Discuter (0)

Imprimer

RSS

Home » Salescast » Here

Frequently Asked Questions (FAQ)

Table des Matières [Cacher/Visualiser] Frequently Asked Questions (FAQ)    Compatibility and integration    Trialing    Input data    Forecast report    Forecasts results    Lead time    Service level    Reorder point    Lead demand    Stock cover    Accuracy    Period    Horizon    Promotion forecasting    Slow movers exclusion    Cost    Executing a forecast    Security

Compatibility and integration

Salescast can be integrated with the vast majority of ERP solutions, order management systems and eCommerce platforms in use today. The only requirement is that your data are stored in a relational database. The established industry standard is SQL, if your system has a SQL backend you are good to go. If in doubt simply get in touch.

Accessing QuickBooks data requires the services of specialized 3rd party providers. We are currently switching providers, during this period we can not support new QuickBooks users. Please register your interest with us and we will inform you as soon as we are ready to provide this service again. We apologize for the delay!

The first step consists in creating an automated setup for the extraction and formatting of the entire sales history in your system. You have two options: You can do the integration yourself. This requires some IT skills and typically 1-3 days time. Please see the documentation. Alternatively we can handle the integration for you. In this case we build an "adapter" for your system. All you need to provide is a remote, read only access to your database. Please contact us for a quote.

Salescast is hosted on Windows Azure, which means we don't have a fixed IP address but rather an IP range. You therefore need to grant access to the IP range of the Microsoft data-center is 94.245.0.0/16. In case you want to grant access to a MS SQL server instance please make sure you read these instructions.

SQL is an established industry standard for storing business data. Nearly every single modern business app comes with a SQL backend nowadays. Your company probably uses SQL too. SQL comes with several flavors: MSSQL, MySQL, Oracle, ... Salescast supports all major SQL providers.

Database credentials are typically managed by the IT division of your company. If you're unsure about the process, don't hesitate to contact us.

Trialing

The first step in testing Salescast consists in setting up the automated extraction and formatting of the sales history. This phase is required no matter which inventory optimization/forecasting tool is being used. In our experience, producing a clean data extract represents more than 75% of the overall effort to test any inventory optimization/forecasting tool or approach.

Unfortunately this is not possible. While Excel is a handy tool for many things, it is not well suited for data management and data export from a technical perspective. This is precisely what databases such as SQL have been developed for. Please see our extensive explanation on why Excel as a data repository will not work.

We highly recommend a full scale test. Forecasting only a few products, or by providing only a few historic data points, will not yield meaningful results and time and effort will be wasted. Salescast's strength over traditional methods and forecasting toolkits comes from the analysis of patterns and correlations in the full product portfolio and the analysis of all available data. This advantage will otherwise be lost.

We propose to compare the accuracy of a Salescast forecast with the accuracy for your current forecast for the same time period using the absolute error. Please see our video on accuracy for an explanation how to execute it.

Input data

Salescast will use as much sales history as available. More data means more information which in general leads to better results in statistical forecasting. Ideal is therefore daily sales data and more than two years of history, but weekly or monthly data for a shorter time period will work too. Lead time and desired service level are required for the calculation of the reorder point. In case you currently do not hold this information in your system we recommend to start with average default values, which you can subsequently improve over time.

Quantity ordered is ideal as it is closest to the actual demand that has been observed. However, forecasts can also be produced with quantity shipped or similar information.

We recommend including all orders as received independently of their status in order to be closest to actual demand.

Both integration options intermediate schema or adapter exclude discontinued SKUs by default.

When using the intermediate schema for integration you can select the SKUs you wish to include in the report by introducing a ‘nullable boolean field’, please see our documentation. If you are integrating using an adapter, a similar function will be implemented.

By default Salescast will forecast a product as soon as there is a single data point. Forecasting based on very little data has its obvious limitations - very naïve models need to be applied and the accuracy will be low. As the sales history of the product grows the forecasting engine will increasingly be in a position to use more sophisticated models and the forecast accuracy will increase. Products with little or no history will benefit particularly from the introduction of Tags.

