Conversational Analytics lets you type a hiring question in plain English — like you'd ask a colleague — and get a chart and a short summary back, instead of building a report by hand. It's currently in Beta. This article covers how it works, what it's reliable for today, what it isn't, and specific known accuracy issues — so you know what to trust and what to double-check before you act on or share a number it gives you.

How it works

• Type a question into the chat box.

• Conversational Analytics returns a chart, a short written summary, and usually a couple of clickable follow-up questions.

Questions it handles well

Aggregate, scoped questions — counts and breakdowns, not lookups about one person — get the most reliable results:

• "How many applicants are sitting in Application Review across every open role today?"

• "What's the breakdown of scheduled interviews by interview type?"

• "How many recruiter screens were scheduled today across all roles?"

• "How many applicants do we have in every stage except Application Review, Hold Queue, and Reference Check across all open roles?"

The pattern: a specific metric (a count or a breakdown), scoped to something concrete — a time window, a stage, a set of roles.

What to avoid asking

• Questions about one specific candidate or application ("why is X's application paused," "what's X's current status") consistently return "I don't know this answer, I'm still learning." Open the candidate directly instead — Conversational Analytics is built for aggregate analytics, not individual-record lookups.

• Requests for a list of names don't return names, even if you explicitly ask for them — "give me the names of everyone hired this quarter" comes back with a count, not a list.

• Referral-source questions are currently the least reliable category and often come back with no answer at all.

• Month-only date questions without a year ("candidates in August") can come back with no data, while the same question with the year specified ("August 2025") returns results — it isn't resilient to a month name on its own.

• Follow-up questions meant to narrow a previous answer aren't always applied correctly. In at least one documented case, asking to "break down by source" after building a specific pipeline report produced a general breakdown instead of narrowing that same report — treat each answer independently rather than assuming full conversation memory across turns.

Good to know

• Conversational Analytics is Beta, built on an earlier version of the underlying engine. Kula's team has said they're actively working on an upgraded version, though no release date has been confirmed.

• It doesn't work like a general-purpose AI prompting tool. It's designed to help you derive insights and generate reports through a chat interface — it won't take actions on your behalf, and it behaves differently from tools like ChatGPT even once the upgraded version ships.

• It respects the same data visibility as your normal reports — anyone with Conversational Analytics access can see the same data through it that they could see in Reports & Analytics, no more and no less. It isn't a separate, more restricted (or more permissive) view.

• Chart values can occasionally render as obviously-wrong, extremely large numbers — a display issue distinct from whether the underlying computed number is correct.

• Complex, multi-condition, or subjective questions aren't supported — things like "how many candidates went through 2 interviews and then got hired," bias-pattern analysis across interviewers, or open-ended "any red flags in this feedback" questions are outside what it's built to handle today.

• Some data categories aren't covered at all, including Flow/sequence data, EEOC data, approval-related data, and change-log-style questions (like how many times an offer was revised).

• It may not be turned on for your account yet. Rollout is account by account, not automatic — if you have the right permissions and still don't see it, check with Kula support before assuming something is broken.

FAQ

• This isn't giving me an accurate answer — what's going on? You're not imagining it. Conversational Analytics is in Beta, and a recent internal review found real, specific accuracy problems — mismatched aggregate counts, wrong rate calculations, and fabricated dates among them — all still being worked on. For anything you need to rely on or share, cross-check the number against the matching standard report.

• How do I best make use of Conversational Analytics? Use it for quick aggregate questions — counts and breakdowns scoped to a time window, stage, or set of roles — rather than as a general-purpose command tool or a lookup for one specific candidate. Treat its numbers as a starting point to verify, not a final answer.

• Can Recruiters or Hiring Managers use Conversational Analytics? Not by default — only Super Admin and Admin have any Analytics access out of the box. A Super Admin has to create a custom role and grant Analytics permissions, including Conversational Analytics specifically, before anyone else can use it.

• Can I ask it for a list of candidate names? No — even if you ask directly for names, it returns a count instead of a list. Use standard Reports or an export for that.

• I asked a follow-up question to narrow down a previous answer, and it gave me something unrelated — why? Conversational Analytics doesn't always correctly apply a follow-up as a refinement of the previous answer; it can run the new question more generally instead. Ask precisely scoped, standalone questions rather than relying on it to remember exactly what you meant by a prior one.

• Is "Ask Kula" the same as Conversational Analytics? Unclear — see Needs Verification above. Don't assume the two behave identically or share the same accuracy profile.

Need help?

For a step-by-step introduction to the Reports & Analytics page as a whole, see "Your First 5 Minutes in Analytics." For permission questions, see "Analytics roles and permissions." For anything else, reach out to us at support@kula.ai or use the in-app chat.