The problem of scattered planning
Often the following situation develops in companies: Each division has its own, unique version of the sales plan. Such plans are used in work, for example, by the departments of marketing, sales, financiers and logistics.
These plans have different formats, varying degrees of detail, and, most importantly, different and conflicting figures.
A logical question arises, how to build an integrated planning system in the company and what is needed for this.
Building a business process
I think it is important to approach the issue from the perspective of creating a streamlined business technology .
As a rule, planning is a regular process (often monthly or weekly), in which there is a coordination and adjustment of the sales plan and related plans (for example, supply and production).
(Often use the terms: S&OP - Sales and Operations Planning, IBP - Integrated Business Planning).
In the planning process, participants and their roles, specific tasks and dates must be clearly defined. For example, sellers provide customer (or channel) plans. Marketing checks the assortment and informs about new products, etc.
For the planning process and its participants, KPIs should be determined and reports developed, according to which it will become possible to control the quality of the results. For example, data completeness, planning accuracy, inventory turnover, and service level.
Organizational Challenges
Discipline of participants
Planning requires the involvement of different employees of the company, as well as providing them with quality data on time. (The IT system can partially compensate for these problems by using automatic calculations.)
Correctness and completeness of directories (data wizard)
It is necessary to ensure timely updating of directories in the accounting system. For example, for a product, the current status, sales start / end dates, category and other fields that will be used in planning and analysis should be determined.
“Top-down” adjustments
When coordinating plans at the top level, inevitably, top-down corrections can automatically occur. In this case, the responsibility for planning is eroded by the performers, as the numbers were “adjusted above.”
In any case, it is necessary to establish tracking / audit of edits and planning versions.
High degree of uncertainty
Changes in the market and the actions of competitors can nullify all planning efforts. It will be useful to introduce a method of comparison with the “Naive forecast”. Those. for example, how much better is the result of the process than a simple moving average or another simple forecasting method. (Unfortunately, in practice it may turn out that the naive forecast is comparable in quality with the result of the process).
Analytical Data Warehouse
Now it is difficult to find a company that does not have its own system of analytical reporting and a single repository of analytical data.
Nevertheless, such a system is a prerequisite for building a planning system.
Statistics of actual sales, prices, additional external analytics, Warehouse supplies, Turnover, Goods in transit - all this is necessary both for the preparation of a sales plan, and for its subsequent analysis.
Therefore, it is possible that before building a planning system, you will have to build a data warehouse and a business intelligence system.
There are many approaches and solutions, but I want to dwell on a few key points:
Data quality
Because The data warehouse is a separate system, then I think the differences in numbers with the main accounting systems from which data is loaded into the warehouse are inevitable. A significant part of the effort can be spent on cleaning, checking and rechecking downloaded data. So that this does not come as a surprise to management, it is worthwhile to put these tasks into the plan / budget of the project.
Data visualization (dashboards)
In general, dashboards are useful for internal marketing of a project and for its effective presentation to company management. However, a significant disadvantage is the rather high cost of their creation and lack of flexibility in configuration on the side of the end user. In fact (in my opinion), a dashboard is more likely an IT product and most even advanced users are not ready to master any data visualization system other than Excel.
Performance
Performance can be a big problem, which will greatly affect the attitude of users to the system and their willingness to work with it. A good way to improve performance is to use OLAP technology, while minimizing the number of on-the-fly calculations.
Machine learning
Of course, this topic is a key “hype” and there is a lot of advertising information around it.
Let's see what machine learning can give us in practice and what we are faced with.
In my opinion, machine learning in the planning sphere, as a rule, does not provide higher accuracy than manual planning (although perhaps this is only a matter of time).
An important benefit from its implementation is the simplification of routine operations, especially for goods that are classified according to ABC classification to B and C.
Significant gains can be achieved if the planning process requires a large degree of detail and the volume of combinations of goods / channels / shops / periods, etc. in the millions of records.
Now about the difficulties:
90% of the effort is spent not on building an algorithm, but on cleaning and preparing data
As in the matter of building business analytics, the data supplied to the input of the machine algorithm must be verified and converted into “features” (or predictors). In my opinion, at this stage, the risk of both logical errors and bugs is highest. You can deal with the problem by visualizing and checking data at intermediate stages.
Result and cost of work is difficult to predict
In my opinion, this is the biggest problem. The construction of forecasting algorithms is a process inherently close to scientific research. It is easy to make it endless and there is a high risk of failure with low quality forecasts. The reason is the infinite number of predictor and model options that you can try to improve the quality of the forecast.
Business distance
In data science projects, in my opinion, there is a high risk of business users not understanding the language that data science experts speak.
For teamwork, it is important to be able to convey in simple words the results and progress of the work. Avoid mathematical and other complex terms, interpret the results of models from the standpoint of common sense.
To reduce risks and increase the manageability of a Data science project, Agile project management technologies are well suited.
The iterative approach, the frequent demonstration of the results to the customer and the launch of the “minimally useful” parts of the solution into the product is essential.