A stock-out cause zero sales until the product is back in stock. When forecasting, these artifacts should not be interpreted as 'zero demand', as this would decrease the accuracy of the forecast. Salescast is applying heuristics in order to identify stock-outs, and differentiate to the situation where there is 'true' zero demand. The majority of stock outs is thus captured automatically and no intervention is required.

Forecast report

Please see the sample report and the Salescast demo video.

Salescast will calculate forecasts, lead demand, reorder point, days inventory (stock cover) and can report the accuracy of the forecast. Please see the sample report.

Salescast can report extra columns. By default we propose to report product name, product identifier, stock on order and stock on hand. Please see the documentation when integration via documentation. When integrating via an adapter the fields to be reported need to be specified during the adapter creation.

Lead demand is a function of lead time, the reorder point is a function of the lead time and service level. In case this data is not available, the fields are left blank. Please check and fill in the missing inputs. When you next time refresh your forecast lead demand and reorder point will be calculated.

Salescast does not provide an order quantity calculation by default. However, implementing an algorithm that automatically calculates the reorder quantity is not a problem. Please read more on choosing the economic order quantity and download our free Excel EOQ calculator.

Please check that the report you use for verification is using the exact same definition of orders and dates. Examples include "shipment date" versus "order date" or "orders after cancellation and returns" vs. "initial customer orders". If you are integrating via our intermediate SQL schema please check the data you upload for potential filters and business rules that might be implemented on your side. In case the discrepancy still persists please contact us.

Salescast by default does not account for assembly SKUs. We recommend to implement a simple script that adjusts your sales data for the effect of assembly SKUs prior to upload into the intermediate SQL schema. Please contact us in case you require support.

By default Salescast will apply a slow mover’s exclusion to reduce Lokad subscription's cost for our customers. Items not sold a least 3x in the last three month when forecasting monthly, not sold at least 3x in the last three weeks when forecasting weekly, or not sold at least 10x over the last 15 days when forecasting daily will be excluded. An item can alternate between slow and non-slow as time goes, depending on its sales patterns. Hence, each execution might generate a slightly different set of items to be forecasted. Please note that in your Salescast account you can turn the slow mover's exclusion feature off.

The reported 12 data points are shown purely for convenience, the forecasting engine however will use the full available data history. Please check the second sheet in your report (‘details’) for a reporting of the oldest data point used by Salescast.

Forecasts results

A refresh can take from 5 minutes to over an hour for very large clients.

Trends play an important role in statistical forecasting. Growth and seasonality are good example of such trends. For this reason we spend a great deal of effort on developing seasonal models that capture them. To do this a trend must be observed and verified. Observing refers to the identification of a pattern as opposed to random fluctuations. Sales that increase before and drop after Christmas is such an example. The verification of a trend requires this pattern to be identifiable at least a second time. In our example this requires that a sales increase during Christmas the previous year can be observed.

In cases where no trends are observed, or verified, a flat forecast is almost always the most accurate forecast even though it often feels counter-intuitive. This might seem simple, yet the complexity lies in determining the right level of the flat forecast.

Typically two cases apply when a flat forecast is observed where a trend is expected: Too little data history makes it impossible to verify an observed trend, or there simply is no (observable and verifiable) pattern despite the intuition of the observer.

We recommend two steps in such a case: First, verify that sufficient data is given to allow a verification of a potential trend, i.e. two Christmas seasons in our example. Secondly, verify that a trend is actually observable. It helps to plot the time series in weekly or monthly aggregation.

This is a very common pattern, which is the result of business growth driven by the continuous addition of new products while the existing products are going through their life cycle. Given that only the existing products in the different stages of their life cycle are captured in the forecast (including decline and phase out) the aggregate forecast is down. Please read more on this phenomenon.

Salescast auto-detects calendar-based patterns, you therefore don't need to tell Lokad that a product is seasonal. Cornerstones of this feature are a large library of seasonal models and Lokad's multiple time series approach.

Salescast will apply seasonality effects to the forecast if they are identified and validated. A validation is given when a pattern repeats at least once or is present in similar products. When looking at the Salescast report it is easy to spot potential seasonality in the 12 historic data points which actually is not present in previous periods.

