AI assistant · N4 Executable plan
Digital Crisis Protocol Builder for Restaurants
Writes the protocol for the first hours when something turns against you online: three levels with countable criteria, who speaks and who stays quiet, what gets checked before anyone writes, and where this stops being marketing.
N4Executable plan
A plan with owners, deadlines and the metric you review. Not advice: an agenda someone can execute on Monday.
◆ This assistant 23 of the 164 published assistants are at this levelIs this for you?
What it does and what you get
- On the day a video starts moving or a food-poisoning claim lands, nobody sits down to write a protocol: people answer from their phones, four of them at once, without anyone having pulled the ticket.
- This assistant writes it beforehand, with a cool head, and puts countable criteria where adjectives usually sit: how many pieces, across how many channels, with or without an image, and which words jump straight to the top level.
- It also brings the two parts almost nobody writes down: when to stay quiet because answering amplifies, and the exact border where the matter moves to the health authority, a lawyer or the insurer.
What you end up with
Five chained deliverables: the three-level table with its clock and its owner, the checking window that comes before the first sentence, the rule for when to stay quiet and when to speak, the watch round with its hours, and the border with the calling order.
EXAMPLE This is what you getAn example built with Sazón de Origen, the house's fictional restaurant.Not your data: your result is built with YOUR restaurant's.See the full example →Sample result
The owner asked this: "A customer's video complaining about the new location kept me awake for three nights, and four different people answered it, each on their own. I want the protocol written before the next one."
The baseline first: without it the levels are adjectives. A quiet day on Instagram (4,100 followers) brings 12 comments and 4 direct messages; the search profile, 2 reviews a week across both locations; the delivery platform, 6 a month. That is the zero everything else is counted against.
SUPUESTO: those four figures were built from the screenshots of an ordinary week you handed over, not from an industry average. If your real quiet week brings twice the comments, the level-3 threshold rises in the same proportion and the first column has to be rewritten before the table gets used.
| Level | What defines it, counted | Clock | Who decides | Who speaks |
|---|---|---|---|---|
| Nuisance | Up to 2 pieces about the same event in 24 hours, one channel only, no photo or video, nobody on the team named | Normal review clock: 24 hours on the search profile, 12 on the delivery platform | Andrés Gil, admin | Andrés Gil |
| Incident | 3 or more pieces about the same event in 24 hours · or a single one with photo or video · or the same event on 2 channels · or it names one of the 24 people | Checked within 2 hours · decision written before hour 6 | The owner | Marcela Ruiz, floor manager |
| Crisis | 12 pieces or more in 1 hour, a whole Instagram day packed into 60 minutes · or a video past 12,000 views, nearly three times the follower count · or any of these four words: poisoning, allergy, foreign object, authority | The owner knows within 60 minutes · in public, nothing until checking closes | The owner, nobody else | The owner |
The nuisance does not enter this protocol, and that is the first discard: it is answered like any review, on its own clock, and it wakes nobody on a Sunday. Raising everything to crisis costs exactly what having no levels costs.
What the house has on the table in that first hour, so the call is made with the figure in sight: on a Sunday midday the location moves some 74 covers at an average ticket of 48,000 COP, close to 3,552,000 COP through the till. Half an hour of checking does not cost that. Speaking too early can, and more than once.
Half an hour, five checks, three people. The clock of the first hour runs for checking, not for posting. Nothing public goes out before this table closes.
| What is checked | Where you look | Who brings it | Minutes | What it settles |
|---|---|---|---|---|
| The POS ticket for the hour named: table, order time, delivery time | Till system of the location the piece names | Marcela Ruiz | 8 | Whether it happened here, and that day |
| Who was on shift in kitchen, floor and own delivery | Shift sheet for that day | Marcela Ruiz | 5 | Who gets asked first, in private and without the piece in hand |
| The piece itself: date, recognisable location, dish recognisable against the menu | The video or photo, zoomed in | Andrés Gil | 6 | Whether it is this house at all. A dish that is not on the menu closes the matter in 6 minutes |
| If the complaint is about portion: the weight served against that dish's standard recipe | Recipe card and kitchen scale | Head chef | 4 | Whether the shift's portioning has to be fixed before anything gets answered |
| The supplier lot of the dish named, and whether a second person reports the same symptom | Delivery note, walk-in with a photo of the label, and reviews of the last 72 hours | Andrés Gil | 7 | Whether the dish comes off the menu today and whether this moves to level 3 |
Thirty minutes in total. When the piece mentions health, the five rows still get done, but in parallel with the call to the owner: checking delays the public word, never the internal alert.
