Monday, February 27, 2012

Analytics as a Strategic Capability

Analytics. The next frontier. First personal computing allowed us to work faster and more efficiently. Then web 1.0 brought an environment that allowed us to work with more flexibility. Web 2.0 made us social and gave us information about who we are, how we behave and what we like. Now it is time for analytics to bring it all together. The efficiency, the flexibility, the information - coming together to generate insights.

While the quantitative tools and techniques that are employed in analytics are nothing new, the convergence of personal computing, the web and social data position analytics to now be a truly strategic capability. The missing piece is organization that leverages each of these components into a cohesive package that delivers the right results to the right people at the right time with the right perspective. This is more than BI, more than a technology. It is a new way of doing business that recognizes the value of information, BUT ALSO is strategic about what problem it is trying to solve, and how that solution is delivered. All this takes people power and smarts that can't come off the shelf, and that understands business strategy, not just quantitative techniques and IT.

The reason why the business focus and perspective is SO important is because with out it, analytics can come up with a million answers, a million solutions, none of which address the real business problem at hand. Many analytics tools on the market make this very hard as they make it very EASY to thrash about in ones data, and to lose focus on why you were knee deep (now neck deep) in it in the first place. It is this axiom that drives the truth that business analytics be driven by business STRATEGY and those that define it. Business analytics should not be driven by technology - in fact technology should be the last consideration. This is not to say that technologists should not be included in the conversation, but they should certainly not be leading it.

So who is this uber businessperson that can lead the analytics of an organization? It is someone that understands technology, quantitative techniques AND business strategy. They are strategic partners with those that set business direction, but yet have the skills to guide or self execute technology and quantitative needs. They must elevate themselves above "number cruncher" status if analytics are to truly shape business strategy. They must see themselves at a business leader more so than a doer. It is this type of leader that can see the big picture and know how analytics fits in that will truly position analytics as a strategic capability within any organization.

Tuesday, January 3, 2012

The Data and Analytics Catch-22

It goes without saying that without data, analytics can't exist.  As such, most organizations invest in data collection systems (ERP, CRM, HRIS, etc), and then get into the business of analyzing the data collected.

Although the main role of these systems is to manage and automate operational business functions, the real value of the data collected is with the analytic insights that they generate.  The problem is that oftentimes analytic needs don't play a major role in these enterprise system implementations.  Compounding this issue is that there is a lag between the time when a system is implemented, and when data actually become available (collected) to analyze.  The result of all this is that analytics can become an afterthought or not practiced at all.

The alternative to this approach is to lead the implementation of a data collection system with analytics.  But how the heck are you supposed to analyze data without a data collection system to collect data?  This is the catch-22.

The reality is that prior to ERP, CRM, HRIS, etc implementations, most organizations are capturing related data in some way (be it spreadsheets, local databases, legacy systems or *gasp* paper records).  These "data systems" while perhaps not elegant, still contain the potential for analytic insights.  What's more is that if analytics, driven by business goals and processes, are practiced prior to enterprise system implementations, then they can sort out what data are really strategic or business critical in addition to operational or regulatory requirements.

The misperception is that big, robust data systems need to be in place before any analysis can happen.  This is just not true.

The message here is that a practice of analytics with available data, ahead of an enterprise system implementation, can lead to a much more informed and productive implementation (among other things).  The key word here is "practice".  Analytics should not be a one time activity or administered only towards a discrete goal (like an enterprise system implementation).  Instead it should be an ongoing effort, employed at all levels of an organization, and across business functions, with whatever data might be available (be they spreadsheets or ERP systems).

Affordable analytic technologies exist today that make it easy to pull data in from these rudimentary and disparate data sources, and blend them together to generate analytic insights that address specific business goals and needs.  The practice of analytics not only enables these insights, but also informs potential investment decisions related to enterprise data systems.

So before you think about that new ERP system, think again!  You might be better served first practicing analytics with data you have right now!

Thursday, December 1, 2011

Hosted Analytics

Hosted IT solutions, or cloud computing is a technology trend that is enabling businesses to offload much of their IT infrastructure to a remote location that is managed by a third party.  Hosted services provide a variety of solutions: databases, file servers, applications and even computing power.

