My employees each keep their Hong Kong Phone Number List Excel file, with a sales history of the references and customers that concern them. My store managers receive sales forecasts out in the pain of the data center but they want to have room for maneuver on the orders, because they know them, that the reference X is a little more requested in this period of the year. My supplies are in trouble every year, because a large customer creates peaks without knowing when they will intervene. I don’t know if I can trust a computer forecasting tool, can it take into account the complex intricacies that a human can make their own? You recognize yourself in one

of these examples. You may be considering a Proof of Concept (POC). But you either don’t have the skills for a specific development, or you don’t have the size to invest in massive computing time and capacity. Above all do not start from scratch The best algorithms are able to take into account several elements, which constitute selection criteria: Seasonality: The classic example is the sale of beer, which increases in summer Metadata: Network traffic at routers depends on the characteristics of those routers The demand for clothes depends on their color, style, brand External factors: Holidays Promotional events Series without

Above All Do Not Start From Scratch

history: Release of a new product No data collected on a quantity that interests us The first good news is that you (probably) don’t have to redo everything. Literature review, choice of architectures, implementation, training, comparative tests… All these challenges were taken up by giants like Amazon , because they too were faced with the need for forecasting. And for good reason, managing tens of millions of references is a tedious exercise. Add to this all the warehouses that accommodate stocks, workers, traffic on its websites … this is a lot of elements that interact and that the company has learned to tame and pack for easy use, on


demand: c t is the role of services like Amazon Forecast. The second good news is that it’s relatively inexpensive, especially on a small scale. With consumer pricing, it is not shocking to end up with an invoice that hardly exceeds a few hundred euros for the most modest, a few thousand for the ambitious for its POC. I take two examples given here : First example: Let’s say you own a clothing business and sell 2,000 items in 50 stores around the world. Each reference combined with a store represents a series, so you will have 100k (2000 references x 50 stores) series to plan. Suppose you load 5GB of data for this task, and a model

The First Good News Is That You

trains on this dataset for 20h. Total cost: $ 185.24 Second example: Say you are doing financial advice. Your customer has 2,000 ice cream outlets, and wants to forecast the cash flow for each outlet. Each combination of a flow and a point of sale corresponds to a series. So you have 2000 series to plan. Suppose you load 1GB of data to perform this task, and the chosen model trains for 4 hours on your data. Total cost: $ 4.65 Your costs will therefore be mainly linked to the load consumed by your employees or to the consultant who will have to find the data to provide and set up the necessary piping to provide the histories and enhance the forecasts. Concretely, how does this happen ? That’s it, my account is created on Amazon

Web Services . In terms of ergonomics, no need for a seasoned programmer to get started. I choose to tackle the avocado price prediction in some US markets to test. In a few clicks and a few minutes, I’m already able to train a model and output my first predictions (apparently it’s an ARIMA type model that will suit me the best!) I am also able to combine other series or metadata that could influence prices to refine forecasts and take advantage of correlations that are just waiting to be exploited. These data will be exported to a storage space, or directly called by my ERP using the APIs offered. I create a data group.

Leave a Reply

Your email address will not be published.