Deep dive — the longer AI read on a competitor

You’re looking at the AI insight panel for a competitor pizzeria — the panel that tells you what their reviewers love, what they complain about, where they’re weak, what to differentiate from. It’s the daily-glance view: clean, scannable, four bullets per column. But sometimes that’s not enough. You want the bar chart of their posting cadence, the KPI grid with their follower count and engagement rate side-by-side, the table of their menu prices versus yours, the longer narrative synthesis. Deep dive is the button that gets you there.

What it does

Every AI insight panel on a competitor’s page (Instagram, Google reviews, Facebook) now has a Deep dive button next to See full analysis. Click it and a fullscreen drawer opens with a richer view of the same competitor — the same data the static panel reads, but spread across charts, KPI tiles, comparison tables, and narrative blocks that the AI picks based on what’s interesting about this venue. A pricing-heavy competitor gets a menu table; a cadence-heavy competitor gets a posting-frequency bar chart; an engagement-heavy competitor gets a follower-growth KPI grid. The AI picks the right shape for the data, you read it.

The first time you open the deep dive on a given competitor, the system generates the analysis — that takes ten to thirty seconds and uses credits from your meter. The second time you open it within six hours, you get the cached version instantly at zero cost. The third time, same thing. After six hours, opening the drawer again auto-generates a fresh version (it’s likely the competitor’s posted something new in that window).

You can also force a refresh at any time using the Refresh button at the top of the drawer — that always re-generates and re-meters credits. Use this when you know the competitor just shipped a big post or moved their prices and you want the latest read.

The rule

The static panel is the daily glance. The deep dive is the weekly read. Same data, two depths.

How to use it

Open any competitor’s page from Admin → Growth → Competitors, then pick a channel tab — Instagram, Google reviews, or Facebook. The static AI insight panel sits in the middle of the tab content. In its top-right, alongside the existing Refresh and See full analysis controls, sits the Deep dive button.

Click it. The fullscreen drawer takes over the screen. If there’s already a cached analysis fresh enough to use, you see it immediately — title, subtitle, then a stack of blocks (charts, tiles, tables, narrative) in the order the AI thought made sense for this competitor. The header tells you how long ago the cache was generated (“Last refreshed 2h ago”). Press Esc, or click the X in the corner, to close.

If there’s no cached analysis, the drawer shows a calm Generate button instead. Click it, watch the spinner for fifteen to thirty seconds, and the analysis lands. It’s cached automatically the moment it’s generated — the next time you open this drawer (or the same drawer for the same competitor on a different day) you’ll see the saved version straight away.

The Refresh button at the top of the drawer is always there. It re-generates from scratch — credits get re-metered, the previous version is replaced. Use this when you want the latest read on a competitor who’s been busy, or after a big update on their end you know about and want the AI to see.

When you want to compare prices side-by-side

If the competitor has a menu we’ve extracted (via the Menu & pricing tab), the deep dive will usually include a table block listing their priced items — name, price, category, and the previous price if it changed recently. That’s the AI’s pick when pricing data is rich; on a competitor without a menu, the AI substitutes a different block (engagement KPIs, posting cadence chart) instead.

When you want the bar chart of their cadence

The AI usually surfaces a bar chart showing the rhythm of their posting — by day of week, by post type, or by month. The exact axis depends on what’s interesting about the data; the AI picks.

When you want the longer narrative read

Every deep dive ends with one or two narrative blocks — the synthesis the AI wrote about what this competitor does well, where they’re weak, what’s worth copying, what to differentiate from. It’s longer than the panel’s two-bullet Our move card; treat it as the strategic read you skim on a Sunday afternoon, not the operational hint you act on at noon.

Side-by-side with your venue

Every KPI tile in the deep dive now carries a small second line showing your own venue’s equivalent number underneath the competitor’s. The competitor’s follower count sits big on top; below it sits You: 4,300 and a coloured delta — −63% in rose if they’re ahead, +170% in emerald if you are, an amber not tracked when we don’t have the matching number on your side yet. The colours read from your point of view: green is winning, red is losing. The format adapts to what you’re looking at — percentages for counts (followers, likes, comments), percentage points for rates (response rate), bare currency for prices.

The comparison line is there so you don’t have to flip back to your own analytics page and do the math. If your osteria has 4,300 Instagram followers and the competitor pizzeria has 11,627, the deep dive tells you “they’re roughly three times your size” without you opening a calculator. If your owner-response rate on Google is 100% and theirs is 78%, you read +22pp in green and move on — that’s a fight you’ve already won, not one to spend the next week on.

The KPI grid answers “where are they?”. The comparison line answers “where are we, against them?” — both questions at once, on the same tile.

The AI decides per-tile whether a comparison is meaningful. On metrics where your side has no equivalent yet (a brand-new venue, an untracked channel), the comparison line either disappears or shows an amber “not tracked” hint instead of inventing a number.

Reading the analysis in your language

Every AI insight panel — and every deep dive — now has a small Translate ▾ button in the header next to Refresh. Click it and a 12-language picker drops down with the locale code on the left (EN, IT, TH, ES, FR, DE, PT, JA, ZH, AR, VI, MY) and the native name on the right (English, Italiano, ภาษาไทย, Español, …). Pick one and the entire analysis swaps to that language — title, recommendations, themes loved, themes needing work, the long narrative paragraphs.