One service that still seems to remain local, however, is analytics.  For the most part, businesses are conducting their analytics in house and keeping the results somewhat centralized.  While business intelligence (BI) and analytics technologies exist that allow businesses to distribute analytics via the web and mobile devices, the price tag on these tools is quite high and prohibitive for small to medium sized businesses to own.

An emerging service to host analytics is starting to gain traction, however.  This is the notion that a third party would host reports, models, dashboards and analytics on web and mobile enabled platforms.  Since development of analytics requires a specialized skill set combining mathematics, IT, economics, business and statistics, these hosted solutions also often come packaged with business analytics consulting services that help transform business needs and overwhelmingly confusing data into insights and actions.  The full solution, therefore, is a service that enables an organization to understand the baseline of their current state, to set goals for their business through predictive and optimization models, and to monitor against these goals on an on-going basis.  All of this is delivered and consumed through web enabled tools.

While this notion, practice and consumption of analytics is not new, the offering of hosted solutions now allows small to medium sized businesses to get in on the analytics game that was previously enjoyed mostly by larger organizations.  All for a fraction of the price that would be required to make hardware and software purchases in the traditional self-hosting model.

The benefit of all this is that hosted analytics tend to be agnostic to enterprise IT platforms.  What this means is that a hosted analytics solution can tap into your ERP, CRM, HRIS, legacy and home grown systems (whether they are hosted locally or remotely) and gain integrated insights across these disparate platforms.  While some of these platforms do offer analytics modules or components, they often lack this ability to integrate with other systems to gain those cross-platform insights.  Additionally, these platforms are often competing technologies and don't "play nice" with each other when it comes to data integration.  This is why a hosted, independent analytics platform is so important.

While the concept of hosted IT has been around for a number of years now, the notion of hosted analytics is is just now being enjoyed by early adopters.  When it catches onto the mainstream like other hosted solutions, it just may change the way companies do business.

Tuesday, November 1, 2011

Small Data and Analytics

We hear a lot these days about "big data":  petabyte sized databases, millions of records, in-memory data processing.  I was at a conference recently where the speaker discussed his 70 BILLION record database!  Several large organizations that deal with data at these volumes have successfully leveraged their "big data" with analytics to better manage their business.

But what about the vast number of businesses that are NOT part of the Fortune 500?  Those that are not capturing massive data volumes of data as part of their core business?  Those that have  Mere gigabytes, perhaps.  Can analytics still benefit them?  The answer is: yes.

One key to leveraging small data for business success is to integrate data from disparate business systems: Google Analytics,, SAP, legacy databases, etc.  Big data systems that for many house their small data and potential for business insights. 

Integration need not happen within the systems themselves, but rather with the data that they generate, or expel, for the purpose of analytics.  Many business intelligence, reporting and analytics tools allow for dynamic and virtual integration of disparate systems, at the point of analysis.  This eliminates the need to monkey with the architecture of the systems themselves, and the need to develop ETL processes, data warehouses and data marts.  The best part of "small data" is that computing power exceeds data volume demands, allowing integrated analytical data sets to be generated dynamically. 

If performance does become an issue, some business intelligence tools will create stand alone "extracts" that are mini-data marts specific to the analytics at hand.  These extracts are a consequence of the analysis and do not need to be separately designed or optimized.  In addition, scheduled refreshes can be performed such that data stay current and live.

But what of these tools, systems and databases?  Will insights come simply by cross system integration?  Well...maybe, but probably not.  Unlike big data environments, where analysts can go swimming in the data and come upon insights by thrashing about, small data requires a bit more finesse (not that big data analysts don't have finesse).  Given the relatively small size of "small data", and the somewhat complex nature of how any business defines, measures, characterizes and organizes themselves, the different combination ways to look at the data begin to dwarf the actual amount of data available.  Something statisticians call "degrees of freedom". 

As such it is imperative when working with small data to begin with clear and concise business goals and objectives (see previous blog post: Micro-goals).  This focus will help narrow down the perspective applied to small data and will increase the degrees of freedom necessary for valid, significant and insightful analysis.

While "big data" does seem to be getting a lot of attention, it is the collection of vast "small data" insights that will will propel change in our businesses and economy.  Let's get started!

Thursday, October 6, 2011

Micro-goals and Rapid Analytics

Let's face it, the future of our economy is uncertain.  With the wild swings in the stock market over the last few months, and the on-going turmoil in Europe, long range, rigid, business planning is not only nearly impossible, it is unadvisable.  Instead, many business are in "wait and see" mode, putting them in a position to be reactive to what the market brings. 

