Predict your store demand
Anticipate how much you could sell in the coming days, how much stock to order from your supplier and when to run a promotion without running out of product. With a real error margin calculated from your own sales, not industry averages.
Only available within Shopify Admin.
FIND THE RIGHT ANALYSIS ENGINE FOR YOUR STORE
A forecast doesn't replace your judgement, but it can help you decide better
1. Questions
How often do you check your sales?
How long have you been selling?
Do you run offers or discounts?
Does having too much or too little stock bother you?
How do you usually decide your supplier order?
TRANSPARENT ARTIFICIAL INTELLIGENCE
What we recommend in each case
If you answer Do you run offers or discounts?: Yes · How long have you been selling?: More than a year
For your store, the better one is:
XGBoost (the spreadsheet with rules)
It doesn't arrive trained: it has to be taught with the business's own data. With enough history it can learn from discounts, holidays and other recorded variables.
You track discounts and you have also been selling for a while. It is the two together that change the answer, not either one alone.
Karlsruhe compared both ways of working across 30 datasets. On 6, XGBoost or the classical methods beat the already-trained models, with between 21 and 2,768 samples of their own. On another 9 there was a crossover point, between 24 and 8,361 samples. What decided it was not having extra columns: it was how much history of their own was available. Karlsruhe (KIT), July 2026, break-even analysis.
Virginia Tech gave nobody those columns: price only. There XGBoost finished behind copying the last value, 0.823 against 0.775. Without additional information it does not hold up; with it, and with history of your own, it does. Virginia Tech, January 2026, Table 2.
One thing worth knowing before deciding: Chronos-2 also takes those columns natively, including the ones known in advance, such as a discount already planned. Tracking discounts does not force you away from Chronos-2.
And a warning: XGBoost has to be trained and maintained with the business's own data. If there is going to be nobody to build it and keep it up to date, this is not your answer.
Karlsruhe did not measure a Shopify store, and it did not measure discounts or holidays either. It measured from how much history of your own the classical methods catch up with the already-trained ones. Karlsruhe (KIT), July 2026.
Today you copy last year. In Virginia Tech that came 14th of 17 (error 0.973). Copying just the last figure came 6th (0.775): much harder to beat. Virginia Tech, January 2026, Table 2.
Today you use a spreadsheet, but with sales only. In Virginia Tech that idea (XGBoost with no extra columns) came 8th, error 0.823: worse than copying the last value (0.775). Virginia Tech, January 2026, Table 2.
The bar is copying the last number. If a method doesn't beat that in the same race, it isn't adding anything. Virginia Tech, January 2026, Table 2.
If you answer How long have you been selling?: A few months · How often do you check your sales?: Every day · Does having too much or too little stock bother you?: Yes
For your store, the better one is:
Chronos-2
It arrives already trained. It takes the store's history and works out a central forecast plus a range for a slow week and a strong week.
You order often or have little history. The study that measured exactly that is the Texas one: hour-by-hour electricity demand, with only 24 hours of history. Texas (ERCOT), February 2026.
In that window, copying the same hour yesterday got 0.591 and Chronos-2 got 0.659: with only 24 hours, Chronos-2 does not beat copying yesterday. Moirai-2 went to 1.304. Prophet broke down, with error above 74, and the accountant SARIMA too, 15.1. Texas (ERCOT), February 2026, 24-hour context.
What hurts you is running out. The warning was measured with 512 hours of history: Chronos-2 covered 95% of cases when it claimed to be 90% sure. Moirai-2 covered 70%: its warning fell short. Texas (ERCOT), February 2026, coverage of the 90% interval.
The method is Chronos-2, not because its middle number is the best, but because it warns better when the week runs slow. Texas is electricity, not your shelf: it's the race that exists for little history. Texas (ERCOT), February 2026.
Today you copy last year. In Virginia Tech that came 14th of 17 (error 0.973). Copying just the last figure came 6th (0.775): much harder to beat. Virginia Tech, January 2026, Table 2.
Today you use a spreadsheet, but with sales only. In Virginia Tech that idea (XGBoost with no extra columns) came 8th, error 0.823: worse than copying the last value (0.775). Virginia Tech, January 2026, Table 2.
The bar is copying the last number. If a method doesn't beat that in the same race, it isn't adding anything. Virginia Tech, January 2026, Table 2.
