IDG Contributor Network: 3 tips for getting started with cloud-of-things analytics
Cloud-of-things analytics has a simple, powerful appeal. It offers the opportunity to know more about your business, faster. For industries ranging from retail to manufacturing, that means better operational visibility, more responsive customer service, more automated reaction to problems, and improved preventative maintenance. That’s not to mention the potential to increase revenue by launching innovative services based on new insights or event-triggered actions.
But there are challenges to moving analytics into the cloud and building a meaningful framework. Each company has different skills, capabilities, experiences, and needs when it comes to analytics. Some are highly proficient and looking to operationalize their deep learning models, while others are still introducing more contextual data sources. Here are some tips that can help companies at any stage move to the next analytics level.
1. Just build the first frame
There is nothing more daunting to a writer than a blank sheet, to a painter than an empty canvas, or to a data scientist than an empty pipeline. You have in your mind’s eye the types of insights you are looking for, but the task of transforming the data, building and testing the model, creating visualizations, and then turning the output into action causes analytical block.
I advise companies to just start. Build a basic data frame on a relatively manageable and familiar dataset, process some basic statistics against it like counts and baselines, and create some simple views. Then start to layer in new data, add more complex analytics, and build out new visualizations. These will be the foundation of the initial solution. The aim is to keep it simple, flexible, and understandable. It builds confidence as the process starts to unveil the direction of the project.
I call this approach agile analytics. Instead of building a monolithic, complex model aiming to answer a specific set of questions, today’s data scientists and analytics managers should adopt an agile approach. This takes a page out of the agile software movement, building specific features in sprints and layering up the model, rather than meticulously planning out every step in a traditional waterfall plan. The reasoning goes that if it takes two years to build, test, and deploy an analytics pipeline, the data sources it uses and the questions it answers will be irrelevant by the time it enters production.
Agile analytics builds the equivalent of an analytics MVP (minimum viable product) to start getting the insights into the hands of the operations team. The data sources and context will get richer, the regressions more accurate, and visualizations easier to interpret, but the baseline data frame will be there. This helps build consensus internally, increase buy-in from lines of business, and get a faster time to value. So just start.
2. Avoid the lure of the point solution
Build or buy, on-prem or cloud, point solution or platform—these are just three of the crucial questions for software buyers today. When it comes to analytics (and granted I might be a little biased here), the last one is straightforward. Many wonderful point solutions in analytics solve specific issues with poster-worthy graphs. For some companies, these are ideal—the proverbial round analytical peg. But for the others, looking for answers to questions they haven’t even thought of yet, they are too limited. Most organizations are evolving so quickly they don’t know tomorrow’s questions. They need a flexible analytics bench that can adapt and evolve as they do.
The challenge here is that a platform still requires some in-house capabilities to make it useful, whereas a point solution can work out of the gate. That’s why I recommend the start-small approach. Marathon runners will tell you that the “couch to 5K” is the hardest part, but that’s where most runners stop, assuming the rest is equally challenging. It’s not. In fact, it’s easier and more rewarding. Have that marathon mindset as you go into the cloud-of-things analytics project, and don’t be lured by the shortcut of a point solution. It’s often a dead end.
3. Build on what you have
Make the systems you have smarter. Don’t just add a smart new system. Organizations have a huge investment in current systems, whether a CRM system, or a solution for incident management. Instead of building a new dashboard for those operators to use, or giving them a new task queue to manage, make their current systems smarter and timelier.
For example, take agents working on customer support in a contact center. They’re monitoring the queues with support tickets, trying to manage response times. Perhaps they have a few queues for different channels, such as phone, text, and social media, or for different product lines. In trying to improve response times to key customers, it’s best not to introduce a new VIP queue to watch, but to integrate the data about which customers to respond to first into the current support software. This avoids retraining, transition projects, and overhead. Instead, the agents get better data about which customers to prioritize, and the analytics stay in the background. Then, later, the analytics frame can start to match agents with tickets based on how likely the agent is to know the answer or how well the agent might get along with the customer, or eventually to eliminate the need for the interaction completely through automated self-care. Each iteration improves the service to the customer, enhances the efficiency of the contact center and avoids the forklift replacement of familiar systems.
By adopting these practical steps—start small, stay flexible, andimprove current systems—organizations can take advantage of the complex, ever-changing world of the cloud of things.
This article is published as part of the IDG Contributor Network. Want to Join?
Source: InfoWorld – Cloud Computing
Recent Comments