There is a middle ground, however: micro-goals.  Micro-goals represent short term targets within the context of a longer term, but fluid organizational vision.  For example, if the long range organization vision is to double annual revenue over the next five years - the micro goal would be to grow revenue by 3.7% in the next quarter (3.7% is the quarterly growth necessary to double revenue in five years).  The feasibility of this short term target is confirmed by baseline analysis of past results, executed with the help of analytic decision models and tracked real-time by business dashboards.  If, at the end of the quarter, the micro-goal is not met - or better if it is exceeded, then the fluid organization vision may shift to reflect the latest reality, with analytic decision models revisited to discover what they missed.

The methodology for applying micro-goals is to:
  1. Define a flexible and fluid vision and roadmap for the future
  2. Develop a process map and conduct baseline analytics of current practices
  3. Re-engineer the process for short-term improvement, and develop analytic decision models to evaluate the likely impact of changes
  4. Execute the improved process and monitor progress real-time, against decision model benchmarks
  5. Revisit and improve process and decision models, adjust the long-term vision and elevate performance
The key tool that supports this business practice is "rapid analytics".  While the long term fluid vision draw on professional judgment of the business owner or manager, it is business analytics that support the actual execution of micro-goals driving towards this vision.  Traditional "big data", enterprise business intelligence (BI) lead by IT will not cut it, however.  While IT plays a crucial role in providing the infrastructure to manage and store data necessary to feed business analytics, they are not in the best position to deliver "rapid analytics".  Instead this should be conceived of and generated by the business user and analytics experts - in direct response to their micro-goals.  Like the analytics that they produce, the technology to accomplish this must also be agile, easily deployed and programable and usable by business users.  These technologies exist today.

Micro-goals do not serve a specific business function, but instead require support from a number of often independent business functions.  As such data and analytics that support micro-goals must also cut across disparate data systems that tend to exist in stovepipes like their respective business functions.  In order to keep pace with the rapid cycles of micro-goals, data feeding the "rapid analytics" must too be agile, lightweight and flexible.  The best way to accomplish this is not through large data warehouses or data marts, but instead through virtual data stores that dynamically integrate data sources and apply to very specific micro-goals.

This methodology, that leverages business strategy, analytics and technology, keeps the benefit of a long term, big picture vision, without being blindly commited to unrealistic and unpredictable long term plans.  Additionally, focusing efforts on execution of short term micro-goals makes for a more flexible, agile yet intentional organization that is constant learning and evolving.

Saturday, September 24, 2011

Drivers of Finite Promotional Periods

Marketing campaigns are a tool used to promote products and services. Campaigns can use various tactics such as email, social media, ads, webinars, direct mail, etc. Marketing campaigns could also be part of an on-going promotion of a brand, or a finite promotion leading up to some discrete event. This article focuses on a scenario when the campaign is directed at a discrete event - in this case a charity running race. The obvious trait of this environment, since the campaign is focused on a discrete event, is that the promotional period is finite.

The race directors utilized email campaigns to past runners to promote their event, as well as facebook ads and postings on various running websites. These activities drove potential registrants to the race website, and ultimately to register for the race online. While facebook ads and the running website postings did drive traffic to the website, very little of that traffic converted to registrations. The main drivers of web visits and ultimate registrations were email campaigns to past runners and the time remaining until the event.

The chart below shows the rate of registrations to web visits. It shows that on average there was a registration for 8 or so web visits (0.13 registrations per visit). This rate was variable, but stayed fairly constant throughout the promotional period.

The magnitude of registrations and web visits did not stay constant during the promotional period, however.  The chart below show a steep increase in both web visits and registrations as the event date drew near.  This chart is also annotated with the dates when email blasts were sent to past runners.

Careful inspection of the trends show that there might be a slight lift in the number of web visits in the days following an email campaign.  This lift seemed to be greatest the day after the campaign, and then showed a decreasing impact as more time passed.  Since the influence of days remaining until the event is so strong in this trend, the email campaign lift is difficult to spot.  A statistical analysis was run on these data to filter out the significance of the email campaigns within the context of the days remaining until the event.

