How to ML-fy Your Small Business

ML is everywhere and those who do not know it, it is Machine Learning. The Machine Learning is out of the labs and being democratized by technologies. So, it can be used by all and sundry and it is not costly too. It is available in the micro-fraction of the cost it was used to be. Objective of this article is to understand how a small business can reap the benefits of machine learning and AI without hiccups. 

Before delving into machine learning and AI for small business, we must also know why a small business should even care about it? Can't it live without it. Yes, definitely a business can survive and be profitable without AI. It is a matter of competitive advantages if a company wants to have. AI will make small businesses what online eCommerce has done to the marketing function of small businesses. The eCommerce has uplifted the marketing and sales performance of hundreds and thousands of small businesses through pan-country reach and delivery systems. Now, any small business can take rest as far as marketing is concerned because eCommerce companies can take your products to consumers, they do marketing, they convince customers, they run sales promotion schemes....... almost everything, isn't it. Let us take an instance, what can you do if your competitor are also on same platform, selling same stuff, at same or lower price than you? Can you lower the price by capital financing the discounts, or delay the break-even or move out of the market. All these options are very scary. 

Here, ML-fy your business and this is what you can do as a logical step to sustain, survive or grow your business in fierce competitions. ML-fying your business means you delve into data generated in your small businesses and explore the opportunities of maximization and minimization. With the information of maximization and minimization opportunities, small business can foresee future business environment and internal performance. If that is known, you can have better planning, better approach to deal with competition. Over and above, your customer will feel the difference which means more customer loyalty and more referrals.

Now, we have got feeling of how ML-fication of small business can benefits in terms of minization and maximization opportunities.Following are the broad steps in implementing machine learning into business--

Let us understand this process of machine learning implementation with an example. Suppose there a sweet manufacturing company-- Sheetal Sweets. This company manufactures sweet "RASGULLA". It is a family owned business. This company is facing a problem that she has issues of either stockout situation or over-inventory of raw materials especially sugar and it is very frequent. Stockout situation of sugar leads to halt in manufacturing process and delay in sales order fulfillment. Oversupply leads to blocking of working capital which means interest has to be paid to bank on overdraft facility.

Let us dive deeper into the problem. Suppose working capital employed in business is Rs.30 lac. Bank charges interest rate @ 12% p.a. If working capital is paid in 6 months on an average, the interest burden on company is Rs.180000 for six month. On yearly basis, it is Rs.360000. Even if it is not borrowed from bank, we can assume the opportunity cost of employing the working capital. If Rs.30 lac amount is available with Sheetal Sweets, it should earn at least 12% return on business operations because bank might be offering same interest rates on bank deposits. Let us assume that working capital is completely absorbed in RASGULLA manufacturing.

The stockouts and over inventory adversely impact the working capital management. In case of stockouts, Sheetal Sweets may loose customer or sales and if there is over-inventory, the working capital rotation is delayed. So ultimately, either way it is harmful. Thus, Sheetal Sweets needs the predictability capability so that it can plan its inventory in optimal manner that minimizes the stockouts and over-inventory situations.

PREDICTABILITY CAPABILITY DEVELOPMENT

For predictability capability can be built through various measures but our focus in this article is to use machine learning for predicting the next day sales, next week sales, next fortnight sales and next month sales. There are range of models in machine learning that can predict the sales in coming periods but we will focus only on Time Series method. In time series method of forecasting, the variable data like sales is available for past periods and with suitable model, we try to predict the sales for coming periods.

                                   Picture 2 Process of Forecasting Model Development

The picture 2 shows the development process of forecasting model. Sheetal Sweets needs to collect its sales data for past periods. It is better to have daily sales data of at least 3 years or more. After collecting data, it should be stored into nice format in MS Excel. In Excel file, it sales data should be filled in cells corresponding to the periods they belong to.

Now, a small programming code in R can be used to build a forecasting model. The forecasting model programming in R can be asked separately via dropping the requirement in comment box. This model should be reviewed time to time to see how close the predicted values of model are to the actual sales realization. The lesser the gap, the better the model is.

Self Learning Technologies for eCommerce

The eCommerce business cannot survive without incorporation of Self Learning Technologies (SLTs) due to changes in scope of decision making from years and months to hours and minutes. Self Learning Technologies can be adopted with budget as minimum as $10000 and budget as big as $1000000. 

There are various ways to implement SLTs from as simple as hiring a freelancer on temporary engagement basis, crowdsourcing, renting the Machine Learning-as-a-Service (MLaaS) and as complex as buying SLT solution, ML cloud services and in-house SLT development. The allocation of human resources depends on approach of SLT implementation chosen. For simple SLT implementation, a data scientist and digital marketer may suffice but for complex SLT implementation, a dedicated team of data engineers, data scientists, business analytics professional and digital marketers should be deployed. 

The SLT deployment may take three months to one year or more which again depends on what kind of SLTs are being implemented. Simple and intermediate SLTs implementation may take three months to nine months and complex SLTs implementation may take twelve months to eighteen months. The decision of SLT implementation is very critical. 

For simple to intermediate SLT implementation, crowdsourcing and MLaaS are better options and for complex SLT implementation, renting MLaaS to buying SLT Solutions from Microsoft, Google, IBM etc. or build SLT solution in-house are suggested. If company wants to keep the SLT intellectual property confidential, limited to itself and to do full blown business transformation; it should go build SLT solution in-house or if company wants to deploy SLT implementation faster and battle tested SLT solution without actually owning it; the company can choose buy SLT solutions from reputed vendors.    

For more detail on how to implement self learning technologies (SLTs) in eCommerce, please click link below---

https://imojo.in/e9s5oc