Skip to content
In private rollout: onboarded personally, usually live within 24 hours.
Sarah is on the floor

Give every product page a closer.

Cervito turns browsing into assisted buying. Sarah greets, compares, bundles, hands off, and shows the revenue she helped win.

Every number on this site carries the date we measured it and a way to check it yourself.

Not ready for a call? Request an invite

A/B holdout proof
44 languages
One lightweight script
EU-hosted, sub-processors published

Pick a store. The tiles below are that catalog. Open it and talk to her for real, no login and no sales call first.

app.cervito.io/demo/northbrook

Picked for this shopper, from this catalog.

Real product pages, live prices, and reasons the shopper can trust.

Yirgacheffe

Yirgacheffe

€19

V60 Dripper

V60 Dripper

€24

Gooseneck

Gooseneck

€39

Northbrook Coffee Co.

with SarahAI

Open Northbrook Coffee Co.(opens in a new tab)

Plate 01Northbrook Coffee Co., demo storefigures illustrative

Contents

The whole product, with the limits printed.

8 things she does on a shift, each with its own page. Under every one sits the same paragraph that page carries: what makes it real, and where its edges are. Published here rather than three clicks deep, because the edges are the part worth reading before a call.

The associate

Conversational selling

She greets, recommends, and closes. She never invents a SKU, never promises free shipping you do not offer, and never quotes a price she has not verified in your live catalog.

She fires onProduct dwellCollection bounceCart stalland five more

What makes it real

Every recommendation is grounded in your live catalog, so she never surfaces a SKU you don't sell or a price that isn't current. She is identified as AI in the widget, and her answers typically arrive in two to eight seconds. There are no canned objection scripts behind her: how well she sells depends on the coaching and training you give her.

Live catalog checkgrounded
  • YirgacheffeSingle-origin · whole bean€19.00in catalog
  • V60 DripperBrew gear€24.00in catalog
  • Espresso machineShopper request·not in catalog

Plate 02Northbrook Coffee Co., demo storefigures illustrative

Cross-sell

Style boards & bundles

Complete-the-look sets, one tap

What makes it real

Style boards are merchant-curated, not generated. You own the combinations, which means the pairings are always sensible, always on-brand, and never invented. The bundle suggestions read the shopper's current cart, so the cross-sell fits what they're actually buying rather than a generic upsell. Two cards, both built to move from interest to a bigger basket without a guess.

Comparisons

In-chat comparisons

Side-by-side picks, honest tradeoffs

What makes it real

The comparison card only fires when a shopper explicitly asks to compare, and the pick is a recommendation, not a lock: the shopper is free to choose either product. A deterministic backstop rebuilds the structured card if the model tries to bury the comparison in a paragraph, so the side-by-side stays reliable across very different catalogs, from cat food to model-numbered electronics. The pros and cons come from your real products, so there's nothing invented to walk back later.

Promotions

Campaigns

Date-boxed promotions she leans into

What makes it real

Campaigns affect on-site widget chats only, not email or SMS. The numbers are honest: attributed revenue is net of refunds and handles multiple currencies, and conversion is measured per campaign rather than guessed. Status is computed from the dates you set, so a campaign goes live, ends, and stops influencing Sarah without you touching it.

Automations

Outbound & webhooks

Real shopper signals, your tools

What makes it real

Every webhook delivery is signed with HMAC-SHA256, so your endpoint can verify the call genuinely came from Cervito before acting on it. Destination URLs pass SSRF-safe validation, so a misconfigured or hostile URL can't be used to reach internal addresses. To be clear: Cervito does not send email or SMS itself. It wires Sarah's signals into the platform you already use.

She improves

Training and coaching

She learns your voice, every shift

What makes it real

Training examples only guide Sarah after you approve them, so nothing reaches a shopper unreviewed. Coaching notes are tenant-isolated, scoped to your shop and never shared across merchants, and capped per chat to keep her fast. Each accepted rule carries a "used in N chats" count so you can see what is actually working. Moderation is fail-open by design, which keeps your authority final: you decide what Sarah learns.