This analysis naturally showed that the days remaining until the event were the main driver of website visits.  After correcting for that factor, however, the days after an email campaign still had a statistically significant lift in web traffic.  This lift was inversely proportional to the number of days past since the campaign (the fewer days since the campaign, the greater the lift).  The results of this statistical analysis were built into a predictive model that smoothed the noise out of the trend and allowed the race directors to identify email campaign latency and frequency that could increase the number of web visits, and ultimately registrations (for next year's race).  A live and active version of that model appears below.
While there are obviously many endogenous and exogenous factors that drive web visits and ultimate registrations (sales) in a finite promotional period, this model offers a simple perspective on how one of those factors could vary to positively influence outcomes.  Naturally, this model could be expanded to include other factors such that their collective influence may be understood.

Friday, August 26, 2011

Bootstrapping a Marketing Program with Analytics

With the continued pressure of a down economy, marketing executives are driven to not only expend marketing budgets more effectively, but also to ensure business growth from those budgets.  Any request for additional marketing funds requires a solid business case analysis that illustrates a direct impact on revenue growth.

With the explosion of various marketing channels in the last few years (web, mobile, email, social media, etc.) has come a tidal wave of related data. As these marketing channels are usually distinct entities, data related to each channel is also distinct, and often not integrated. While fully integrated marketing channels that track customers across channels from "first touch" to sale may be the Holy Grail of marketing analytics, the reality is that many organizations do not have systems in place that can serve up data this way.

Investment in marketing automation tools can provide part of the solution, but this potentially costly investment can be a tough sell to for the marketer that is trying to justify additional marketing funds in the first place!  Leveraging available data across disparate channels, with the right mix of historical data analysis and statistical modeling can provide a business case for a growth strategy that requires additional marketing spend.

Since a common customer identifiers are not available across disparate marketing channels, analytics at this level of detail is not realistic.  Spend justification relies on illustrating relationships between marketing spend and revenue, however.  This relationship need not be defined at the customer level, but instead could be defined across some common time interval such as days, weeks, months, quarters or even years.  Naturally, the more granular the time period, business variability related to marketing spend will be better understood. 

The example below illustrates this relationship for an online retailer that wished to justify increased sales through an expanded marketing program budget.  This retailer used a multi-channel marketing program that includes Facebook, Google Ad Words and Search Engine Optimization (SEO) to promote their products.  The chart below shows daily spend on Facebook to ad clicks that lead to website visits.  The strong relationship between these two factors indicates that Facebook spend is leading to website visits (naturally with a pay-per-click campaign).  Similar relationships are seen with Ad Words and search.

Given the multi-channel nature of this retailer's marketing program, and the fact that data from Facebook, Ad Words and SEO were neither integrated nor tracked at the customer level in their sales system, direct insight of channel spend to sales was not possible.  Instead, this retailer looked at daily website visits (total) to daily sales (total).  The chart below tells the story that higher daily web site visit volume leads to higher daily sales.

This insight, while valuable, does not justify an expanded marketing program budget, however.  In order to make the business case for additional marketing spend, with the constraints of un-integrated and customer blind data, the retailer built a simple marketing forecast model that drew upon the relationships seen between spend and web visits, and web visits and sales.  This model projected web visits from each source (Facebook, Ad Words and SEO) based on the known relationships and multiple future spend scenarios.  It then aggregated projected visits across all three channels and used that result to project sales, again based on known relationships of visits to sales. 

The result was the model below which provides a sales forecast for two separate growth scenarios.  Both scenarios assume a baseline spend for each marketing channel that is equal to historical levels.  Scenario 1 assumes that spend will not grow from this baseline over then next 15 months.  Scenario 2 assumes a 2% monthly growth from the baseline for each marketing channel.  Based on the underlying relationships, the model projects expected sales for each scenario.  The model displayed below is a live, interactive model that allows for creation of various scenarios (try it!).

Although the results are based on a rather simplistic model of a multi-channel marketing program, they do provide a sufficient business case to justify increased spend.  This simple model could be easily adopted to more complex programs, to include more channels or seasonality, for example.

While having well integrated customer level data across marketing channels is a preferred starting point for marketing analytics, the reality is that this unattainable and costly endeavor for many organizations.  Leveraging available data with common attributes does have the very real potential to provide insights that can bootstrap a marketing program for growth and success.