Introducing Forecast

Contribute to Forecast

We're looking for help developing Forecast. If you have specific business needs, or specific forecasts or reports you'd like to see developed, please get in touch

What is Forecast? 

Put simply, Forecast is a new reporting module to create accurate forecast metrics for turnover and rebate.

Would you like to know how much the group is likely to spend on Power Tools this year? Would you like to know how rebates will be affected if your largest member leaves next month? Would you like to know the effect a new member joining in Scotland will have on overall purchasing power?

These are all question we're trying to answer by developing Forecast.



Before we dive into what Forecast actually does (or will do), first, let's backup and talk about forecasting in general. Companies today use everything from simple spreadsheets to complex financial planning software to attempt to accurately forecast future business outcomes. These tools build forecasts by looking at a historical series of data, which is called time series data. For example, such tools may try to predict the future purchases of a member by looking at their previous purchase data with the underlying assumption that the future is determined by the past. This approach will generally fail to produce accurate forecasts because data has irregular trends and shifting variables - both internal and external. Even when forecasting uses seasonality models, it's still looking only at historical purchase trends, and not taking into account other data series that change over time (such as category trends, member branch size) with relevant independent variables like store locations or the general state of the economy at the time.


Seasonality is not good enough

Using Machine learning and AI, when you provide data to Forecast it will examine it, identify what is meaningful, and produce a forecasting model capable of making predictions that are up to twice as accurate than looking at time series data alone.

Your data alone is not good enough

If you forecast only using your own internal data, regardless of how verbose it is, you will not be able to factor in external data which can affect your historic data and therefore your forecast.

Forecast automatically imports publicly available data which compliments your internal data. For example, Forecast automatically includes the following time related data:

  • Unemployment rate per month for the UK
  • All majoir exchange rates averaged per month, including
    • GPB > EURO
    • GBP > USD
    • GBP > YUAN
    • EURO > YUAN
    • EURO > USD
  • UK Weather, including:
    • Average temperature per month
    • Cumulative amount of rainfall per month
    • Cumulative number of sunny days per month
    • Cumulative amount of snowfall per month
    • Maximum windspeed per month

All of these things can influence your historic data in ways you cannot understand without spending weeks or months analyzing the data at a micro and macro level. Using advanced AI, Forecast finds relationships between these variables which it uses and factors into its forecasts.

Now, you may be thinking "that's great, but a wet July affects my sales, not my purchases!". It doesn't matter; if these are lagging indicators or if they have no bearing on your purchases or inventory levels, Forecast work that out and weight your data accordingly, automatically.

Time for an example

Le's say it's August, and you want to forecast group spend to the end of the year. With completely static variables, forecasting with simple seasonality data is generally adequate. However, the world is not static, and neither is your business.

So, let's add some variables. Let's say in the last few months, two of your members have changed.

  • One member has recently left the group. They had 5 branches.
  • Another has made an acquisition, and that new company has 5 branches.

Ok, on the face of it you've got five new branches, and have lost five branches - so it would be tempting to see these i reductionist form, but let's add some further variables we know about to demonstrate how unwise that would be:

  • The member who has left was predominantly a fencing supplier in Scotland.
  • The new branches the second member has added are kitchen showrooms in London.
  • Kitchens and bathroom sales are up 8%, YoY
  • Timber deals generally have lower rebate earnings.
  • Kitchens generally attract high rebates
  • Members with branches in London are up 12% YoY
  • Fencing sales are down 4% YoY
  • Members with branches in Scotland are down 1% YoY

‚ÄčLet's add some further external variables to really muddy the water:

  • We had an unseasonably long and hot summer this year
  • The pound has weakened against the Euro this year by 4%
  • The UK unemployment rate ticked up by 2% last quarter

You can hopefully see that simply looking at last years spend and thinking it is going to any better than throwing spaghetti at the wall is a fools errand. Furthermore, if you include last years spend in a seasonality seasonality template that wouldn't help your forecast either.

eBiz Forecast can factor in all these variables using ;Artificial Intelligence developed over 10 years and used in the retail business at to analyse, and find patterns in your data by using deep learning algorithms to provide predictions that can be trusted, back-tested, and that are considerably more accurate than using seasonality and time series data alone.



What can Forecast do?

Forecast can:

  • Predict future spend and rebates - up to 25% of your time series length into the future (for example, if you have four years of data, it can predict one year ahead)
  • Can be filtered and queried by supplier, member, deal and category
  • Use 5 different AI algorithms for it's predictions: Autoregressive Integrated Moving Average (ARIMA) Algorithm, DeepAR+ Algorithm, Exponential Smoothing (ETS) Algorithm, Non-Parametric Time Series (NPTS) Algorithm and the Prophet Algorithm.
  • Can use backtest windows - so you can "predict" a past window and compare these predictions against actual results, in order to analyse prediction accuracy and to test different algorithms effectiveness and accuracy on your data
  • Uses probablistic forecasting at three different quantiles: 10%, 50% and 90%, allowing you to choose a forecast that better suits your business needs depending on whether the cost of capital (over forecasting) or missing customer orders (under forecasting) is of importance.
  • Automatically add related historic time series data to suppliment your forecast, such as weather (average monthly temperature, inches of rain etc.), average monthly exchange rates, public holidays
  • Do what if analysis by creating scenarios where you can simulate future events, such as:
    • What if a member leaves or closes a branch next month?
    • What if a member opens a new branch or a new members joins next month?
    • What if the pound weakens by 10% against the Euro next week?
    • What if we have an unseasonably long and cold winter?
    • What if the UK unemployment rate drops by 2% next quarter?

How much does it cost?

We're still working on that. Training vast sets of data using AI is currently very expensive. However, we plan to work on getting the costs down, so at present we estimate that we can deliver a free tier (for example, 10-20 forecasts for free so you can evaluate), but after that there will be a fixed cost to use Forecast.

When will it be available?

We're developing Forecast now, and we expect general availability to be Q2 2020.

What happens now?

We're excited to start building Forecast, but we'd love your input so we can make it as useful as possible. If you'd like to contribute, please get in touch so we ca discuss your ideas in more detail.