Coaching noteSaved

“Shoppers who mention pour-over convert on the starter set.”

What makes it real

Ten distinct touch types are tracked across the journey, and attribution is computed at order time against the recorded path. The holdout is the honest part: because a real slice of traffic is held back as a control, the revenue figure reflects causation, not correlation. A note on scope: the holdout is opt-in and set up by you, and the email and SMS touch types are framework-only today, not yet wired into the live measurement.

What makes it real

Handoff is soft by design: there is no guaranteed agent response time, and an unclaimed conversation auto-releases back to Sarah after 24 hours rather than stalling. Voice coverage depends on the shopper's browser. The follow-up email after a handoff is delivered through your own email or SMS tools, the same way outbound works. Every claim here maps to behavior already running in the widget: the 10-second pickup timer, the generated handoff brief, the 24-hour release, and the 44 detected languages.

This one needs a human. I’m bringing in the team.

Handed to your team · full context
The real thing

Three moments, exactly as visitors see them.

The production widget on Northbrook Coffee Co., our demo store. The same Sarah your visitors would meet. Recommend, compare, bundle.

01

Recommend

Gift & occasion

Northbrook Coffee Co.

with SarahAI

Something for a coffee-nerd friend, under €40?

09:41Read

The Yirgacheffe (€19) is a crowd-pleaser for enthusiasts. Add the ceramic pour-over mug (€16) and it lands at €35.

2m ago

YirgacheffeYirgacheffeIn stock€19.00

Bright and floral, and it is the bag coffee people gift each other.

Plate 03Northbrook Coffee Co., demo storefigures illustrative

A bed for a senior dogAsk her yourself at Pawfield Pet Co.(opens in a new tab)
02

Compare

Side-by-side

Northbrook Coffee Co.

with SarahAI

Yirgacheffe or Sumatra if I drink it black?

10:12Read

Compare options

AI Pick

Yirgacheffe

€19.00

In stock
  • Bright, floral cup
  • Delicate under milk

Sumatra

€18.00

In stock
  • Deep, earthy body
  • Low acidity, heavier

AI Pick: you drink it black, and the Yirgacheffe is made for that.

Plate 04Northbrook Coffee Co., demo storefigures illustrative

Compare two noise-cancelling headphonesAsk her yourself at Voltline Electronics(opens in a new tab)
03

Bundle

Complete the set

Northbrook Coffee Co.

with SarahAI

New to pour-over? Here’s the starter set: a V60, a gooseneck kettle, and filters.

just now

Goes well together

3 items selectedTotal: €69

Plate 05Northbrook Coffee Co., demo storefigures illustrative

Bundle a pour-over starter setAsk her yourself at Northbrook Coffee Co.(opens in a new tab)
Check it yourself

See her work a real catalog.

Real storefronts, very different catalogs, the same associate. Open one and ask her anything. This is the live product: no login and no sales call first.

Live catalog prices, from each store’s own catalog. Open a card and check them.

Prove it

See exactly what she earned, down to the cart.

Five attribution models side by side, a cross-session bridge that ties a phone visit to a desktop purchase, and the number that matters: conversion lift against a hidden control that never sees the widget. Causal, not correlation.

Conversion lift
+23%

vs. a hidden control cohort: causal, not last-click credit.

Plate 06Northbrook Coffee Co., demo storefigures illustrative

Proof of impact

Attribution

northbrook-coffee.com · last 30 days

Revenue influenced
€12,840
+18.2%
Carts saved
214
+24%
AOV with Sarah
€94.50
+11.6%
Significance
95%
Models
5

All attribution models, side by side

  • Last-widgetActiveFinal widget interaction before purchase gets full credit€12,840
  • LinearAll widget touches share equal credit across the journey
  • Time-decayRecent touches weighted higher, rewards near-purchase chats
  • First-touchFirst widget interaction gets all credit, rewards discovery
  • Position-basedFirst and last each get 40%, middle touches share 20%

Four of the five split every order across the whole journey, so their totals match. They differ in which conversation gets the credit. Only the active model reports revenue.

