Vous êtes sur la page 1sur 3

blogs.sas.

com

7 tips for operationalizing analytics


Randy Guard

4-5 minutes

One of the biggest challenges we’ve been hearing from customers


lately is that they need help operationalizing analytics to extend the
value of their modeling efforts. In many cases, they’ve hired smart data
analysts who can transform data and create sophisticated models, but
their work is used for a single purpose and then it sits on a hard drive
somewhere.

It’s difficult for most organizations to figure out how to move models
from a single machine into widespread operational use. What steps
can they take to make sure this critical work is used widely to
transform the organization?

Specifically, I’ve heard customers ask questions like:

How can we bring this model into our ERP system?


How can we bring analytics into our production control systems?
How can we embed advanced analytics into our customer
intelligence systems?
How can we share these models with other teams who could
benefit from similar analyses?

The answers are multifaceted, involving culture change, technology


adoption and process adjustments – but the results can be truly
transformational. If you can embed analytics into your operational
systems, you can automate more decisions and further spread the use
of data.

For example, GE Transportation has embedded analytics into its


EdgeLINC application, which collects locomotive data while a train is
traveling. Using event stream processing from SAS, the data can be
analyzed in near real time to automate decisions for individual trains –
and to make broader decisions about the entire fleet.
So instead of using that IoT data for a single command-and-control
purpose, the data also augments the work of other systems to improve
operational efficiencies, optimize crew performance and to consolidate
the management of applications across the fleet. How can you
operationalize your models and achieve similar results? I recommend
these seven tips:

1. Know your data


Deploying and operationalizing analytics more broadly requires that
you have a good handle on your data, where it comes from and how it’s
organized. This step is essential so you can be confident you’re making
decisions on the most relevant data.

2. Use a flexible modeling environment


It should be easy to share and collaborate on modeling efforts in an
environment that is flexible, fluid and creates a sense of community
development.

3. Apply governance and model management


When working across systems, be prepared to integrate disparate
code, processes and information into one hub that provides consistent
delivery of information – regardless of its source or destination.

4. Partner with IT and the business


IT operations, business units and analytics teams need to work
together at all levels of the organization. In particular, your chief
information officer, chief analytics officer and business leads should
be meeting to establish shared goals.

5. Work towards zero latency


Data from sensors and IoT devices is everywhere. If you can use APIs
and event stream processing to bring analytics to the data, you can
automate decisions on the spot, as the data is created. When latency is
reduced, the speed to decision is improved.

6. Learn new modeling techniques


The increased use of deep neural networks provides more accurate
decisions – and more applications of analytics than ever before.
Computer vision, rare-event modeling and natural language processing
are just a few of the latest advances.

7. Establish a heterogeneous IT environment


Your data scientists need the ability to work with multiple tools and
code bases – and to deploy those results in multiple ways, through
APIs, voice-to-text and other technologies that move far beyond
standard reporting. You should be able to consume analytics from
anywhere and deliver answers to anyone or anything.

As these tips indicate, operationalizing analytics involves the entire


analytics life cycle – from data to discovery to deployment. If data
scientists can easily share data, ideas and models between teams and
operational systems, the results of their work can be multiplied
exponentially.

Learn How to Go From Data to Decisions as Quickly as Possible

Vous aimerez peut-être aussi