Tuesday, 12:30. This autumn an online appliance store in Almaty started promising "Delivery today" for orders placed before 14:00 within the city, if the item is in stock. At 10:30, after an update, the warehouse software stopped: the stock team can't see orders and can't pick them. The store has 11 orders to deliver today, some of them from last night: 8 from the website and 3 from its shop on Kaspi.kz. Meanwhile the website banner and the ads on Google and Instagram keep promising same-day delivery to new customers. The marketer finds out about the outage from the support chat.
This is the store from the lesson on the digital ecosystem, with about 115 orders a week from the website and Kaspi. The numbers are made up.
Marketing's first job here is to stop handing out a promise that can't be kept today, and to help the people who have already been given it. The crisis strategy can wait until evening.
What broke and who decides
Business crises work in different ways, and marketing has a different job in each:
| What happened | Who decides | What marketing does |
|---|---|---|
| An outage: warehouse, payments, website | Operations director, IT | Takes down the promise, keeps customers updated on status |
| Supply: the item is out of stock or takes longer to arrive | Purchasing | Pulls the item from ads, corrects stock and delivery times |
| A supplier raises prices | Owner and finance | Updates prices in ads, recalculates how much an order can cost |
| A manufacturer's recall | Manufacturer, lawyer | Provides the channels so every customer learns what to do |
Crisis communication researcher W. Timothy Coombs makes this his first recommendation for any crisis: everyone who may be affected receives instructions, including recall information (Crisis Management and Communications). The manufacturer and the lawyer write the text, and marketing makes sure every buyer of that model sees it.
Roles are best assigned before the first outage. In the UK Government Digital Service's incident management standard, an incident has two key people: the incident lead and the communications lead, who updates everyone according to a priority table, every hour for a critical incident (GDS Way: Incident management); the page itself was last reviewed in July 2024 and warns that it may be out of date. At the store, the operations director leads, the marketer handles communications, support messages customers and the owner decides on compensation. Announcing that the warehouse is back up is the incident lead's job.
Where the promise lives
Pausing one campaign takes the promise off that campaign only. The marketer works through a list:
| Where | What it says | What to do |
|---|---|---|
| Website banner | "Delivery today" | Take it down, replace with "Delivery times being confirmed" |
| Product page | Delivery time under the price | Show "tomorrow" or remove the time |
| Google Ads | "Delivery today" in the ad text | Pause the ads with that phrase |
| The "We'll deliver today" ad | Pause it | |
| Shop on Kaspi.kz | The delivery time customers see | Change it where the store sets it |
| An auto-reply about same-day delivery | Replace the text | |
| A scheduled "Order before 14:00" email | Reschedule it | |
| Phone sales staff | "We'll bring it today" | Warn them in the team chat |
After each change, the marketer opens the website in an incognito window and checks the ads' status: a change in the ad account doesn't yet mean the customer sees it. The list with the names of whoever edits each place will be needed again when the promise is brought back.
A "Delivery today" ad that can't be honoured today becomes, as we read it, unreliable advertising. The Law on Advertising (as amended up to 17 May 2026) requires advertising to be reliable (Art. 6(1)) and classes untrue information about delivery as unreliable (Art. 7(2)(4)).
What to tell people whose order is already in
At 13:10 support messaged all 11: the eight website customers on WhatsApp, the three Kaspi customers in whatever way the platform's rules allow. The website customers got this message:
Good afternoon! Your order has been accepted and the item is in stock, but it won't go out today: our warehouse software has stopped. We'll send you a new delivery time before 14:00. If waiting doesn't suit you, reply "cancel". Card payments are refunded the same way they were made.
The message doesn't explain the breakdown, but it has what the customer needs: what's happening with the order, what they can do and when the next message will come.
At 14:00 the warehouse was still down, and the update went out anyway: what's known (an outage after a software update, IT rolling back the version), what isn't yet (when picking will restart) and when the next message comes (16:00). In its incident communication tips, Atlassian suggests acknowledging the problem right away, briefly describing who's affected and promising further updates, then updating every 30 minutes or at whatever cadence suits the situation (Incident communication tips). The tips are written for service status pages. A two-hour cadence is enough for the store, as long as every message names the time of the next one.
By evening the picture looked like this:
| What | Orders |
|---|---|
| Due to arrive today | 11 |
| Customer replied to the message | 9 |
| Delivered the same day, after 16:00 | 7 |
| Delivered the next day | 3 |
| Cancelled | 1 |
A sent message only confirms that it was sent. Two people didn't reply, and the courier called them before setting off. An order counts as closed once it has been delivered or refunded.