If you answer How long have you been selling?: A few months · How often do you check your sales?: Every day
For your store, the better one is:
Chronos-2
It arrives already trained. It takes the store's history and works out a central forecast plus a range for a slow week and a strong week.
You order often or have little history. Texas measured exactly that: 24 hours of history. Texas (ERCOT), February 2026.
In that window: copying the same hour yesterday 0.591. Chronos-2 0.659. Moirai-2 1.304, worse than copying yesterday. TinyTimeMixer 5.73. SARIMA 15.1. Prophet above 74. Texas (ERCOT), February 2026, 24-hour context.
The method is Chronos-2, and it is worth saying it with the figure in front of you: with only 24 hours it does not beat copying yesterday. Where it rules is the warning: when it said «I'm 90% sure», it covered 95% (measured with 512 hours of history). Prophet and the accountant are useless here. Texas is electricity, not your shelf. Texas (ERCOT), February 2026, coverage of the 90% interval.
Today you copy last year. In Virginia Tech that came 14th of 17 (error 0.973). Copying just the last figure came 6th (0.775): much harder to beat. Virginia Tech, January 2026, Table 2.
Today you use a spreadsheet, but with sales only. In Virginia Tech that idea (XGBoost with no extra columns) came 8th, error 0.823: worse than copying the last value (0.775). Virginia Tech, January 2026, Table 2.
The bar is copying the last number. If a method doesn't beat that in the same race, it isn't adding anything. Virginia Tech, January 2026, Table 2.
If you answer How often do you check your sales?: Every month
For your store, the better one is:
Chronos-2
It arrives already trained. It takes the store's history and works out a central forecast plus a range for a slow week and a strong week.
You order monthly, have decent history and only track what sold. That resembles the only 2026 race where almost every method on this list ran together: Virginia Tech, monthly grain prices. Virginia Tech, January 2026.
Chronos-2 scored 0.736. Time-MoE finished ahead (0.693); it's not on the list we compare, and it's not Amazon's. TimesFM 2.5 tied with it, also 0.736. Moirai-2 0.751. Copying the last value 0.775, and that already beat the spreadsheet (0.823), the accountant SARIMA (0.859), ETS (0.921) and Prophet, last of seventeen (1.291). Virginia Tech, January 2026, Table 2.
The method is Chronos-2: it finished ahead of everything classic on the most complete grid on the list. The test is grain, one data point per month, not your Shopify. Virginia Tech, January 2026, Table 2.
Today you copy last year. In Virginia Tech that came 14th of 17 (error 0.973). Copying just the last figure came 6th (0.775): much harder to beat. Virginia Tech, January 2026, Table 2.
Today you use a spreadsheet, but with sales only. In Virginia Tech that idea (XGBoost with no extra columns) came 8th, error 0.823: worse than copying the last value (0.775). Virginia Tech, January 2026, Table 2.
The bar is copying the last number. If a method doesn't beat that in the same race, it isn't adding anything. Virginia Tech, January 2026, Table 2.
If you answer Does having too much or too little stock bother you?: Yes
For your store, the better one is:
Chronos-2
It arrives already trained. It takes the store's history and works out a central forecast plus a range for a slow week and a strong week.
You order weekly and only track sales. There are two studies. Their figures are never mixed.
In Virginia Tech, the most complete picture on the list, Chronos-2 finished ahead of Moirai-2 and of copying the last value. Prophet, last of seventeen. Virginia Tech, January 2026, Table 2.
In Texas, with lots of history, Moirai-2 estimated the number slightly better (0.307) than Chronos-2 (0.334). Texas (ERCOT), February 2026, 2048-hour context.
What hurts you is the empty shelf. In Texas the warning, measured with 512 hours of history: Moirai-2's covered 70% when it promised 90%. Chronos-2 covered 95%. That's why the method is Chronos-2: its «if it runs slow» warning truly covered. Neither Virginia Tech nor Texas is a weekly store; they're the races that exist. Texas (ERCOT), February 2026, coverage of the 90% interval.
Today you copy last year. In Virginia Tech that came 14th of 17 (error 0.973). Copying just the last figure came 6th (0.775): much harder to beat. Virginia Tech, January 2026, Table 2.
Today you use a spreadsheet, but with sales only. In Virginia Tech that idea (XGBoost with no extra columns) came 8th, error 0.823: worse than copying the last value (0.775). Virginia Tech, January 2026, Table 2.
