Science fact: “Model Factory” means getting the basics right.

0

I press a button, a miracle machine churns through all the calculations in the world and the answer to the Ultimate Question of Life, the Universe, and Everything[1] is produced as a single number. Oh hang on, that’s 42. Alright, for our microcosm, let’s stick to the answer to my customer’s behaviour, my products’ demand, when the widgets of my machines are going to fail or how my customer feels. Too much to ask?

Once upon a time this was science fiction. Not anymore.

As people across all parts of the organisation jump onto the potential of data, we naturally start asking more, bigger and harder questions. And we want the answers now. Minority Report[2] may be in the near future but for today, how do we just keep up with everyday demand and still have time to think of new questions to ask?

We may need our own miracle machine to give us space and time back.

Model FactoryIn practice, we want something that works in the background to analyse data and derive predictions, with little requirement to change or interact, but at the same time is robust and trustworthy so that the results and justifiable and make logical sense. Our machine for producing these predictions, a “model factory”, should automate, accelerate and maintain governance over a series of logical processes across the Analytics Lifecycle from data preparation to exploration to model development to deployment & monitoring.

This comes down to a machine that has well-oiled technology, has clearly defined logic and is up-to-date and maintained. Building and running this machine will take 8 Ps: people, process and product (technology), possibilities and the old saying “preparation prevents poor performance”. Being prepared just means getting the basics right.

Possibilities: look beyond the "safe zone" for goals to strive towards.

  • Engage people (internal and external) who have done it before to develop on ideas and plan a realistic roadmap.
  • Keep an ear out for new trends in process improvements and analytical techniques e.g. conferences, association meetings, publications.
  • Allow dedicated time and/or resources to experiment with existing and new data sources to learn dynamically and determine the next innovation.

People: create a culture to attract and retain the right people to create, maintain, update and interpret the machine.

  • Give people direction and guidelines but room to create and innovate e.g. by providing separate processes and technology environments.
  • Create a team of people with various skill sets – business, domain, technical, unicorns – but who speak a common language and have a common goal.
  • Keep the day job interesting with side projects e.g. enablement, secondment, research.

Process: implement, enforce and reinforce processes which improve productivity and question others.

  • Automate standard reports and make the others, as much as possible, self-service e.g. use interactive visualisation and Microsoft Office add-ins.
  • Give access to the right information to the people who need it e.g. common intranet site, locked down operational vs. dynamic discovery.
  • Document stages of the process in standard templates for reusability and governance.

Technology: match the right technology for the task at hand and leverage modern infrastructure.

  • Project objectives, user skills, time constraints and the format for consumption will dictate the technologies required to solve tasks e.g. exploration, operational, experimentation, integration with front-line interfaces.
  • Provide high-performance technology – machine learning, multi-threaded, in-database, in-memory, template-driven, workload-managed – to accelerate the cogs of the machine.
  • Integrate each stage of the Analytics Lifecycle with common metadata to improve seamlessness.

These basics are the first principles to a big bang “model factory” – everything else will fall into place. This machine is a living entity that will evolve over time as technology advances, analysis trends change and objectives are redirected, and in this way will need to be kept up to date and modern. But by getting the basics right, the machine, at whatever stage of its and your organisation’s evolution, will form the foundation for your future intergalactic purposes.

May your “model factory” not be in a galaxy (too) far, far away.

Learn more about Machine Learning, High-Performance Analytics, Unicorns (the people kind) and try SAS Visual Statistics at the links. For those in Australia and New Zealand, keep an eye out for webinars, Hands-on Workshops and conferences throughout 2015.

[1] The Hitchhiker's Guide to the Galaxy, Douglas Adams (1979)

[2] Minority Report, Twentieth Century Fox Film Corporation and DreamWorks SKG, Dir: Steven Spielberg (2002)

Share

About Author


Business Solutions Manager

Annelies believes that there is potential for Analytics everywhere. She works as an evangelist, enabler and execution strategist to empower individuals and organizations with good Analytics practices. Annelies is the Advanced Analytics technology lead at SAS Australia and New Zealand responsible for product management and enablement. During her career, Annelies has held various positions supporting the customer lifecycle from strategy and requirements to implementation and adoption. This experience gives her a practical view of the end-to-end process of data analysis across government and industry, including engagements in several customer analytics, demand forecasting, text analysis and allocations optimization projects. Annelies is the current co-chair of the NSW Chapter of the Institute of Analytics Professionals of Australia, the largest analytics community in Australia. She has a research Masters in Mathematical Statistics and guest lectures at several tertiary institutes in Australia.

Back to Top