Artificial intelligence (AI), machine learning (ML), and advanced analytics have become hot buzzwords in the business world. But for many business analysts and business owners they are nothing more than that – just words.
By Leandra Webb-Ray, data scientist at Decision Inc
Many companies feel an urgent need to apply these concepts but are at a loss as to how and where to implement them, and what true value can be gained.
Sadly, this leads to many rushed projects where methodologies are applied incorrectly or to the wrong use cases, simply for the sake of saying that AI is being used. This only works to harm the reputation of AI, and make users more resistant.
Many professionals are also hesitant to use analytics because they fear that it will make their skills obsolete, instead of looking at how it can free up their time to spend it on more critical and analytical tasks.
Sufficient time and skills need to be dedicated to a successful implementation of analytics, and a phased approach is needed to ensure that the outcomes are accurate and adopted by all key stakeholders.
Start with the Key Question
So, what is the starting point? For most companies, this is the key question that comes up daily.
In retail it could be: ‘How can I plan for future customer demands?’, ‘Which products should I promote?’ or ‘Where are my biggest losses?’
In procurement: ‘How can I save money while still purchasing the same products?’ or ‘Are my vendors giving me a consistent price?’
Implement a Proof of Concept (POC)
Once a key question has been identified, the next step is the Proof of Concept (POC). Instead of putting together a hurried solution and using it for potentially flawed decision making, a POC allows for a prototype to be applied and tested on a subset of the data, and for all stakeholders to become comfortable with the proposed solution.
The question: ‘How can I plan for future customer demands?’ is a perfect use case for applying a POC of afForecasting algorithm. After analysing a subset of sales data and finding the most appropriate forecasting methodology, the new sales figures of that subset can be monitored for a trial period and compared to the forecast, and the methodology can be adapted if needed. When these figures match closely (within reason), the case for AI has proven itself – building trust in the concept.
To answer the question ‘Are my vendors giving me a consistent price?’, a regression algorithm can be used to determine which factors drive the price of a product the most (such as time of year or specific vendor) using another POC. This will give business owners the data to make key purchasing decisions and support the integration of analytics tools.
Deploying an Analytics Strategy
Deployment of analytics is ideal once the POC has created an appetite for analytics within the business.
This involves three key elements: people; process; and technology stack.
People: The key to user adoption
As with most new concepts, analytics will only be successful if there is a key user to drive the process. It is important to identify a champion early and help them take ownership. A good candidate generally has an interest in the analytics field, as well as a strong motivation – such as their own key question that needs to be solved.
This champion, together with their team, would then be enabled to take the outputs of the analytics and ensure that the relevant actions are taken, and their effects measured – ensuring that the true value of the analytics is realised.
Process: Driving solution entrenchment in the business
Early implementation of a solid process in an analytics deployment will ensure long-term success. Without defined steps and assigned responsibilities, this valuable solution could get lost in the background of pressing day-to-day tasks.
Important first steps include: defining a business user who will be responsible for data extraction (if the solution does not interface directly with the database), who or what will initiate the analytics processes and when will it happen, who will check the output and take relevant actions, and how will success be measured.
Technology Stack: Solutions that are intuitive and lead to self-service capabilities
Lastly, appropriate technology should be chosen for each step of the process. Various considerations such as cost, ease-of-use, capabilities of analytical engines, and what technologies are already familiar to the business should be considered.
It is important that all business users are enabled to use the relevant technologies and feel comfortable doing so. Support also needs to be readily available for any areas that cannot be supported by internal teams.
Whatever the industry, the key is to start by addressing a critical pain point using analytics. Start small to allow the concepts to be implemented effectively, become adopted well by teams, and then scale up from there.