The bar is copying the last number. If a method doesn't beat that in the same race, it isn't adding anything. Virginia Tech, January 2026, Table 2.
If you answer How long have you been selling?: More than a year
For your store, the better one is:
Chronos-2
It arrives already trained. It takes the store's history and works out a central forecast plus a range for a slow week and a strong week.
Moirai-2
It also arrives trained in advance. With plenty of history its central forecast was slightly better in Texas, but its safety range fell short.
The two studies don't pick the same method. Below you get both, each with its paper. There is no blended winner.
You order weekly, have more than a year of sales and only track units. The two studies don't pick the same one. There is no blended champion.
Virginia Tech (almost the whole list, one data point per month): Chronos-2 ahead of Moirai-2 (0.736 vs 0.751). Virginia Tech, January 2026, Table 2.
Texas with 2048 hours of history: Moirai-2 0.307, Chronos-2 0.334. Texas (ERCOT), February 2026, 2048-hour context.
If you want the picture of the list, Chronos-2. If you want the number for lots of hourly history, Moirai-2. Don't add the figures. Your store's week is neither the grain's month nor the grid's hour.
Today you copy last year. In Virginia Tech that came 14th of 17 (error 0.973). Copying just the last figure came 6th (0.775): much harder to beat. Virginia Tech, January 2026, Table 2.
Today you use a spreadsheet, but with sales only. In Virginia Tech that idea (XGBoost with no extra columns) came 8th, error 0.823: worse than copying the last value (0.775). Virginia Tech, January 2026, Table 2.
The bar is copying the last number. If a method doesn't beat that in the same race, it isn't adding anything. Virginia Tech, January 2026, Table 2.
For your store, the better one is:
Chronos-2
It arrives already trained. It takes the store's history and works out a central forecast plus a range for a slow week and a strong week.
You only track sales and your history is months, not days. The most complete picture on the list is Virginia Tech. Virginia Tech, January 2026.
Chronos-2 finished ahead of Moirai-2 and of copying the last value. Copying the last value already beat the spreadsheet, the accountant and Prophet (last of seventeen, 1.291). Lag-Llama doesn't run on that grid. Virginia Tech, January 2026, Table 2.
The test is monthly and agricultural. The name comes from that picture, not from a miracle on your store. Virginia Tech, January 2026.
Today you copy last year. In Virginia Tech that came 14th of 17 (error 0.973). Copying just the last figure came 6th (0.775): much harder to beat. Virginia Tech, January 2026, Table 2.
Today you use a spreadsheet, but with sales only. In Virginia Tech that idea (XGBoost with no extra columns) came 8th, error 0.823: worse than copying the last value (0.775). Virginia Tech, January 2026, Table 2.
The bar is copying the last number. If a method doesn't beat that in the same race, it isn't adding anything. Virginia Tech, January 2026, Table 2.
See the technical detail and the studies
TRANSPARENT ARTIFICIAL INTELLIGENCE
Picture one product: a t-shirt
8 weeks ago
Sold 40 t-shirts
7 weeks ago
Sold 38 t-shirts
6 weeks ago
Sold 42 t-shirts
5 weeks ago
Sold 41 t-shirts
4 weeks ago · there was a discount that week
Sold 55 t-shirts
3 weeks ago
Sold 39 t-shirts
2 weeks ago
Sold 44 t-shirts
Last week
Sold 42 t-shirts
What each method would do with these sales
Chronos-2
How it reads the sales
It looks at the eight weeks as they are. Since we haven't told it the 55 t-shirts were sold during a discount, it reads that rise as a week of higher demand.
What it returns
It doesn't give a single figure. In this example it works out three possibilities:
A slow week
36
A normal week
42
A strong week
52
These figures are only here to explain the example. Chronos-2's position in the comparison comes from the Virginia Tech study (Table 2).
In money
If each t-shirt costs $20, that would be between $720 and $1,040, with $840 as the central scenario.
Moirai-2
How it reads the sales
It looks at the same eight weeks and forecasts with a model that arrives already trained.
What it returns
It also works out a slow week, a normal one and a strong one, but its range is narrower:
A slow week
38
A normal week
42
A strong week
47
That narrower margin may look better, but in the Texas study it fell short: when it should have covered 90% of cases, it covered 70%.
