Home »
Salescast » Here
Looking at your first forecast report? Expect the unexpected
Unless you're already experienced in statistics, the first time you have an in-depth look at the numbers produced by Lokad, you will, most likely, be puzzled.
Statistics is a counter-intuitive science, which produces counter-intuitive results. Nevertheless, those results have a very real impact on your business. Let's try to shed some light on this first report.
Validating the data import
When getting your first report, the most important thing to do is to
validate the data used as inputs. In computer science, there is an old saying:
Garbarge In, Garbage Out. If historical data is not correctly imported, then there is little chance that forecasts will make sense. Hence, we suggest to pick about half a dozen lines in the Excel report and see if the historical data reported by Lokad matches the figures you're familiar with - even better if you already have a reporting system in place to perform the comparison.
Then, Salescast can also import inventory-specific values such as stock on hand, lead time, service level, etc. In particular, the lead time and the service level values have a strong impact on the final reorder point suggestion (when available). If reorder point seems way too low,
check that the lead time is correct; same goes for the service level. In case of doubts, don't hesitate to question the Lokad team, we are here to help.
Generating and interpreting your report
We recommend you review the
sample report which explains how to interpret all the columns in your report. In case you are not familiar with the Salescast web application yet please watch the
demo video which explains how to
generate and access your reports.
Choosing your forecasting period and horizon
Reports produced by Salescast have
two major settings named
Period and
Horizon. Those settings can be configured from the web user interface. Go to
Your Forecasts » Settings. By default, Salescast produces 6-months ahead forecasts. Here, we discuss how to choose the value for those settings.
Salescast supports 3 distinct periods namely
day,
week and
month. The period represents the aggregation level of your data. If you choose
week for the period, you will get weekly forecasts, i.e. one point per week, starting where the data ends. Then, the horizon is a number representing how many forecasting points should be produced. For example, if
Period=week and
Horizon=3, then a 3 weeks ahead forecast will be produced. The maximal value for the horizon is 100, but opting for high horizon values is frequently a poor idea (this aspect will be discussed further below).
The Period
We suggest to
chose the period that fits best your operational requirements. For example, if your company is passing orders on a weekly basis to its suppliers, then weekly forecasts are more appropriate. As a natural effect of aggregation, the forecasting accuracy increases when switching to longer periods (and vice-versa, switching to shorter periods decreases accuracy).
The Horizon
The horizon should chosen so that it is
large enough to cover to your lead-times but not larger. Indeed, Salescast is
priced by the number of forecast points generated for your reports: the larger the horizon, the more you get charged. On the other hand, if forecasts are shorter than lead times, Salescast uses a
linear interpolation of the forecasts to compute the stock cover and the reorder points. Needless to say, results produced through interpolation are less accurate than those directly produced though our core forecasting technology.
Then, Salescast does not enable to tune a distinct horizon value for each item (*), despite the fact that items are likely not to have the same lead time. Yet, thanks to the interpolation behavior, you are not forced into using the MAX lead time as forecasting horizon for your report. Instead, we suggest to take
a horizon large enough so that it covers 90% or 95% of your lead times. We believe Lokad to be low cost, and micro-optimizing further the forecasting horizon is typically a waste of time.
Rule of thumb: history 5 times longer than forecasts
Technically, Salescast can forecast 100 days ahead out of a single day of historical data. Yet, the accuracy will be, most likely, abysmal. As a rule of thumb, we suggest not to expect too much from a statistical forecast if the depth of the historical data is not at least 5 times the length of the requested forecast.
Rule of thumb: weekly forecasts beyond one month
Daily forecasts are rarely needed further than 28 days ahead (i.e. 4 weeks). Indeed, we have observed that companies that try to leverage daily forecasts looking far ahead have a tendency to
micro-optimize their activity in a way that is usually not profitable. It's typically more sensible to preserve some flexibility for last minute adjustments when looking one month ahead. In practice, while daily forecasts are handy for very short term inventory optimization, it's a better practice to switch to weekly forecasts beyond one month.
In order to benefit from both daily and weekly forecasts, we suggest to have
2 distinct reports in Salescast:
- one report with daily forecasts, updated every working day
- one report with weekly forecast, updated once a week.
Rule of thumb: monthly forecasts beyond one year
Similarly, weekly forecasts are rarely justified for forecasts beyond 1 year ahead. When looking far in the future, weekly statistical forecasts give a false sense of precision. Monthly forecasts should be preferred.
Setting the right service level
Service level is an important input for the calculation of the optimal reorder point. We recommend you
make yourself familiar with the concept and how to choose the right service level for your products.
Thinking the forecast accuracy
Just before the forecasts there lies an
accuracy column. Lokad delivers
not only forecasts, but also anticipated forecast error rate as well. You can think of this feature as of a self-diagnosis of the system. Without getting into technical details of actual definition of this accuracy indicator, let's say it's a percentage: 100% being a totally accurate forecast and 0% being a totally inaccurate forecast.