And what comes out of here is kept: screenshot with the hour, ticket number, shift sheet, photo of the lot label, all in one folder named with the date. If this ends up with the lawyer or the insurer, that folder is the difference between a version and evidence.
Silence is decided, signed and reviewed. Not answering is not doing nothing: it is a decision with an hour on it.
| Case | Speak or stay quiet | Why | Review hour |
|---|---|---|---|
| Account with no posts and no followers, event unchecked | Stay quiet | Answering in public lends 4,100 readers to someone who had none | 6 hours |
| Video with the fact checked, already seen by more people than you have followers | Speak once, in the same thread, without moving it to private on the first line | Your own customers are reading it now, not the person who filmed | 2 hours |
| The piece shows a server's face and name | Stay quiet in public · speak with him the same day | Defending him in the thread exposes him more than it protects him | Before the shift closes |
| The matter is already with the health authority, a lawyer or the insurer | Stay quiet, always | Any public sentence is evidence | No review: it lifts when the lawyer says so |
| A customer asks on your own profile, "is this true?" | Always answer | That is your customer, asking in your house | Every 4 hours while it lasts |
| A claim spreads that the sancocho contains something it does not | Speak, with the standard recipe and a figure | A false detail left unanswered becomes the detail | 24 hours |
The holding line, at the 2-hour mark. Three things only: that you saw it, that you are checking it, and what time you come back. It sounds like this: "We saw the video this morning. We are pulling the ticket and the supplier lot, and we will come back today before six with what we find." Twenty-seven words, none of them spare.
Everything else stays out of that line: the cause, which is not conceded while it is unchecked; the employee's name; any promise about the outcome; money or food in exchange; and the "it's false" said before the previous table closes.
Illustrative example generated with a fictional restaurant. Not a promise of results, not a client case.
This is 3 of 5 parts. Behind the wall sit the two parts that get operated. The **watch round** in five columns, capped at 30 minutes a day between two people, with its hours and the two alert-group rules. And the **border**: five signals with the exact calling order —owner, lawyer, insurer, authority if the local rule requires it—, what goes in writing to each, what is never posted about any, and the 5-point checklist for the dry drill.
See the full example — free accountWorks with these AIs
How to use it
Which of your data it uses
This assistant works with 13 of the 23 fields in «My Restaurant», among them Restaurant name, Business type and City and country. It does not ask out of curiosity: these are what make the answer speak about YOUR scale instead of an industry average — the same calculation on a business of another size returns a number you cannot decide with. You fill them once and they apply across the library, so the second assistant you open already starts with them in place. Whatever you leave blank, the assistant asks for one thing at a time instead of inventing it, and says so before answering, so you know what it is working from.
Always up to date
Every assistant carries a visible version and date. When an AI changes how it works, the assistant's version goes up and you see what changed.
v1.1.12026-08-26The body stops contradicting the answer ladder. It used to say «do not move on until it arrives», the very order the ladder was born to replace: the assistant asked the owner for what the owner does not have and stalled. It now says what the house doctrine says — ask one at a time, and if they do not know, continue with a marked assumption. FORMAT OF THE RESULT was not touched, so the example still corresponds and its seal moves up with the card.v1.1.02026-08-25Body rewritten to pay down duelo debt: it scored 75.3 out of 90.0 with method and format already full. What sat on the floor was `lexico` —1 of 10: a restaurant crisis protocol naming only the POS ticket, when a portion complaint is settled by checking the served weight against the standard recipe, and a piece arriving through delivery is not checked against the dining-room shift sheet— and `cuidados`, at 2 of 6, because the traps of the trade lived in their own section («TRAMPAS DEL OFICIO») outside the house's seven headings. That section is now split between ROLE and BEFORE YOU DELIVER, where whoever is working actually reads them. A SUPUESTO: line is added so the quiet-day baseline is declared for what it is —a figure built from the owner's own data, not an industry average— and the example shows it. Rises to 96.7. `limites_es`/`limites_en` were already written and are left as they are.v1.0.02026-08-25The assistant is born.More assistants in this category
Writes the messages your restaurant answers automatically -welcome, booking, delivery, opening hours, address,…
N4 · Executable planWrites the replies to four and five-star reviews: sorts them by what each one brings, hands out a long reply o…
N4 · Executable planBuilds the flow your restaurant uses to win back a guest who left angry: who enters and who does not, the priv…
Included in the always-growing library
Access to every published assistant, adapted to your AI and personalised with your restaurant's data, updates included.