In money
At $20 per t-shirt, that would be between $760 and $940, with $840 as the central scenario.
XGBoost (the spreadsheet with rules)
How it reads the sales
XGBoost works differently. As well as the sales, it can use information such as discounts, holidays or other conditions for each week.
If we tell it there was a discount the week of the 55 t-shirts, it can take that into account. If we don't, to it they are simply 55 more sales.
What it returns
In this example it estimates around 40 t-shirts.
A slow week
—
A normal week
40
A strong week
—
Unlike the other two methods, here we have no slow week and no strong week around the forecast. In the Virginia Tech study, XGBoost without extra information finished behind simply repeating the last value.
In money
At $20 per t-shirt, a forecast of 40 t-shirts would be around $800.
And that is where the order comes from
Say the central forecast is 42 t-shirts. You want 10 more as a margin in case the week runs strong, and you already have 15 in stock.
42 + 10 − 15 = 37
You would order 37 t-shirts from your supplier.
The ranking isn't from this example. It's in the studies.
TRANSPARENT ARTIFICIAL INTELLIGENCE
Information
Last updated: 2026-08-30
What each method is
Chronos-2
It arrives already trained. Before ever seeing your store's sales it has learned patterns from large collections of time series. That is why it can take a new history and produce a forecast without first training on that store's data.
In the studies on this page it is tested exactly that way: it receives the available history and works out what may happen next. As well as a central forecast, it can give a range for a slow week and a strong week.
Beyond the history it also accepts additional columns —discounts, holidays and other conditions known in advance— without needing to be trained. They were not used in the studies on this page: there, every method received the history alone.
Moirai-2
It also arrives trained in advance. Salesforce has trained it on large collections of time series, so it too does not need to start from scratch each time it reaches a new store.
It takes the available history and forecasts using what it learned during that earlier training. In the Texas study, with plenty of history, its central forecast was slightly better than Chronos-2's, but its safety range fell short: when it should have covered 90% of cases, it covered 70%.
XGBoost (the spreadsheet with rules)
XGBoost works differently: it does not arrive trained for your store. It has to be taught with data from the specific problem before it can be used to forecast.
It can learn from several columns — sales, discounts, holidays or other variables that have been recorded. The more useful data it has to train on, the more chance it has of learning relationships particular to that business.
That advantage has a cost: it needs data to learn from. That is exactly why Karlsruhe studies at what point having enough history of your own lets XGBoost or the classical methods match or beat the models that already arrive trained.
The engine is Chronos-2, from Amazon. What ApisDom puts on top of it —validating the history before running it, deciding whether the data supports a defensible forecast, calculating the error margin from that store's real sales and marking the closed days— is what turns that forecast into something a shop can actually decide with.
Documentation of ApisDom's predictive technology
The studies
Virginia Tech, January 2026
The Promise of Time-Series Foundation Models for Agricultural Forecasting: Evidence from Commodity Prices
Source: Virginia Tech · Le Wang, Boyuan Zhang · 2026
They compared 17 methods using monthly prices of corn, soybeans, wheat and cotton between 1997 and 2025. All of them received exactly the same information: past prices only. They were given no discounts, no holidays and no other additional variables.
There is an important difference here: models such as Chronos-2 and Moirai-2 already arrived pretrained and were used without being trained on those commodities. Other methods had to learn from the data available for the test.
It is not a Shopify store or a daily order. It serves to show what happens when everyone competes using history alone, with no additional information.
One of the authors, Boyuan Zhang, works at Amazon, the company that develops Chronos-2. The study states that the research is independent of his position. Also, the method that finished first, Time-MoE, does not belong to Amazon. We keep this visible so you can weigh the results with the full context.
| Time-MoE | 0.693 |
| Chronos-2 | 0.736 |
| TimesFM 2.5 | 0.736 |
| Moirai-2 | 0.751 |
| Copy the last value | 0.775 |
| Spreadsheet with rules (XGBoost) | 0.823 |
| The accountant (SARIMA) | 0.859 |
| ETS | 0.921 |
| Copy last year | 0.973 |
| Seasons and holidays (Prophet) | 1.291 |
Texas (ERCOT), February 2026
Time Series Foundation Models for Energy Load Forecasting on Consumer Hardware: A Multi-Dimensional Zero-Shot Benchmark
Source: Investigador independiente (Turín) · Luigi Simeone · 2026
This study used Texas electricity demand between 2020 and 2024. It compared what happened when the models were given very little history, just 24 hours, and then a great deal more, up to 2,048 hours.
