Milly Software
InsightsUpdatesMilly Chat in April 2026: CRM Routing, Quiz Launch, and a New Retention Model
Update··8 min read

Milly Chat in April 2026: CRM Routing, Quiz Launch, and a New Retention Model

A round-up of what shipped in Milly Chat during April 2026 — native CRM routing across four platforms, intent-classified lead capture, the full quizzes launch, a transparent conversation retention model, and a five-page analytics rebuild.

V
Viet Le
co-founder · Milly Software

April 2026 was Milly Chat's biggest shipping month to date. Native CRM routing across four platforms, the full quiz launch, a new conversation retention policy, a rebuilt five-page analytics dashboard, and a long tail of smaller improvements — most of them shipped because a merchant asked first.

For full details on each feature, see its dedicated post on /writing.

Native CRM routing — four platforms

The marquee April release. Captured leads now write natively to four CRM platforms — no custom code, no Zapier in the middle — with per-platform success or failure visible to merchants on the lead detail page.

  • Klaviyo — profile import plus automatic list subscription. Drives abandoned-cart, browse-abandonment, and back-in-stock flows on chat-captured leads from day one.
  • HubSpot — Contacts and Tickets — contact upsert (already there) alongside a new Tickets path. Support questions now create a properly-pipelined ticket with a short summary, not just a bare contact. Custom properties get provisioned on first use.
  • Zendesk — ticket creation via standard token authentication. The body summary lands in the ticket description, ready for the support team to triage.
  • Maestra — already in place, now joined by back-in-stock notifications through Maestra (more on that below).

On the Core tier, the AI routes each lead to a different destination based on what the customer was asking about — newsletter sign-ups go to Klaviyo, support questions become Zendesk tickets, sales inquiries land in HubSpot, back-in-stock alerts go to Maestra. All from one widget. And if a specific page or campaign needs different routing, you can override platform-by-platform per conditional rule.

Lead capture, fully overhauled

The CRM integrations are the visible half; the routing logic underneath is the half that took the most thought. Five things changed about how a lead moves through the pipeline:

  • Intent tagging. The AI tags every captured lead by what the customer was asking about — promotional / business / support / general. Routing rules and coupon eligibility flow from the tag.
  • AI body summary. A one- or two-sentence summary attached to each lead, visible at a glance on the Leads page. Saves the support team from opening the full transcript to triage a ticket.
  • Smart routing (Core tier). Each intent can route to a different destination, with per-rule overrides on top of that.
  • Custom webhook. Fire any HTTPS POST alongside platform routing on every capture — URL, headers, body template, configurable per-store and per-rule.
  • Honest AI confirmations. The AI now grounds its "I created a ticket" / "I subscribed you to the list" claims in the real outcome of the integration call. If a Zendesk ticket fails to create for any reason, the AI no longer cheerfully claims success — it falls back to "I've passed your details to the team for follow-up."

And one quieter-but-important change: smart deduplication. The duplicate-email window now adapts to intent. Newsletter sign-ups still dedupe at 24 hours (preserves coupon-spam protection), but support tickets and business inquiries dedupe at 5 minutes — so a customer who emails a ticket request, gets a follow-up, and submits again isn't silently swallowed.

Quizzes — full launch

Product recommendation quizzes are now fully shipped end-to-end as a Core-tier feature.

  • Admin builder — create unlimited quizzes per store with inline editing of questions, options, and per-product scoring.
  • Conditional results — top recommendations and complementary cross-sells per result page.
  • All four widget formats support quizzes (chat bubble, search bar, slideout, smart banner). A persistent sparkles icon in the widget header lets shoppers re-trigger any quiz mid-conversation.
  • Quick Question routing — Quick Question pills can either send a chat message or trigger a quiz. One unified surface, two functions.
  • CSV bulk import — for merchants with quizzes already drafted in spreadsheets.
  • B2B-aware — quizzes auto-hide for B2B, wholesale, and collaboration intents.

Live across multiple merchants and being tuned against real shopper flows.

Out-of-stock → back-in-stock notifications

When a customer asks about an out-of-stock product, the AI can now offer to subscribe them to a back-in-stock notification flow. The Maestra adapter shipped in April; the Klaviyo adapter followed in early May (so it's already live for any merchant who wants it).

Back-in-stock is decoupled from general lead capture — separate configuration and UI, so merchants can run back-in-stock notifications without enabling lead capture (or vice versa). Useful for stores where the chat is meant to be a product-discovery tool, with email captured exclusively at the out-of-stock moment.

Conversation retention model

We replaced the previous "unlimited" retention with a transparent, scalable policy:

  • Starter — 30-day rolling replay window, save/flag exempt.
  • Essentials — 90-day rolling, save/flag exempt.
  • Core — 90-day rolling, save/flag exempt, plus a monthly export of older conversations for warehousing.
  • Enterprise — 1-year window, flagged conversations kept indefinitely, monthly export.