There are many patterns that happen once a year but not always at the same date. For example, Mother’s Day (which falls on different dates depending on the year and also varies between countries) and other holidays like Ramadan, Easter and Hanukkah (which fall on different dates depending on the year). Many of these events, such as Easter, vary only slightly in calender day, while the corresponding sales increase will happen over several days. In this case the irregularity is usually less relevant. For more extreme cases we use quasi-seasonal models.

Please see the second sheet of the Salecast excel report (called ‘details’). The date provided under ‘Sales start’ is the oldest data point observed in your historic data.

Please see the second sheet of the Salecast excel report (called ‘details’) for a list of the names and the number of tags that have been used.

We do not report the model that has been used for a specific product. However, for a general overview of which models we have in our library, and how the best model is selected, please watch this video.

Currently this information is not reported, however observing the forecast should give you a good idea whether a pattern such as seasonality has been identified and a seasonal model applied. Did you know that seasonality does not only occur as yearly seasons? Our algorithms are looking for a large number of possible seasonalities. Examples include weekly, monthly (‘paycheck’) and yearly seasonality.

Yes, you choose automatic refresh in the Salescast settings.

Salescast will produce forecasts and inventory optimization metrics that can be manually overwritten in the Excel report or in your replenishment or order management system. However, we strongly recommend to distinguish between the demand forecast and the replenishment optimization. An adjustment of the suggested reorder point to reflect business constraints should be part of the replenishment optimization and risk management. For the demand forecast historic performance is in most cases the best proxy for future performance, and manual overwrites should be treated as exceptions.

Lead time

The lead time refers to the delivery times of your suppliers (as opposed to your delivery time to your customers). It is defined as number of calendar days from purchase order to renewed availability in your warehouse. Please find more information on the definition and calculation of lead time.

Lead time is defined as number of calendar days. Working days would be a more accurate measure of lead time, albeit the impact on the forecast is rather small and accounting for national calendars worldwide is therefore not reasonable. Please read more on lead times.

Varying lead time from your suppliers can seem very confusing, as you can only work with a single value for lead time in Salescast. However it is fairly straightforward to determine the lead time you should use. Please read more on modeling varying lead times.

Not as part of the standard features. However we can implement such a feature for you, please ask us.

In such a case an extrapolation is used to determine the lead demand. This is never as good as a forecast, we therefore highly recommend to choose a forecast horizon that covers most of the lead times in the product portfolio (bar the outliers). Please read more on choosing the correct forecast horizon.

Service level

Service level is the desired probability of being able to fulfill a customer order. This value represents the performance goal in terms of stock availability. For example, a service level of 90% in practice means you will have sufficient stock on hand to fulfill 9 out of 10 orders. The service level allows you to define the balance between inventory cost and stock out cost. A high service level will increase your inventory and stock availability. A lower service level will lower your inventory, but also increase the chances of not being able to serve a customer.

The choice is highly business and product specific. Service levels can range from as high as 99% to less than 50%. You will find a simple, hands-on explanation on how to get started with a reasonable set of service levels for your product portfolio here. Please see our free service level Excel calculator and in-depth service level analysis.

The higher the service level, the higher the required safety stock to guarantee this level. In fact, the inventory level is rising exponentially with the service level. As a rule of thumb, the safety stock will double when halfing the service level’s distance to 100%. For example, when going from 95% to 97.5% service level, the safety stock will double. Please read more on choosing the right service level.

Yes, each SKU has its specific service level. Salescast will retrieve this information together with your order history from your system and report the values in the forecast report.

Yes you can, although we propose to rather make service level part of your inventory management policy. The inventory reduction and availability increase proposed by Lokad comes from two pieces of analysis: The forecast, and the reorder point (safety stock) calculation. We usually recommend to get started by assigning a free field (i.e. custom field) to service level and populate it with a uniform ‘average’ service level. Subsequently you can then refine the service levels on a product level. Please read here how this is done.

Reorder point

The reorder point is the number of units on hand that should trigger replenishment. When the stock on hand falls below the reorder point, a replenishment order should be launched. Intuitively, the reorder point is calculated to cover two things: First the stock you continue to sell during your lead time (your lead demand). Secondly the volatility in your demand (the safety stock). Therefore the reorder point = lead demand + safety stock. Read more on calculating the reorder point.