If a customer submits a claim in person, by registered mail or to the email address the store has published (Art. 42-4(1)), and the store disagrees with it, the Law on Consumer Protection (as amended up to 20 August 2026) requires a reasoned written reply within ten calendar days (Art. 42-4(2)). This is a summary of the law, not legal advice.
When to bring the promise back
At 15:10 IT rolled back the update, and the warehouse started picking orders. But the banner stays down even after "the warehouse is back": last night's, this morning's and new orders are all in one queue, and the couriers' routes are already set. The promise comes back when three conditions are met:
- the queue of orders promised for today has been cleared;
- the couriers have confirmed they can take new same-day orders;
- support knows what to say if the outage happens again.
The store brought back "Delivery today" on Wednesday morning, at first only on the website and for orders before 12:00, and in the ads by lunchtime, once the first orders had left on time. The condition for pausing it again was written down in advance: if by 13:00 more than five of today's orders are still unpicked, the banner comes down again.
When costs rise and orders are below normal
A month later, suppliers raised their prices. The store raised prices on part of its range, and now keeps about 5,500 ₸ per order instead of the 7,000 ₸ used in the awareness test calculation. For two weeks running, orders were around 100 a week. The owner asks to cut all marketing by 30%.
Demand first. A hundred orders a week is the bottom of the store's normal range, 100 to 130, and two weeks can't tell a real drop from ordinary fluctuation. Conclusions about demand can wait, but the money has to be recalculated now.
How much the store can spend on ads per order is set by what it keeps from an order. Google Search brings about 40 orders a week at roughly 5,000 ₸ of ad spend each. At 7,000 ₸ per order the store kept 2,000 ₸; at 5,500 ₸ it keeps 500 ₸. Search is still in profit, but with almost no margin: if ad cost per order rises above 5,500 ₸, every order from search becomes a loss.
Cutting everything by 30% means cutting three different kinds of spending equally:
| Spending | Example at the store | What to do |
|---|---|---|
| Brings orders now and pays off | Google Search, product ads on Kaspi | Keep it and check cost per order against 5,500 ₸ every week |
| Needed by people who already bought | Emails to past buyers, warranty reminders on WhatsApp, replies to 2GIS reviews | Cut last |
| Works for later | Awareness videos, flyers | Pause first |
For search the rule is simple: if ad cost per order is above 5,500 ₸ two weeks in a row, lower the bids in the categories where the store keeps the least per order. Other scenarios and review conditions are written down the same way as in defending a budget, before the numbers start moving.
A cheap auction is a reason to expand, but only while ad cost per order stays below 5,500 ₸ and the warehouse and couriers can keep up.
When costs rise, marketing has one more duty: prices in ads. As we read it, the Law on Advertising requires the price to be true at the time the ad is distributed (Art. 7(2)(3)), so new prices go into ads, Kaspi listings and 2GIS on the same day as on the website.
Reviewing the outage
A week later the incident lead, the marketer and support rebuild the timeline: 10:30 the update and the warehouse stops, 12:30 marketing hears about it from the support chat, 12:50 the banner comes down, 13:10 messages to customers, 15:10 the rollback, Wednesday morning the promise returns. The weakest point turned out to be the two hours between the outage and marketing finding out: all that time, the ads kept bringing customers in under a promise that no longer worked.
The review is blameless, as the same GDS standard advises, and the fixes are written down with owners and deadlines. IT reports outages in a shared chat with marketing. The list of places where the promise lives is kept as a checklist. Warehouse software updates are scheduled for the evening, after the day's orders have gone out.
Exercise: a promise map for your project
Take your own project or a client's business.
There may be more places on the map than you'd expect: besides the ads, a delivery promise lives in auto-replies, scripts and marketplace listings nobody has opened since launch. The rows marked "don't know" are the main risk: in a crisis there's nobody to fix them.
An agreement with operations works when it's a named chat and a named person, not "we'll let you know if anything happens". Otherwise marketing finds out about an outage the way the appliance store did, from the support chat two hours later.
In the calculation with a fifth less per order, look at the channels where ad cost per order is already close to what an order brings in. They'll be the first to turn unprofitable, so write a rule for them in advance: at what cost per order bids come down, and who lowers them.