look at your support ticket inbox right now. sort by tag. i'll bet "how to" or "where is" or "can't find" fills the first page. these aren't bugs. they're not feature requests. they're navigation problems — users who got lost in your UI and had no one to ask except your support team.
the reason they called support isn't that your product is broken. it's that the gap between "stuck" and "figured it out" had no bridge. ai screen guidance is that bridge.
the math
let's be specific. the average saas company sees 40–50% of its support volume in how-to questions. these are questions with a definitive right answer — an answer that doesn't change, that any agent can give, that an AI can give faster.
if AI resolves 80% of how-to questions before a ticket is filed, and how-to questions are 40% of your volume, you've cut total ticket volume by 32%. at an industry average of $15 per ticket, a 500-ticket/month team saves $2,400 per month. in the first month.
$2,400/month is $28,800/year. enough to hire a part-time contractor. enough to justify a dedicated UX researcher for a quarter. and that's a conservative estimate — teams that deploy aside consistently see deflection north of 40% by month two, as the AI learns the most common stuck points in their specific product.
why chatbots don't cut it
every support platform has an AI chatbot now. they all promise ticket deflection. most teams i talk to report meaningful deflection for about three weeks, then the rate plateaus — and users start figuring out how to phrase questions in ways that route them to a human.
the fundamental problem: chatbots answer from docs. they're trained on your help center, your release notes, your FAQs. when a user asks "where is the export button?" the chatbot pulls a paragraph from a help article and pastes it in.
users don't read docs. that's why they filed a ticket. giving them a chatbot that reads docs back at them is not a solution — it's the same problem in a different font.
screen guidance is different because it operates in the user's actual context. the AI isn't guessing what screen they're on — it can see it. when a user asks "where is the export button?" aside looks at their screen and says "i can see the data table you have open. the export button is in the toolbar at the top right, it looks like a downward arrow." that's a different class of answer.
the deflection loop
this is the full loop for a deflected ticket, from first click to problem solved:
- 01
user gets stuck. clicks the aside widget in the corner of your product.
- 02
browser asks which tab to share. user clicks confirm. no extension, no download.
- 03
AI sees their screen. within three seconds: "i can see you're in the billing section. the payment method settings are one level down — click account settings in the left nav."
- 04
user follows the instruction. the AI watches and confirms when they land on the right page.
- 05
user solves their problem. session ends. no ticket filed.
total time: 30–90 seconds. total agent time: zero. no ticket created, no queue position, no context-switching for your team. the user just... got the answer and moved on.
measuring deflection
aside tracks deflection automatically. the formula is simple:
a "handoff" is any session where the user explicitly asks to speak to a human, or where aside determines it can't resolve the issue. everything else counts as deflected. the aside analytics dashboard shows this breakdown by session, by day, and by the specific stuck points that generated the most handoffs — so you can prioritize fixes in the product itself.
after four weeks, you'll have a heat map of your product's worst friction points, ranked by the number of times users needed help with them. that data is valuable beyond support — it tells product what to fix next.
getting started
the integration is one script tag. drop it in your app's html, set your workspace ID, and the widget appears. no SDK, no npm install, no webhooks, no backend changes. the whole thing lives in the browser.
the 30-day pilot gives you five live guided sessions, white-glove setup, and direct access to the founder. it's designed to get you to a measurable deflection rate within a week — not to sell you a contract, but to prove the thing works before you commit.
most teams see their first deflected ticket in under an hour. the first week of data is usually enough to build the business case for a full rollout.