Conversion lift: A/B measurementSignificant

+23%

Visitors who saw the widget converted more often than visitors in the control group, over the last 30 days.

Treatment (saw widget)
Every shopper Sarah greets
Control (no widget)
Held out, never sees the widget
Measurement window
Last 30 days
Method
Two-proportion z-test, two-tailed
How this is computed

Plate 06Northbrook Coffee Co., demo storefigures illustrative

On every shift

She runs the whole conversation, not a single reply.

She greets browsers before they bounce, compares two options with a clear pick, handles the “is €19 too much?” objection, and pulls a bundle at checkout, by text or voice, in 44 languages.

  • Proactive: she opens the conversation on dwell, cart-stall, exit-intent and four more signals.
  • Recommends, compares, bundles, and picks the right size or grind from plain language.
  • Grounded in your live catalog: never invents a SKU, a price, or a policy she can’t verify.

Northbrook Coffee Co.

with SarahAI

Something fruity for my morning V60: Yirgacheffe or Sumatra?

09:24Read

Compare options

AI Pick
Yirgacheffe

Yirgacheffe

€19.00

In stock
  • Bright, floral cup
  • Made for pour-over
  • Too delicate for milk
Sumatra

Sumatra

€18.00

In stock
  • Bold, earthy body
  • Holds up in milk
  • Heavier than fruity

AI Pick: for a fruity morning pour-over the Yirgacheffe wins. Go Sumatra if you ever switch to milk drinks.

just now

Plate 07Northbrook Coffee Co., demo storefigures illustrative

The ledger

Cervito vs. a generic widget vs. hiring a team.

A generic widget is cheap and answers FAQs. A retail team is expensive and slow to scale. Cervito matches the human column on the things that move carts, and does not match it everywhere.

Comparison of Cervito, a generic chat widget, and hiring a retail team across 8 storefront capabilities.
CapabilityCervitoGeneric widgetanswers FAQsHire a teamknows the floor
On your storefront 24/7, every timezoneYesYesNo
Knows your real catalog coldYesNoYes
Speaks 44 languages on day oneYespartialpartial
Refuses to invent a SKU or fake priceYesNoYes
Greets, recommends, handles objectionsYesNoYes
Remembers what worked last weekYesNopartial
Hand-off with full context to your teamYesNoYes
Multi-touch attribution back to revenueYesNoNo

Cervito: yes in 8 of 8 rows · the human team matches on 4

Book a demo
The hire process

Three steps, and one line on your storefront.

One-click OAuth from your store after the demo, or drop the script tag on any storefront. She reads your catalog, you write three coaching notes, and she starts her first shift. No theme edits, no developer required.

01

Install in under two minutes.

After your demo, it’s one-click OAuth from your store. She reads your products, collections, and any policies you add. No theme edits, no developer required.

steps = 3 · theme_edits = 0 · script_tags = 1

02

Name her. Coach her.

Pick a name. Write three coaching notes about what you sell and how you sell it. Her first shift is already on-brand.

coaching_notes ≥ 3

03

Watch her sell.

Live conversations, multi-touch attribution, daily debriefs in your inbox. You’ll know exactly which carts she saved, and what to coach next shift.

shifts_per_week = 7

theme.liquid
<script async  src="https://app.cervito.io/widget.min.js"  data-shop-key="YOUR_SHOP_KEY"></script>
  • One lightweight script, loaded async
  • Shadow-DOM isolated
  • No theme edits
  • Zero third-party fonts or scripts

110.76 KB gzipped, against a budget of 116 KB we enforce at build time

Measured 2026-08-10. Run gzip at level 6 over the shipped file, the same method our build gate uses. Check the file we measured

The questions you’d ask

What we actually get asked on calls.