The save/flag exemption is the part worth calling out. Bookmark a conversation from the replay list and it stays forever, regardless of tier. Your analytics aren't affected by retention either — totals and trends are calculated nightly and stored separately, so purges only affect per-message replay. The model lets us scale replay storage predictably without losing the conversations you actually care about.

Analytics — restructured into five pages

The single Analytics page split into five focused ones: Overview / Impact / Speed / Queries / Zero Results. Daily rollups make 30-day ranges load in under two seconds, even on high-volume stores.

Two new things on the Impact page worth flagging:

  • Post-purchase support engagement. Returning customers who came back to chat after purchasing — warranty questions, returns, "where's my order." This captures the support-side widget value that was previously invisible alongside the pre-purchase "chat → buy" story. On one of our larger merchants in a recent 30-day window, the post-purchase surface accounted for ~444 returning customers and ~$33K in associated revenue. That was already happening; we just weren't surfacing it.
  • Configurable attribution window. Set 7 / 14 / 30 / 60 days per store. High-AOV merchants can bump from 7 to 30; impulse merchants stay at 7. All four windows precompute nightly so changes are instant — no rebuild lag when toggling.

And on the Queries page, two operational additions:

  • Gaps. Surfaces unanswered queries with a mark-as-addressed workflow. Instead of digging through transcripts to find what your knowledge base is missing, the gap surfaces itself and you decide whether to add an article, update a product description, or accept the gap.
  • Weekly Recap. A 7-day standup view — conversation trend, top queries, top gaps, leads, add-to-cart activity, assisted revenue. Designed to be the page you open once a week without having to compose your own dashboard.

Knowledge base & conditional rules

Two smaller-but-load-bearing surfaces also moved in April.

Knowledge base picked up custom categories — every store gets eight defaults (FAQ, Policies, Product Info, etc.) that you can rename, recolor, reorder, hide, or delete. Color-coded filters make browsing large knowledge bases fast instead of bottomless. And markdown upload for batch-importing .md files, with title detection and heading-aware chunking — so the AI retrieves well-scoped snippets instead of pulling in loosely-related content.

Conditional rules got reorganized into three sections — Branding (visibility, format, leading question), Response (AI instructions, quick questions), Lead Capture (per-rule webhook, body template, CRM platform, coupon toggle). Plus three new proactive trigger types on Core+:

  • Time on page — fire after a chosen number of seconds.
  • Scroll depth — fire when a visitor scrolls past a threshold.
  • Exit intent — fire when the cursor moves toward the browser chrome.

Smaller things, in one paragraph

The widget now ships in four formats (chat bubble, search bar, slideout panel, smart banner), all with a sticky add-to-cart bar on product pages and a variant-aware picker that renders all variants — out-of-stock ones stay visible but greyed out, instead of hidden. The AI now mentions which variants are in stock when some are sold out ("available in Seabreeze Matte and Frostbite, but Obsidian Black is currently out of stock"), and products with mixed availability stay visible. Toast notifications replaced the top-of-page banners across the dashboard. Out-of-stock toggle in the preview surfaces, so you can verify what shoppers will actually see. Manual store-name sync — explicit button on the Branding tab, replacing the previous background-sync that ran on every page view. Daily ingestion heartbeat — an internal monitor alerts us if your store stops receiving orders within 24 hours, so silent webhook failures get caught in a day instead of a week. Per-visitor rate limit on the analytics endpoint — protects against runaway behavior without affecting legitimate flash-sale traffic.

What's coming next

A few things queued for May and beyond:

  • Reviews integrations — Yotpo first, with one or two more on deck. The goal: surface review content inside the chat so the AI can ground product recommendations in real customer feedback alongside the catalog and your knowledge base.
  • Cross-sell recommendations — post-add-to-cart complementary product surfacing. Pairs naturally with the sticky add-to-cart and variant-aware work that landed in April.
  • Leads insights — a richer view of what your widget is capturing. Capture-rate trends, intent breakdown over time, top sources, and a summary of which corrections have been applied so you can see how the AI is being tuned.
We build features the way we'd want them ourselves if we ran a Shopify store. The feedback loop with merchants is the most reliable signal for what to ship next — and most of the April release shipped because someone asked for it first.

If something on your store would benefit from a feature we haven't built, or a CRM integration we haven't added yet — say so. The easiest way: drop a note via the contact page, or reply to whatever email thread you're already on with us. We read everything, and the request list is where the next month of roadmap mostly comes from.

Try Milly Chat

Want to see how this fits your store?
We'll set up a working session.

Get it on ShopifyTalk to sales →