Why We're Building Moss
Why We're Building Moss
A real-time AI layer that turns every SaaS UI into an interactive, self-teaching experience.
1. SaaS Has a Silent Churn Problem
Most of the users who leave a product never file a complaint. They sign up, hit a screen they do not understand, try once or twice, and stop logging in. The onboarding tour ran fine. The help center exists. Nothing in the product noticed the moment they got stuck.
Customers do say onboarding matters to them: in a 2020 Wyzowl survey of 216 consumers, 86% said they would be more likely to stay loyal to a business that invests in onboarding content that welcomes and educates them after purchase (Wyzowl). That is a consumer survey, so treat it as directional for B2B. What we saw in our own discovery calls was more specific: prospects who never did anything in their trial because they did not know where to start, and support queues dominated by variations of the same question.
"Where's that filter again?" → support ticket → wait → frustration → churn.
2. The Cost of Doing Nothing
Every team we talked to was already paying to close this gap, just in ways that do not scale:
- A support queue full of how-do-I questions. "How do I configure this, give access to that, connect this integration." Each one is answered by a person, and the answer is usually a screenshot of a screen the user is already looking at.
- Human-led onboarding as the only option. For complex products, a specialist walks each account through setup. That works for the largest customers and is not offered to the rest, which caps how many customers the product can take on.
- Static tours that break. Tour tools anchor to the UI as it was when the tour was built, so every release is a chance for guidance to point at nothing.
Despite AI chatbots, users still type-describe their screen while reps ask:
"Do you see a button that says …?"
3. GUI, Chat & Voice Need to Converge
We believe the next era of human-computer interaction is hybrid: LLMs that see the interface, chat in context, and guide visually, blending the best of GUI precision with conversational ease. It is also a different route from MCP. Where MCP needs backend API integration for every tool, a web agent like Moss needs a line of code in your frontend and works with the UI you have already built.
4. Introducing Moss
Moss is a real-time interactive AI assistant that lives inside your web product. This is what it does today:
- It sees the screen. A lightweight SDK, embedded with a script tag, captures the page the user is on: visible state, interactive elements, and where they are in the product. Sensitive fields can be marked for redaction.
- It answers with sources. Answers come from your help center and documentation, with citations, so the user gets the why alongside the how.
- It guides on the live UI. Moss highlights the next element to click and stays with the user step by step as the screen changes. Walkthroughs can be authored ahead of time, saved from a good AI session, or generated live for the long tail no one wrote up.
- It can take the action. With permission, Moss fills the form or files the ticket with the conversation summary attached. Elements like a "Pay now" button can be marked so the assistant never clicks them on its own.
- It gets better with your approval. Moss reviews conversations, finds where users got stuck or where the docs fell short, and proposes concrete changes. A person approves each one; nothing changes silently.
5. What Changes When Moss Is On
We are careful here. We do not yet publish outcome numbers, because the only ones we have come from pilots that are not cleared for public use, and a pilot result on one product does not transfer to another. What we can say is what the mechanism changes, and what we expect to see as a result:
| What changes | Why it should matter |
|---|---|
| The how-do-I question gets answered in the product | The user gets the answer and the next click without leaving the screen or waiting on a reply, so fewer of those questions should reach your team as tickets. |
| Setup help is available at the moment of confusion | Help arrives when the user is stuck, not on the day of the scheduled call, so the gap between signup and the first completed real task should shrink. |
| Guidance follows the live UI | Walkthroughs are resolved against the screen as it is now, so a release should not silently break them the way selector-pinned tours break. |
| You see where users get stuck | Every conversation records what the user was trying to do and where it went wrong, which is input your product and docs teams do not get from a help center. |
How large each effect is depends on the product, the quality of your documentation, and how many of your workflows Moss has been shown. That is why every engagement starts with a measured pilot rather than a promise.
6. How We're Proving It
We're partnering with forward-thinking teams in logistics, fintech, and dev tools to benchmark, on their own product:
- Before/after volume of how-do-I tickets
- Time from signup to the first completed real task
- User satisfaction (CSAT on each session)
When a partner clears their numbers for publication, we will publish them here with the context needed to read them.
7. Help Us Shape the Future
We're opening a design-partner waitlist. Partners get:
- A pilot on your real product (a script tag in your frontend, no backend work)
- Bi-weekly stand-ups that steer our roadmap
- Preferred terms after the pilot
👉 Let's Talk Let's make "Where's that button?" a thing of the past.
Questions, thoughts, or spicy onboarding horror stories? Email me junil@viamoss.ai
See Moss guide a real workflow in your product.