Not the marketing FAQ. The real ones: about training, handoff, mistakes, data, and what happens when something goes sideways.

Still weighing something?

Write to [email protected] · the founder reads every note.

A chatbot answers FAQs. Cervito is a named associate: she has a name and a personality you train, she remembers what worked last shift, and she sells. She greets, recommends, handles objections, and escalates to your team only when she should. You’re hiring her, not configuring an FAQ tree.

She’s built to refuse rather than invent. If a SKU isn’t in your catalog, she says so. If a policy question is fuzzy, she offers a handoff with the full conversation context. Your human picks up where she left off, never from scratch.

After every shift, thumb up the conversations that closed and thumb down the ones that didn’t. She proposes a coaching note in plain English (“don’t push the bundle to gift-shoppers”). You edit, accept, or reject. Accepted notes inject into every future conversation. No prompt engineering required.

She loads asynchronously, after your page has already painted, so she never sits in the critical rendering path. Every style lives inside a Shadow DOM boundary, isolated from your theme in both directions, and she pulls in zero third-party fonts or scripts. The shipped bundle is 110.76 KB gzipped, against a budget of 116 KB we enforce at build time. That is the measured size of the file your shoppers download, not an estimate. We publish the number, the date we took it, and the way to reproduce it, so you never have to take our word for it.

110.76 KB gzipped, against a budget of 116 KB we enforce at build time

Measured 2026-08-10. Run gzip at level 6 over the shipped file, the same method our build gate uses. Check the file we measured

The quickest way is to book a 30-minute demo with the founder (a Google Meet call) from any “Book a demo” button. Cervito is in private rollout: we onboard you personally and you’re usually live within 24 hours. We’re intentionally not listed on the public App Store; we want to onboard merchants we can support well. Prefer email? Write to [email protected] with your store domain.

We quote per store rather than publishing a price list, because the right number depends on catalog size and conversation volume, and we would rather tell you the real one on the call than anchor you to a tier you do not fit. Two things we will put in writing before you ask: no per-seat pricing, and new capabilities ship to everyone rather than being gated behind a higher plan. Invoicing is by bank transfer.

Conversations live in your tenant only: no cross-tenant sharing, no shared model training on your data. Visitor PII is redacted before it reaches the model; merchant API credentials are encrypted at rest with AES-256-GCM. All three mandatory Shopify GDPR webhooks (data request, customer redact, shop redact) are honored within Shopify’s stated timelines. Sub-processor list at /legal/dpa.

Automatically: when a visitor explicitly asks, when her confidence on a non-product question drops below threshold, or when a topic matches your configured handoff rules (“refund,” “complaint,” whatever you flag). Your team picks up with the cart, the catalog she pulled, and the full transcript.

Anonymous sessions are pruned after 90 days. Identified sessions (where a visitor checked out or was recognized) are kept for 365 days. Aggregate analytics (intent counts, conversion lift, attribution) are retained indefinitely in anonymized form.

On the floor

I greet every shopper the second they land, know your catalog cold, and bring in your team the moment it matters, every shift.

SarahSenior associate · Northbrook Coffee Co.

Sarah is software, and the widget says so: a bordered AI badge sits beside her name in the panel header for the whole conversation, announced in the shopper’s own language to screen readers, and if anyone asks her directly she answers plainly that she is an AI assistant for the store.

Your next hire

Ready to put a real associate on your storefront?

Book a 30-minute demo with the founder over Google Meet and watch Sarah work your own catalog. We onboard you personally, usually live within 24 hours. Not ready for a call? Send us a note instead.

Private rollout · one-click uninstall if it doesn’t earn its keep

Or just tell us about your store

A few details is enough. The founder reads every note personally and typically replies within one business day.

0/2000

No spam. Replies from a person.