Chronos-2 and Moirai-2 were not trained from scratch on the Texas data: they were used as already pretrained models. That makes it possible to test one of their potential advantages: what they can do when they arrive at a new problem with barely any history available.
The study also checked whether their margins were reliable. It didn't only look at which model got the central figure closest, but at whether, when it said the result would fall inside a range with 90% confidence, that range really covered what happened.
It is not a store and it does not measure product sales. We use it because it lets us compare little history against plenty of history, and check whether each model's ranges truly hold. XGBoost did not take part in this study.
| Copy the same hour yesterday | 0.591 |
| Chronos-2 | 0.659 |
| Moirai-2 | 1.304 |
| TinyTimeMixer | 5.73 |
| The accountant (SARIMA) | 15.1 |
| Seasons and holidays (Prophet) | above 74 |
| Moirai-2 | 0.307 |
| Chronos-2 | 0.334 |
| The accountant (SARIMA) | 0.370 |
| TinyTimeMixer | 0.450 |
| Seasons and holidays (Prophet) | 0.610 |
| Copy the same hour yesterday | 0.749 |
| Chronos-2 | covered 95% when it claimed to be 90% sure |
| Moirai-2 | covered 70% |
| Prophet | covered 70% |
Karlsruhe (KIT), July 2026
When Do Foundation Models Pay Off? A Break-Even Analysis of Pretrained Time Series Forecasters
Source: Instituto Tecnológico de Karlsruhe (KIT) · Nicholas Tan Jerome, Frank Simon · 2026
This study was made precisely to answer a very practical question: when is it worth using a model that already arrives trained, and when is it worth training one on your own data?
Moirai and Lag-Llama had already learned from large collections of series before entering the test. They could start forecasting without training on each new dataset. XGBoost, ETS and ARIMA start from a different position: they need to use the available data from the specific problem to tune how they work.
The researchers compared both ways of working across 30 datasets, gradually increasing the amount of local history available.
On 15 of the 30 datasets, the models that already arrived trained won at every amount of history.
On 6, XGBoost or the classical methods managed to beat them even with little data of their own, between 21 and 2,768 samples.
On another 9 there was a point at which having more history changed which option was best. That point appeared between 24 and 8,361 samples, depending on the dataset.
The study also produces one especially clear rule: with fewer than 700 data points and appreciable seasonality, it recommends using the pretrained model directly and not retraining it.
About the questions
How often do you check your sales?
If you order daily, the study about very little history is the Texas one (hour by hour). If you order monthly, the one with the most methods together is Virginia Tech (one data point per month).
How long have you been selling?
Texas measured what happens with very few hours of history. Virginia Tech measured series spanning years.
Do you run offers or discounts?
None of the three studies measured discounts: in Virginia Tech and Texas nobody was given that column, and Karlsruhe measured something else, how much history of your own is needed. This answer counts together with the history one, not on its own.
Does having too much or too little stock bother you?
Texas measured whether the warning «you'll sell between this and that» truly covered what happened. A well-calibrated margin protects in both directions: from buying too much and from falling short.
How do you usually decide your supplier order?
This doesn't pick the method. It tells you whether what you do now finished ahead or behind in the same race.
TRANSPARENT ARTIFICIAL INTELLIGENCE
Quality Shield
Before each prediction, Forecast reviews your store sales history and classifies it with a quality traffic light: green, yellow or red. If the data is not sufficient, the engine does not run and no credit is consumed. Every credit spent produces a usable result.
Green
Your store has a solid history. The prediction engine runs and you receive the complete forecast with three scenarios (conservative, central and optimistic) and the real error margin calculated on your own history. 1 credit is consumed because it produces a usable result.
SBC Classification, Syntetos-Boylan-Croston 2005. ADI and CV² metrics on real store data.
Red
Your store does not yet have enough history for a reliable forecast. The engine does not run and no credit is consumed. Your credits stay in your balance, waiting. When your history reaches the required level we will email you so you can use them.
Only triggered by lack of sales history. The way you sell, in bursts or by season, never blocks you.
