How RestroAI works: from your order data to your WhatsApp at 9 AM

The big picture: nightly batch, morning delivery

RestroAI is not real-time. By design. It runs once a day, in the early morning hours when your restaurant is closed and Restrofi's servers are quiet. The compute is cheap, the data is settled, and the next day hasn't started — so RestroAI has the cleanest possible read on what just happened. By 9 AM IST, your WhatsApp is buzzing with a 3-5 sentence message about yesterday and today.

The whole pipeline takes about 30 seconds per restaurant. Hundreds of restaurants get processed in parallel.

Step 1 — Data collection (already happening)

Every order placed through Restrofi — whether QR dine-in, takeaway, delivery, or staff-entered — gets captured with full detail: items, quantities, prices, timestamps, table number, customer phone if collected, GST line items, modifications, and cancellations.

There's no extra setup for RestroAI. If you're using Restrofi normally, you're already producing the data RestroAI needs. The first 48-72 hours of order data is what RestroAI uses to build your baseline; from then on, every new order updates the baseline incrementally.

Step 2 — Pattern analysis (the hard part)

This is where RestroAI's analysis engine does the work. It processes a structured slice of your last-24-hours data plus your last-30-days baseline and looks for things like:

• Revenue or order-count anomalies (more than ~15% deviation from same-day-of-week baseline)
• Menu mix shifts (a normally-popular dish suddenly underperforming)
• Time-of-day patterns (your peak hour drifting)
• Customer behavior changes (repeat rate climbing or falling)
• Margin movements (cost or price changes affecting per-dish profitability)
• Dead inventory (items with zero sales for N days)

It also cross-references calendar context: weekends, public holidays, regional festivals, weather events. A 22% sales drop on a Saturday is suspicious; the same drop on Diwali night is expected.

Step 3 — Insight selection (what makes RestroAI useful, not noisy)

Out of roughly 30 signals analyzed every night, RestroAI picks the 3-5 most actionable ones for that specific morning. The filtering rule is brutally simple: would the owner actually do something differently if they knew this?

• "Your sales went up by 0.8% vs yesterday" → not actionable, dropped
• "Mushroom soup hasn't sold in 8 days" → actionable, surfaced
• "Tomorrow is a public holiday and last 3 holidays paneer ran out by 9 PM" → actionable, surfaced

This is the design choice that keeps RestroAI from becoming notification spam. Most AI products err the other way: send everything you can compute. RestroAI sends only what would change a decision.

Step 4 — Writing in your language

Once the insights are picked, RestroAI generates the message in your chosen language: Hindi (Devanagari), Hinglish (Roman script), or English. The Hindi and Hinglish versions are not translated from English — they are written natively, which is the only way to get the tone right.

The same insight in three languages reads like:

English: "Yesterday: 47 orders, ₹14,200. Mushroom biriyani sold 0 — 8-day streak. Suggest replacing with a weekend special."

Hinglish: "Kal: 47 orders, ₹14,200. Mushroom biriyani 0 sold — 8 din ka streak. Weekend special se replace karo."

Hindi (Devanagari): "कल: 47 ऑर्डर, ₹14,200। मशरूम बिरयानी 0 बिकी — 8 दिन का streak है। Weekend special से replace करो।"

We cover this language angle in much more depth in RestroAI in Hindi and Hinglish.

Step 5 — Delivery via WhatsApp

The finished message goes out through Restrofi's WhatsApp Business integration (powered internally by Zavu). Cost-wise, each message costs Restrofi about ₹0.50 to send via Meta's pricing — but that cost is already factored into your plan, so there are no per-message charges to you.

The message arrives on the WhatsApp number you registered as your Restrofi owner number. Forward it to a manager, send it to the kitchen group chat, screenshot it for your CA — it behaves like any other WhatsApp message.

What RestroAI does NOT do

It's worth being explicit about this:

• RestroAI is not real-time. It's daily. Mid-day alerts are on the roadmap for a future tier.
• RestroAI does not take action automatically. It informs; you decide. It will never auto-update menu prices, auto-disable items, or auto-message customers.
• RestroAI does not replace your dashboard. Restrofi's analytics dashboard remains for deep dives. RestroAI is for daily decisions. See RestroAI vs traditional restaurant analytics dashboards for the comparison.
• RestroAI does not work outside India yet. The festival logic, GST awareness, and Hindi/Hinglish generation are India-tuned.

A worked example: a Tuesday morning

Here's how a real morning unfolds for a restaurant we'll call Biriyani Hub:

• Monday night, 11:45 PM: Restrofi's nightly batch kicks off. Biriyani Hub's last-24h data and last-30-day baseline get bundled and processed.
• 11:46 PM: RestroAI identifies that AOV is up 12% vs Monday baseline (good), but mushroom biriyani sold zero (eighth consecutive day), and tomorrow is the start of a long weekend.
• 11:46 PM: RestroAI selects 3 insights from the analysis: AOV increase (positive note), mushroom biriyani dead-item (action item), long-weekend prep (forecast).
• 11:47 PM: RestroAI writes the Hinglish version (Biriyani Hub's chosen language).
• 9:00 AM Tuesday: WhatsApp message lands on the owner's phone. Owner reads it during morning chai. Decides to drop mushroom biriyani from the menu and prep extra paneer for the long weekend.

That's the loop. Boring on paper, transformational in practice when it happens 365 mornings in a row.

The AI engine: built for Indian restaurants

RestroAI's AI engine is purpose-selected for the Indian F&B context. Its Hindi and Hinglish generation is meaningfully better than generic alternatives — "good enough" Hindi from an Indian restaurant owner's perspective is a high bar. The cost-per-analysis at the volume Restrofi runs is very competitive, and it handles structured-data prompts well — RestroAI feeds it tables of order data, not free-form text.

Restrofi runs the AI as a managed service — no GPUs, no fine-tuning, no model serving on our side. That keeps RestroAI affordable and means AI improvements flow through to your messages automatically.

Where to go next

If you're new to all this, start with What is RestroAI?. If you're trying to evaluate which Restrofi plan unlocks which RestroAI capabilities, RestroAI pricing covers it. And for the full breakdown of what every kind of insight looks like, RestroAI features walks through all 8 categories.