The button then turns purple and reads Show original (IT) with the language code you picked. Click it again and the analysis flips back to the original instantly — no network call, no waiting, no cost. The translation is held in a shared cache, so a phrase translated once anywhere in the app (your own panel, a teammate’s panel, an analysis you ran last week) comes back for free the next time anyone asks for it in the same language. Only fresh strings — phrases the system has never translated before — bill characters against your usage meter, and the cost is fractions of a credit per panel.

The button is meant for sharing analyses with non-native staff or reading a Thai-venue analysis in Italian during your morning coffee. The AI itself doesn’t change — same recommendations, same insights — only the language you read them in.

When the deep dive recommends “steal this dish — they offer Carbonara at ฿420, you don’t have it” you can trust that recommendation, because the AI has been handed your full POS menu before it writes a word. Item names, headline prices, category — all 200-or-so dishes go into the prompt as context. If you have Spaghetti alla Carbonara on your menu under “Classic Pasta”, the AI fuzzy-matches that against the competitor’s “Carbonara” and doesn’t claim a gap that isn’t there. Same for “Carpaccio di Manzo” matching “Beef Carpaccio”, or “Cotoletta alla Milanese” matching “Schnitzel”.

When a match exists, the AI is told to recommend a different angle — reprice, replate, better photography, sharper marketing copy — instead of pretending you need to add a dish you already serve. When no match exists, the “steal this dish” recommendation is a real gap worth piloting. Trust the recommendation either way; the AI has the data to be honest.

Worked example

Anna runs an Italian osteria in Bangkok. She’s been watching Casa Italia, a competitor two streets over, for a couple of months. Their static AI panel on Instagram tells her the usual things — they post about pasta, their reviewers love the carbonara, their engagement is up. She wants the deeper read before her Monday planning meeting.

Tuesday morning, she opens Casa Italia’s competitor page, clicks the Instagram tab, and presses Deep dive in the AI panel header. The drawer takes over the screen. Twenty seconds later, the analysis lands: a title (“Casa Italia — what they’re doing, what to take, what to leave”), a one-sentence subtitle, then five blocks.

The first block is a KPI grid with four tiles. Followers: 12,800, below that You: 4,300 · −66% in rose — Casa Italia is roughly three times her size. Posts per week: 3.2, below that You: 1.8 · −44%, also rose. Engagement rate: 8.4%, below that You: 9.1% · +0.7pp in emerald — Anna’s followers are smaller but more engaged. Average pasta price: ฿340, below that You: ฿320 · +฿20 in neutral grey — she’s priced just under, no clear winner. Anna takes ten seconds to read the comparison and already knows where the fight is: scale, not engagement quality.

The second block is a bar chart showing Casa Italia’s posting cadence by day of week — they ship Wednesday and Saturday consistently, Sundays almost never. The third is a table of their menu items — Carbonara ฿420, Cacio e pepe ฿380, Tagliatelle al ragù ฿440 (was ฿420, up two weeks ago). The fourth is a narrative on what they’re doing well — their behind-the-scenes prep reels consistently outperform their food shots. The fifth is a narrative on what to differentiate from — their wine pairings are weak, their bar program is underbuilt.

Importantly, the menu narrative does NOT tell Anna to “steal Casa Italia’s Carbonara” — the AI knows she already has Spaghetti alla Carbonara on her menu at ฿350. Instead it suggests she reposition the dish, photograph it next to a glass of Frascati, and lead the next Wednesday post with it: Casa Italia is at ฿420 for the same dish, Anna’s ฿70 cheaper, that’s a value angle.

Anna’s head chef is Thai and reads English slowly. She clicks Translate ▾ in the panel header, picks TH ภาษาไทย, and the whole analysis swaps to Thai in about four seconds — title, recommendations, the narrative blocks, the KPI tile sublabels. She forwards the screenshot to the chef on Line. Then she clicks Show original (TH) to flip back to English for her own reread, and closes the drawer. Total time: about five minutes.

When the analysis fails

Sometimes the AI provider is misconfigured, or the competitor doesn’t have enough data yet, or the credit meter is empty. Each of those surfaces as a calm amber card inside the drawer with a short explanation:

“Not enough data on this competitor yet” — refresh their page first so we have something to analyse, then try the deep dive again.
“Your AI provider isn’t set up yet” — open Admin → AI Settings and pick a model for deep-dive analysis.
“We couldn’t build the deep-dive view this time” — try again in a moment. If it keeps failing, the technical detail underneath has the specifics for support.

The static panel keeps working in all cases; the deep dive is the optional, on-demand layer.

Competitors — the page the deep dive lives on. Open a competitor, pick a channel tab, look for the Deep dive button in the AI panel header.
Competitor menus and pricing — the source of the pricing table the deep dive surfaces. Add a menu to a competitor first if you want pricing in the analysis.
Library — Phase 2 destination for pinned deep dives (not yet shipped; the artifact already persists, only the bookmark surface is pending).