Yellow
Your store has enough history to forecast. You receive the complete forecast with three scenarios and the real error margin calculated on your own history. The engine runs and 1 credit is consumed. The range between the conservative and the optimistic scenario helps you decide how much stock to order with the information in front of you.
Variable sales, sales in bursts, or recent pauses. The engine still runs: you get your forecast with a clear warning and you decide.
Blue: Exceptional stability
Once the prediction has been generated, if the engine detects that your sales are so stable that the daily average is already a reliable reference, it tells you with an Exceptional stability notice. In that case, next time you can use that average directly and save the credit, or generate the forecast to confirm that the stability holds. This is information that appears after the prediction, not a state of the prior traffic light.
Triviality test: standard deviation below 5% over the last 30 days.
How it works
Install the app
From the Shopify App Store in seconds. No technical setup or external integrations. On install you receive 2 free credits to try the engine with no commitment.
Connect your sales
The app reads your order history directly from Shopify. Before each prediction the quality traffic light evaluates the data: if the history is not enough, the engine does not run and no credit is consumed. It is similar to a weather forecast: with very few data points any number would be made up.
Get your projection
Select the period, choose the metric and generate the prediction. You get three scenarios with real error margins. Next to the chart, a message evaluates the confidence of the result and, if your sales are so stable that the average alone suffices, you receive an Exceptional stability notice.
Free course: learn to read your forecasts
A 10-module PDF guide in business language: how much data you need, what each traffic-light colour means and how to use the scenarios to order stock without gambling. No coding, no statistics.
Download the course in PDF (Spanish)Download the course in PDF (English)Video: stop guessing, learn to forecast your demand
Seven minutes explaining how to turn your sales history into useful forecasts for your business, with no coding and no statistics. The same thing the app does, explained calmly. (Video in Spanish.)
Watch the video (7 min, Spanish)Watch the video (English)Start today
ApisDom Forecast is now available on the Shopify App Store. When you install it you get 2 free forecasts
Professional support included
Every merchant with credits has access to the integrated support system directly from the application.
Create categorised tickets with automatic diagnostics, attach screenshots if needed and receive a response within 24 business hours.
Your enquiry history is saved so you never have to repeat information.
Try Viral by ApisDom
Want to analyse your own data? With Viral you can upload any time series and get predictions with the same ApisDom technology. Social media, offline sales, web traffic.
Frequently asked questions
What is ApisDom Forecast and what is it for?
ApisDom Forecast is an application for Shopify stores that predicts how much you will sell in the coming days, weeks or months. It helps you decide how much stock to order from your supplier, when to launch a campaign and when to drop prices to clear inventory. It works on your store actual sales history, not on industry averages.
How do I know the forecast is reliable and not a made-up number?
Each forecast includes three scenarios (conservative, central, optimistic) and a real error margin calculated on your own store history. If your data is not sufficient for a reliable forecast, the app does not run it: a prior system, the quality traffic light, blocks the analysis and no credit is consumed. You only receive a forecast when there is a statistical basis to make it useful.
How much does it cost? Do I have to commit to a subscription?
There is no subscription. You only pay when you use the app, through a credit system that never expires: you buy a pack when you need it (from 14.99 USD), generate forecasts when you need them and the leftover credits wait in your account for next season. When you install the app you receive 2 free credits to try the engine with no commitment.
I just opened my store. Will it work for me now or do I have to wait?
If your store does not yet have enough sales history, the engine does not run the forecast and no credit is consumed. Your 2 free credits stay in your balance, waiting, until your store has enough data; we will then email you so you can use them. It is similar to a weather forecast: with very few data points any number would be made up, and that is not a responsible basis for business decisions.
Who is behind the technology? Is this real AI?
The prediction engine is Amazon Chronos-2, the time-series artificial intelligence model developed by Amazon Science. ApisDom has built a microservices layer on top of Chronos-2 that validates your data, manages quality, calculates confidence intervals and returns the real error margin measured on your history. The base technology is Amazon Science. The intelligence that adapts it to a Shopify store is ApisDom.
What happens to my data? Is it safe?
Your data is processed on servers in the European Union (Google Cloud, the Netherlands) and is never sold or shared with third parties. ApisDom Forecast does not store personal information about your customers: it only works with daily aggregated sales totals. We comply with the GDPR and you have a panel to export or delete your data at any time.