Ask Spotlight: How to Interrogate Your Customer Analytics

I went digging through a month of Ask Spotlight usage to find out what people actually ask it and two things surprised me:

• 56% of every message sent is a follow-up. So the most common thing anyone does with Ask Spotlight is push on an answer it already gave them. Brilliant! That’s the whole idea.
• Most conversations stop at two messages (but the longest one I found ran to forty-eight!)

Now, two messages are often exactly right. You had a good question, you got your answer, you got on with your day feeling empowered and informed. Most of what people need from Ask Spotlight can genuinely be one or two well-aimed questions away.

But the long conversations are doing something different; somewhere around turn four or five you stop asking what happened and start asking why, and what to do about it, and can I have the receipts. You come out the far end with a root cause, a coaching plan, and real data you can forward to Product with the subject line “please fix this.”

So let’s do a masterclass; whether you’re using Ask Spotlight or another agentic AI tool to question your contact centre data, these tips are for you.

Lesson One: The power of your first question

Think about prompting like this: you’re briefing a colleague.

1. Say when: “Yesterday.” “Last week.” “1 October to 28 February.” All grand. What doesn’t work is saying nothing and hoping. Ask Spotlight will pick a sensible default, but a default is a polite guess about what you meant, and you can do better than a polite guess!

2. Steer towards the right data: Don’t be afraid to use some filters to help specify the data. Think things like region, channel, queue, market, B2B versus B2C, voice contacts. 

3. Say what you want it to look like: “Top 5, ranked.” “Three sentences.” “Split B2B and B2C, then split B2B again by third-party versus own retail.” Ten extra words, and you skip the reformatting.

Don’t agonise over it though. Your first question isn’t a spell and if you get it slightly wrong, you just ask again. Which brings us to the good bit.

Lesson Two: The follow-up loop

Most of us learned our search habits from Google and our data habits from dashboards. Both taught us a variation of the same thing: you ask, you get, you’re done. If the answer’s wrong you go back and phrase it better from scratch.

Bin that, because it’s costing you the good stuff.

Every follow-up does something different with the same investigation. And you never re-type the filters, the dates, or the context. It’s all still sitting there. Here’s a real one, from an iGaming operator wondering why ‘welcome-offer contacts’ were creeping up.

Turn one: how big is this?

Q: “Welcome-offer contacts have been climbing this month, what are players actually getting in touch about?“

A: 312 contacts, up 38% on last month. Three sub-reasons do most of the damage: bonus not credited after first deposit (44%), wagering-requirement confusion (31%), offer expired before it was used (18%).

Turn two: why is it happening?

Q: “Why the confusion? Is the offer itself unclear, or is something breaking?“

A: Clarity, not a bug. The “not credited” players had deposited but hadn’t met the 40x wagering requirement, and assumed the bonus had failed. They say over and over that the wagering requirement and the 7-day expiry aren’t visible at sign-up, they find out when they contact support. Actual crediting errors are rare.

Turn three: so what do we do?

Q: “What could we change to bring these contacts down?“

A: The biggest preventable chunk is wagering and expiry confusion, about 49% combined. Show the wagering requirement and expiry on the offer page and the deposit-confirmation email. Add a bonus-progress indicator to the player account. Send a nudge on day five of seven. Roughly 6 in 10 of these contacts look self-serviceable if the terms had been visible earlier.

Turn four: prove it

Q: “Give me three example conversations I can send to the product team.”

A: Three representative interactions, links straight through to the transcripts.

Four questions. About two minutes in total to ask. Out the other end: a diagnosed problem, a sized opportunity, a specific product recommendation, and the receipts.

Lesson Three: Role specific usage

Same tool, completely different question. An ops lead’s Monday morning looks nothing like a QA manager planning their review week, which looks nothing like a CX director prepping a board update.

Ops leads: the morning read

EdgeTier’s biggest group, and the most habitual. Volume, themes, and contact reasons make up roughly one question in three across the whole platform.

• The opener: “What were yesterday’s top contact drivers, and how does that compare to the same day last week?“

• What you’re really after: knowing what today looks like before someone tells you at eleven.

• The follow-up that earns its keep: “What drove the extra contact versus last week?” 

Volume says something happened, but deviation says something changed. Chase the deviation.

QA managers: from scorecards to coaching

Second biggest group, about 17% of questions. They have three major jobs: what are our evaluations already telling us, where do my remaining review hours go, and what do I actually say in the 1:1.

• Start with the evaluations you already have: This is the bit people don’t realise is on the table. Ask Spotlight can see your QA scores (auto evaluations and manual ones) not just the conversation analysis underneath them. So you can point it at the scoring itself:

“Across last month’s evaluations, where are we losing the most points, and which sections of the scorecard are weakest?“

Then push on it: “Which agents are struggling most on that section, and what’s the pattern in how they’re handling those calls?” You’ve gone from a scorecard average to a specific, coachable behaviour without opening a single evaluation form.

• Smaller lookups live here too: “give me [agent]’s QA average for the last three months“, or “has this interaction been scored or reviewed by anyone?“

Then work out where to look next. The evaluations only cover what’s been sampled. For everything else: “Show me the agents with the highest customer frustration rate on calls last week.” “What’s driving it for the top three?“

That’s a prioritised review queue built from the whole population rather than the sample, and Ask Spotlight will link you straight to the interactions worth opening.