The reorder point is a function of forecasted demand, forecast accuracy and lead time. It is calculated as the sum of the lead demand (product expected to be sold during the lead time) and the safety stock (stock accounting for the variability in demand). The key is to calculate the variability of the demand to arrive at an optimized safety stock. It is intuitive that a highly predictable, stable selling product requires less safety stock than a more erratic selling, and therefore more difficult to forecast product. Read more on calculating the reorder point.

The reorder point should be considered independent from the forecasting period.

Yes, this is possible. Automation is a key benefit of Salescast, and by automatically updating the reorder point in your system a high level of integration is achieved in a technically rather simple way. Either you use the Lokad_Forecasts table when using the intermediate schema, or Lokad's engineers will build this into your adapter.

The reorder point is the sum of lead demand and safety stock, and safety stock is a function of the forecast accuracy. The more accurate the forecast, the less safety stock is required. Lokad Salescast therefore is using the calculated forecasting accuracy as an important input into the reorder point calculation. Check calculating safety stock with Excel for more details.

The reorder point will increase with lead time. The reorder point is the sum of lead demand and safety stock. Lead demand is the number of units that will be sold during the lead time; the longer your lead time, the more product you will sell.

The reorder point produced by Salescast will replace the Min-value. You can think of it as a rather ‘static’ Min value being replaced by a reorder point that is optimized each time the forecast is refreshed.

The Max value results from the calculation of the reorder point and economic order quantity.

Lead demand

Lead Demand is short for lead time demand and describes the number of units that will be sold during the lead time. This value is inferred from the demand forecasts combined with the lead time. For example, if you expect to sell 10 units in a month, and your lead time is 30 days, the lead demand is 10.

Stock cover

The number of days left before a stock-out if no reorder is made. This value is inferred from the demand forecasts and the stock on hand.

There is always a level of uncertainty regarding timing and quantity of the stock on order. Including stock on order would introduce additional uncertainty (=risk) to your inventory management.

Accuracy

The reported accuracy is the average accuracy that has been calculated by comparing forecast and actual sales. To do this Salescast goes back and forecasts and determines the accuracy along the sales history, moving forward month by month.

By default the accuracy of the forecast is not reported even though it is an integral part of the analysis Salescast performs. If you want to see the accuracy of the forecast for each product please turn on the feature in the Salescast settings.

This value is the expected accuracy of the forecasts delivered by Lokad. The higher the value, the more accurate the forecasts is expected to be. We have the following relationship: Accuracy = 1 - MAPE, where MAPE is the Mean Absolute Percentage Error. The expected accuracy is computed by Lokad - like the forecasts themselves - through advanced statistical analysis.

Mape stands for Mean Average Percentage Error. It is a commonly used measure of accuracy in statistical forecasting.

Often part of a trial is the quantification of accuracy improvement of Lokad's forecast versus the current forecast. This is done in a simple and efficient way by using the Absolute Error. We have published a video on how to assess the forecasting accuracy and how to compare forecasts.

Accuracy close to or at zero is reported in cases where the expected accuracy is very low and little confidence should be put in the forecasted results. Typically this occurs when the product history is very erratic. It is very common that in product portfolios of hundreds to thousands of products this occurs for a few products. Statistical forecasting needs to be taken with a grain of salt – it provides a huge accuracy and productivity gain for the bulk of the products, yet it hardly covers all cases.

Yes. The reorder point is the sum of lead demand and safety stock, and safety stock is a function of the forecast accuracy. The lower the forecast accuracy, the higher the required safety stock to guarantee the desired service level.

Period

The forecasting period can be day, week or month. For example, when choosing a weekly period, Salescast will provide you with a forecast of the weekly demand. In the settings you can choose and change the period.

We suggest choosing the period that fits best your operations. For example, if your company is passing orders on a weekly basis to suppliers, then weekly forecasts are most appropriate.

This behavior is caused by the slow mover’s exclusion feature of Salescast. Each period has its own exclusion rule and a change in period will most likely change the products being excluded. You can turn the slow mover’s exclusion on/off in the settings in Salescast.

As a natural effect of aggregation, the forecasting accuracy increases when switching to longer periods (and vice-versa, switching to shorter periods decreases the accuracy).

