on voice is a different adjective, not a different personality.
A variant that crosses any cap is rejected automatically, with the comparison in your audit log.
You hold the wheel
Copilot is the default: Sigrid proposes, you approve — one tap in Slack. Autopilot is earned, per test type, after verified wins on your account. You can take it back at any time, and the audit log always shows which mode shipped what.
One snippet. No replatform.
Webflow, Framer, Next.js, Astro, Shopify, WordPress, plain HTML — if it serves a page, Sigrid can work on it. Stores too: product pages to checkout, measured in revenue per session, net of refunds.
<head>
<meta charset="utf-8"/>
<title>Your store</title>
+<script src="cdn.heysigrid.com/s.js" data-site="you" async></script>
</head>
<body> — everything else stays yours </body>
1 line added · 0 lines touched
<8KB gzipped · no React · anti-flicker by design · remove it and nothing remains
Lives in your Slack.
Hypotheses, approvals, results — one thread, not another dashboard to forget. The web app is the audit trail; Slack is the conversation. Email digest if you prefer it quiet.
Sigrid APP Day 9 · 09:12
Pricing-page test is trending +9.8% — below statistical power. Holding to the pre-set horizon.
Sigrid APP Day 14 · 08:40
Verified: +11.4% trial signups (95% CI: 4.2–18.9) against your holdout. Winner promoted — the holdout stays out.
✓ holdout-verified · n=21,408
Slack + email at launch · Discord, Teams, Telegram when customers vote for them
Or in your Claude.
Sigrid is an MCP server — connect her to Claude or any MCP-capable tool and ask about experiments, approve variants, pull the audit log. Wherever you already work, she’s a tool call away.
you › status of test 14?
sigrid › Day 9. Trending +6.1% revenue per visitor — below power, n=15,832 of ~24,600. Holding; I don’t call winners early.
you › ping me at the verdict
sigrid › Noted. Verdict lands ~Day 14 — here and in Slack.
MCP-first · one protocol, every MCP-capable tool — IDEs and agents included
Is Sigrid for you?
YES, IF
- ✓ Your site does roughly 30k+ sessions a month (or ~300+ conversions)
- ✓ You spend on ads or other paid acquisition
- ✓ You run a founder-led SaaS or DTC store
- ✓ Nobody on your team owns conversion
NOT YET, IF
Under ~15k sessions a month, no tool can prove lift inside a useful window — Sigrid included. She’ll tell you, not sell you.
Leave your email anyway → — Sigrid checks back when your traffic can carry proof.
05 · PRICING
Flat price. No meter.
The more traffic you have, the faster Sigrid can prove lift — and the price stays the same. No per-impression math, no credits, no surprises.
STANDARD — MOST SITES NEED EXACTLY THIS
€249/mo
At launch · 30-day money-back
- — 1 site, up to ~100k sessions/mo
- — 5 concurrent experiments
- — Full loop: read, reason, render, verify, remember
- — Holdout-verified reporting
- — Slack, email + Claude (MCP)
GROWTH
€499/mo
At launch · 30-day money-back
- — 3 sites, up to ~500k sessions/mo
- — 15 concurrent experiments
- — Priority hypothesis queue
- — Everything in Standard
CUSTOM
Yours
Agencies, portfolios, >500k sessions
- — More sites, more traffic, more experiments
- — Same loop, same holdout, same honesty gate
- — Pricing shaped to your portfolio
read by the founder · answered by a human
Session figures are soft ceilings, not meters — cross one and we talk tiers; nobody gets surprise-billed. · +€99/mo per extra site or extra 5-experiment pack · 30-day money-back on every plan · no free tier, on purpose
Proof before payment.
The €79 one-time conversion audit (available at launch) is the table on the right, built for your site: who lands there, what to test first, and how fast your traffic can prove it. Credited to your first month.
Reserve an audit via the waitlist →why 0.72 — 31% of checkout exits happen at the shipping step · 4 of 6 competitors show shipping on the product page · “surprise shipping” recurs in your reviews
score = expected lift × confidence ÷ effort · every score traces to sources: your analytics · competitor public pages · your reviews
Questions founders actually ask
When does Sigrid launch — and when would I pay?
Soft launch is planned for summer 2026; the waitlist hears first and gets first access, in order. Joining costs nothing and commits you to nothing. You won’t pay anything before Sigrid is live on your own site, and the 30-day money-back guarantee starts then — not at signup.
Will the snippet slow down or break my site?
The snippet is under 8KB gzipped, loads async, and ships no framework. Anti-flicker by design: she briefly holds only the elements she’s changing — never the whole page. A hard timeout guarantees the worst case is your original page, unchanged. Every variant passes a render check before any visitor sees it, and a permanent rollback exists for every change. Remove the snippet and your site is exactly as it was.
Does she work on Shopify and ecommerce stores?
Yes. Same loop, different objective: on a store Sigrid works product pages, carts and checkout flows, and measures revenue per session net of refunds — not just clicks. The snippet does the work on every platform at launch; native Shopify and Webflow publishing is queued on the public roadmap.
What does Sigrid read, and where does my data live?
She reads aggregated behavioral data from the tools you connect — GA4, Microsoft Clarity, PostHog, Stripe — plus her own snippet events. Before the first hypothesis she also does the homework: your competitors’ pages (only what’s public — nothing behind a login) and the reviews or support themes you point her at, to work out who lands on your site and why. EU-built, GDPR-first, hosted in the EU. Testing does not require collecting personally identifiable information.
What happens when a test loses?
It gets rolled back and reported. In Sigrid’s words: “Losing tests are data. I don’t hide them — they’re in your audit log, and they count on the public dashboard.”
Who approves changes to my site?
You do. Copilot mode is the default: Sigrid proposes, you approve in Slack or the dashboard. Autopilot is earned per test type after verified wins on your account — and you can take it back at any time.
What if I have less than ~30k sessions a month?
Honest answer, with the math: under ~15k sessions a month, no tool can prove lift inside a useful window — including Sigrid. Between 15k and 30k it works, just slower. From ~30k up, proof gets fast. Wherever you are, she’ll show you the numbers for your traffic before asking for a subscription.
Why is there no free tier?
Free tiers optimize for signups. Sigrid optimizes for proof. There’s a 30-day money-back guarantee instead — if she hasn’t earned her keep in your first month, you get it back.
How does she not sound like AI slop on my site?
Brand drift is measured per dimension with hard caps — voice 0.07, image 0.10, layout 0.20, CTA 0.10, cumulative 0.30 per page-week (0 = identical to your page today, 1 = unrecognizable). A stylometric review layer reads every variant against your brand voice corpus before it ships. Variants that drift get rejected, with the comparison in your audit log.
How does the 30-day money-back actually work — and what if I leave?
One email to support@heysigrid.com inside your first 30 days gets a full refund, no questionnaire. If you leave — refund or not — every variant comes down, your site is exactly as it was, and you keep the audit log of everything that ran. Leaving has to be cheap for staying to mean anything.
“I’ll show you what I’d test before you pay me a subscription. And when I report a number, it will be one I can defend.”
— Sigrid
Join the waitlist
Launch news first. €79 audits are reserved in order from this list. Costs nothing, commits you to nothing.