100% accuracy is not a realistic expectation for a forecasting process (statistical or otherwise). In particular, when looking at low sales volume, a small error of +1 or -1 can already reduce the accuracy
expressed as a percentage to a value lower than 50%. Contrary to intuition,
the overall level of accuracy is not a consequence of the forecasting method, but of the level of aggregation of the data itself.
For example, if we are forecasting daily nationwide electricity consumption from one day to the next, a forecast with a 99.5% accuracy might be considered as rather poor, while forecasting the promotional sales of a fresh food product, a 30% accuracy might be considered as a significant achievement. Yet, this does not mean that a
better forecasting solution cannot improve the situation...
When it comes to forecasting, there is no such thing as
good or
bad forecast in absolute terms,
the only thing that matters is how good are those forecasts compared to the status quo or alternatives?. You can't just say
those forecasts are too inaccurate, let's give up on forecasting because inventory level implicitly represents a demand forecast - which is likely to be
even worse anyway.
If you have inventory, you are already forced to produce forecasts. The question is:
are those implicit forecasts better than the explicit (statistical) ones or not?
Odd looking yet normal patterns in forecasts
In your report, you will probably observe
flat forecasts, i.e. perfectly steady sales over several weeks or months. We know that it's
very unlikely that such an event will actually happen, yet, keeping in mind that the goal is to keep
inaccuracies as low as possible, a flat forecast is frequently the
most accurate statistical option, especially when historical data are highly volatile. More
about flat forecasts.
Also, you might also observe some
data points that look odd: too high or too low, while the historical sales look reasonably flat. There are many distinct reasons that explain such situations. In particular,
data aggregation is a lossy process and sometime, information is lost when looking at weekly or monthly data. Since Lokad leverages the most fine-grained data available (ex: daily data if available), Lokad can capture patterns that are
not visible in the aggregated visualization.
Then, if you start to sum all forecasts, just to see if overall figures make sense, you're likely to get more odd results. For example, your
business might be growing while individual product sales are decreasing. In short, Lokad takes into account life-cycle patterns which are typically hard to grasp, because you need to think of the constant of products getting in and out of the market.
Requesting a review of your first report
Once you have successfully integrated and produced your first forecast report with Salescast,
we recommend to review your first report with a member of the Lokad team. It will help you to make sense of the numbers produced, give you guidance on how to work best with Salescast and explain what might at first be puzzling to you.
To get the most of the review, please review the information on this page, your forecast report and send us your most important questions upfront. A member of our team will contact you and schedule a call.
Addressing question about your forecasts
In case you have
questions regarding your forecasts at any time we can discuss if a
subset of at least 100 time-series can be isolated AND
if an alternative forecasting method does outperform Lokad. With fewer time-series, the noise level is typically too high to draw any conclusion with a high statistical confidence. Without an alternative that outperforms Lokad, the discussion is very likely to not produce much result, simply because there is no proof that this particular aspect can be improved by anyone.
However, we must advise that
we cannot answer questions about individual forecasts unless you opt for an individual support package.
Integration packages or fees do not cover a discussion about the statistical relevance of the delivered forecasts. As much as we are interested in providing you with the support you need, there are a number of reasons for this:
- Time: Analyzing in detail what is happening on a specific time-series is time consuming and can easily take one hour. A typical account holds hundreds of series, engaging in a discussion with the client on individual series is often very time consuming for both parties.
- Productivity: In our experience, discussions on individual forecasts yield frequently no practical results. The forecasts, good or bad, are the result of a complex and powerful technology which we do not tweak with regards to individual results.
- Statistical relevance: No forecast comes in isolation, tweaking the result of one time-series (i.e. product) impacts the result of other time-series. The improvement that would be brought to the time-series in question would most likely degrade the overall performance.
Remarks about forecasts, as well intended as they usually are, are by definition impossible to validate (unless we learn to look into the future..). Rule-based suggestions (i.e. if product is x then do y to the forecast) will with a very high probability not improve the situation since Lokad routinely outperforms (on average) all rule-based ideas.
Forecasting accuracy is the №1 priority for us, and we will continue doing our best to constantly improve our technology.
Instead, we suggest
you assess the overall accuracy of the forecast, ideally by comparing it to your previous forecasting method. It is not difficult (provided that you have an alternative forecast) and will give you a sense of the accuracy. How this is done can be seen in this
short tutorial video on assessing and comparing forecast accuracy.
Conclusion
When quantitative forecasts are concerned, we firmly believe that you can't improve what you don't measure. If nothing else, Lokad offers you the opportunity to benchmark the
status quo in your company. By doing so, you're likely to be puzzled by the Lokad numbers the first time, yet
we are committed in having a member of the Lokad team reviewing those numbers with you. Our experience indicates that within less than 1h, those numbers will start to make a lot more sense.
Email us anytime to schedule a call.