• Then build the coaching. “Give me themes I can use in a coaching session.” And it works in the flattering direction too, which people underuse: “give me a good and a bad example of an agent response to a billing query.” Your own best conversations, turned into training material, in about fifteen seconds.

One more worth stealing: repeat contact by agent. If the customer came back, the first answer didn’t land. That’s a coaching insight you will never find in a CSAT score, or on a scorecard.

Senior leaders: the answer, mid-meeting

Fewer questions but higher stakes. Senior leaders don’t live in dashboards, but you still need the picture, ideally in the language of your own targets.

• The opener: “What are the biggest drivers of customer dissatisfaction right now, and how long have they been trending?“

• What you’re really after: something evidence-backed enough to steer a decision, fast enough to use while the meeting is still going.

Three gears anyone can use

Underneath the job titles there are three ways of working that show up across all of them.

First gear: the fast lookup

About one question in ten isn’t analysis at all. It’s finding one thing, quickly.

• “Provide the calls for booking 30256635”
• “Search for contact ID 6cec4623-2f96…”
• “Can you look for this phone number in the last 12 months: +44 7960…”
• “Give me all interactions from customer email [address]”
• “Find any cases in the last 24 hours with this exact wording: ‘I received a damaged item…'”

Fun fact, a decent few I found are just a number pasted in with no sentence around it at all, which delighted me as it shows that people are in the habit of using Ask Spotlight and find it a reliable source!

Second gear: the analyst brief

Other end of the scale. Around 18% of prompts run past 500 characters, and they mostly come from people who could have written the SQL themselves.

These read like specs, not questions. Objective, date range, explicit exclusions, the specific angles to dig into, and the exact output structure required, right down to “give me an executive summary, main findings, recommendations, and thirty representative conversations.”

You don’t have to work this way, but if you run the same complex analysis every month, writing it out properly once beats asking it vaguely five times. Brief it like a sharp junior analyst. The tighter the scope, the less you’ll correct.

Third gear: the thing I got slightly wrong

Let me tell you about a question pattern I nearly labelled a mistake. It shows up constantly and looks like this:

Can you do the same search for 2 Mar – 4 Mar? Can you do the same search for 2 Mar – 15 Mar? Can you do the same search for 16 Mar – 31 Mar?

My read: these poor people don’t get it. They’re re-asking instead of digging. They’re using a conversation like a search box!

Then I looked properly, and no, they get it just fine. They’ve got an answer they like and they want the identical analysis with one thing swapped. The dates. The market. The agent. The queue.

“Why” is still the main topic here, users just want to steer the platform towards their way of working rather than having one catch-all report that’s less nuanced. Nothing wrong about that!

The most common missteps 

1. Asking one vague question as a test, then giving up: By far the most common way to be underwhelmed. The learning curve is much gentler than learning a full analytics interface, but it isn’t zero; expect a bit of trial and error before you’re landing exactly what you want first time.

2. No scope at all: “What question is most frequently asked by customers?” No dates, no brand, no market. Ask Spotlight is designed to still work well without parameters (we’re very proud of this in fact!) but if you have a specific focus area, state it in the first question.

3. First names: “Sarah’s messages” is a much harder problem than a full name or an ID. Give it something to grab.

4. Asking for things that aren’t in there: Forecasts and predictions, or taking action on a ticket; not today. Ask Spotlight analyses the data you already have. (More on that below, because it’s changing.)
5. Don’t forget about the Rules: Teach Ask Spotlight, or your agentic AI of choice, about your business, the KPIs that matter, the tone you want, and the things it should never do. The more specific the rules, the sharper the answers. These rules then sit at the top of every output.

What’s coming next for Ask Spotlight

Here’s a workaround I keep spotting: people maintaining a prompt library in a Google Doc somewhere, pasting the same thing every week with the filters swapped. That’s a hack, and hacks are the best product feedback there is.

So next: saving prompts. Repeatable report templates with your own guidelines and branding. Filtering the same report per market, team or region. Delivered on a schedule to the people who need it. Same analysis but none of the copy-paste.

After that, it gets more interesting. Ask Spotlight is EdgeTier’s first agentic feature, but it still stops where you start; it gets the information and builds the report, then hands it over. We’re looking at what happens when EdgeTier can act on its own analysis. Something closer to a colleague with a target to lower contact rates, who comes up with ideas from the data, does something about them, and tracks whether it worked.

And if you’re technical: the free EdgeTier MCP server already lets you pull EdgeTier data into your own LLM tools (Claude, Copilot, ChatGPT, whatever you’ve built) and mix it with your other sources. If you’re the sort of person writing 500-character prompts, go and have a look at that.

The short version

• Your first question is a starting point, not a transaction.

• Say when. Stack your filters. Say what you want it to look like. Then keep going, ask why, ask what to do about it, ask for the evidence.

• The people getting the most out of this aren’t asking cleverer opening questions than you. They’re just asking follow-ups.

---

Already using EdgeTier? Ask Spotlight is in your platform now. Go and ask it something specific!

Not yet? Have a look at what Ask Spotlight does, and book a demo if you want to see it against your own data. See Ask Spotlight →