If connectors are what Raya can see, skills are how she knows what to do with it. A skill is a proper working method for one specific job — how to audit an account, how to diagnose a drop in performance, how to write a brief a creator could actually shoot from. Most of them we write and maintain centrally, and Raya picks the right one for what you've asked — you don't install anything or keep anything up to date. You can also build your own, which is further down this page.
It's the least visible of Raya's three pillars and, honestly, the one that does the most work. It's the difference between an AI that gives you a confident-sounding paragraph and one that comes at your account the way an experienced strategist would — in a particular order, with particular things it refuses to skip.
You don't pick a skill — you just ask
There's no menu to choose from. You type what you want in Ask me anything on the Raya tab (or @ her in Slack) and she routes it herself. What matters is that you say what you actually want, because the phrasing is what she matches on.
The routing is deliberately invisible. Raya won't list her built-in skills for you or tell you which one she picked, and you'll never need to name one — the only skill names you'll see are the ones on skills you've created yourself. So there's nothing to learn and nothing to memorise; ask for the job, not the tool.
These are all real prompts that land on a skill:
"Performance has dropped — what's actually causing it?"
"Which campaigns should I scale, and which should I pull back?"
"Audit my active ads and do a creative gap analysis."
"What are my customers saying about us?"
"Which of my statics should I re-test as video?"
"Write me a UGC video brief."
"What's been trending on TikTok that I could borrow?"
Broadly, they cover four kinds of job: diagnosing performance (why something moved, where the budget should go, whether your audience structure is working), reading your creative (auditing what's live, breaking down your top spenders, finding the gaps in what you've tested), research (competitors, the ad library, and mining real customer reviews for language), and production (concepts, briefs and copy off the back of all that).
What a skill actually changes about the answer
This is the part worth understanding, because it explains behaviour that can look odd the first time.
She'll ask before she spends. If your request could reasonably mean two different things, she stops and asks rather than guessing. Ask for "20 UGC examples" and she'll check whether you mean UGC-style ads or organic posts — they're completely different sets, and searching the wrong one wastes your time and your credits. Say "20 UGC ads" and she won't ask at all. On a scheduled run, where there's nobody to ask, she takes the sensible default and tells you in the output which one she took.
She shows her working. Where a skill is talking about specific ads, it's meant to put the actual ads in front of you rather than just naming them — evidence, not decoration.
She tells you what she didn't do. If part of your request didn't happen, she says so in the same reply, before the deliverable. No "ready to drop into your board" for something that never got saved.
She'll sometimes decline to call it. If there genuinely isn't enough conversion data behind an ad to support a verdict, a skill will say so instead of inventing confidence. We know that's frustrating when you wanted an answer — but it's the right instinct, and the fix is usually a wider window or a bit more spend behind the test rather than a differently worded question.
Getting a better answer out of one
Four things make a noticeable difference, and they're all on your side of the conversation:
Say which account or brand you mean if you work across several — otherwise she has to stop and ask.
Say your window. "Last 30 days", "since the 1st", "this quarter". If you don't, you may not get the window you had in your head.
Say ads or organic when you're asking about content formats.
Fill in your Brand World. Skills that write anything — briefs, concepts, copy — are only as good as what Atria knows about your brand. See Brand World setup.
And the more connectors you've plugged in, the more of your picture a skill has to work from. A diagnosis that can see your Shopify data is a better diagnosis.
Custom skills
Teams keep asking for their own — a house audit format, an agency's reporting method, the specific checklist you run every Monday. You can build those yourself. There's a Skills page in the Raya section of the app where your custom skills live, and you can create one there.
The easiest route, though, is to ask Raya to build it with you. Describe the job you want done your way — what she should look at, in what order, what she must never skip — and she'll draft the skill and refine it with you. Once it's saved, it works like any other: you don't invoke it by name, you just ask for the job and she routes to it.
If you'd rather talk it through first, tell us about the repeatable job and we'll help you shape it — and that feedback is genuinely what shapes which skills we build centrally, too.
How this sits alongside the other two pillars
Connectors are her access — what data and tools she can reach. See Raya connectors.
Skills are her method — how she does a job properly once she's got the data.
Automations are her schedule — the same work, running weekly without you asking. See Automate your reports with Raya.
You don't have to think about any of this to use her. But if you've ever wondered why one question gets you a proper piece of analysis and another gets you a shrug, this is usually the reason — and a slightly more specific question is usually the fix.
If you'd like a hand working out which questions get the best out of her for your setup, just shout — happy to go through a few with you.