In theory a longer period will provide an increased accuracy as a natural effect of aggregation, and a higher accuracy will lead to a decrease in safety stock and therefore lower reorder point. In practice this effect is typically rather negligible.

You can choose only one period per forecast. In case you want to forecast two periods in parallel, for example daily 4 weeks ahead and weekly 26 weeks ahead we suggest to set up two forecasts by connecting your database twice using the wizard.

Salescast will forecast within calendar periods by default. Please contact us in case you require an adjustment to your fiscal calendar.

Horizon

The horizon should be large enough to cover 90% to 95% of your lead times. A horizon above this level will unnecessarily increase your cost, a horizon below this will lead to a degradation in the accuracy of the calculated reorder point. Read more on choosing the correct horizon.

Technically, there is no limit to the forecast horizon. Yet, the forecasting accuracy decreases the further into the future you forecast. As a rule of thumb, we suggest 0 to a maximum of 42 days for daily, 0 to a maximum of 52 weeks for weekly and 0 to a maximum of 36 month for monthly forecasts. In cases where both a low aggregation and long horizon are needed we suggest to combine two or three forecasts following the stated rules.

In case the chosen horizon is shorter than the lead time, Salescast uses a linear extrapolation of the forecasts to compute stock cover and reorder points. Needless to say, results produced through extrapolation are less accurate than those directly produced though our core forecasting technology. We therefore recommend to us a horizon that covers 90% to 95% of your lead times.

Promotion forecasting

Yes, Salescast can automatically forecast the effect of promotions on demand. To access this functionality, the events framework needs to be implemented. By submitting information on historic and scheduled future promotions, Salescast can analyse their effect on sales in the past and factor this into the forecast.

The more data and detail you can provide, the better the forecasts. In the ideal scenario, the promotion data includes promotion type (e.g. 2 for 1, 30% off), communication/advertising (e.g. page space in catalog or brochure), facing (e.g. head of the isle) and packaging. In practice this data is typically not available, or of questionable quality. However, even a limited information about promotion can still gradually improve accuracy.

The technical implementation is as straightforward as the setup of the initial forecast (please see the intermediate schema documentation). However, 'tuning' this feature can be more difficult. We therefore recommend implementing events only in a second step once the core forecast is up and running and with a small support/consulting package from us where we guide you to set up correctly. We offer advice on all aspects of the implementation, including which data to use, planning which data to collect and measuring the results. Please get in touch.You can start by reading the documentation on tags and events.

Slow movers exclusion

By default we exclude very infrequently selling products from your forecast since the forecast will add little value for you and increase the price of our service unnecessarily.

You can choose to include or exclude slow movers in the Salescast settings.

The default exclusion rules are: monthly forecasts - items not sold at least 3x over the last 3 months are excluded; weekly forecasts - items not sold at least 3x over the last 3 weeks are excluded; daily forecasts - items not sold at least 10x over the last 15 days are excluded.

Cost

You are billed on monthly consumption, which is recorded automatically. While you forecast throughout the month, your forecast consumption is recorded. At the end of the month, you monthly consumption will be priced based on our public pricing. You will receive an automatically generated invoice.

Once you are integrated, set up properly and satisfied with your initial tests we will start billing you on a monthly basis.

You only trigger consumption and therefore cost when clicking on the Refresh button in your Salescast account. Any other operation such as downloading reports does not incur cost. In case you have triggered a refresh erroneously don’t worry, you can cancel it any time before completion.

Please see our pricing calculator. For a detailed explanation please read understanding the pricing.

Please check the article on understanding the pricing for explanations and ideas on how to optimally tune your forecasts.

Executing a forecast

A refresh can take from 5 minutes to over an hour for very large clients.

You can choose manual or set dates for an automatic refresh in your Salescast account.

Security

Salescast is hosted on Windows Azure, the cloud computing plateform of Microsoft, which offers state of the art secure computing environment. Salecast itself uses industrial grade data encryption for all connections (SSL). Strict confidentiality policies of Lokad apply.

Sure. Just drop an email to contact@lokad.com with your NDA template and we will review it ASAP. If you don't have an NDA template, we can provide a draft.