Collections

The Transformation of Analytics into Intelligence

Analytics is Not the Same

For almost two decades (1990-2010), Analytics was synonymous with reporting and cool dashboards for executives.  

But since 2010, the Analytics Stack is morphing into a parallel and equivalent entity to the operational/production IT stack and accruing as much if not more budgets, heft, and power. The new Analytics Stack has the mandate to generate closed-loop actionable intelligence in the enterprise that can be consumed as a service by both people and systems within and beyond enterprise boundaries.

True for both pure digital companies and traditional enterprises, the new analytics stack is different from the old analytics stack in the following ways:

1. Dedicated teams with Blend of Business, Data, & Technical Talent vs. IT-Only teams

The New Analytics stack has its own dedicated cross-functional teams, a sharp contrast to the motley collection of report writers and BI analysts in the old stack. The team composition mirrors product management teams in digital companies and key business functions in non-digital or industrial companies. For example, at Neflix, the analytics teams are organized around content development, acquisition, payments, content recommendations, and retention. At an industrial company like, say Daimler or BMW, they are more likely to be organized around marketing, sales, manufacturing, R&D, etc.  

2. They Report to Business CxO, not to IT

The Analytics teams are typically headed by a Chief Analytics Officer, Chief Intelligence Officer, or a Chief Data Officer. They report directly to the CMO, head of products, or in rare cases, the CEO. These teams are funded by large and growing budgets.

3. They Deliver Real-Time Actionable Intelligence with Measurable Business Outcomes

Increasingly, intelligence from the new analytics stack is exposed as real time mission-critical APIs driving key aspects of the consumer experience. Think “people you may know” on Linkedin, recommendations from Amazon, the pre-filtered search list, the real-time translations on Google translate, auto-flagging of NSFW content, etc. 

4. New Revenue Streams and New Business Models vs. Dashboards

For traditional enterprises, the analytics stack represents the next big wave of tech investment to unlock value from the data collected by ERP/ CRM/ E-Commerce/ IoT/ Transactional systems. Think a) A Bank delivering a dynamic list of up-sell/cross-sell offers presented to a customer on an ATM machine, b) A security equipment provider changing field service from “Dispatch to Diagnostics” to “Diagnostics to Dispatch” and thus transforming the economics and effectiveness of service, and c) A Retailer selling customer insights to upstream CPG companies.

These projects also unlock a range of new business models (servitization of physical equipment, dynamic insurance premiums linked to behavior,  spot insurance based on data driven fraud and risk assessment, etc.)

5. Closed Loop from Data to Action to Outcomes and in Real Time

Old Analytics Stack = Data Engineering + Data Science (Batch Insights)

New Analytics Stack = Data Engineering + Real Time Data Analysis (Insights) + Real Time Decision Support (Enabling better decisions and actions by people and systems) = Business Outcomes feeding back to Data Engineering.

The Analytics stack is also expanding into the “Intelligent Application / Agent” opportunity that is more operational / transactional, and starting to blend with MDM type applications for product, customer, and supplier data.

Now that we see how the new analytics stack is just as important (or more) as the mission critical production systems, how can enterprises get the help they need in building competence and capabilities in conceptualizing, implementing, maintaining, and improving the stack?

From the work I do, I believe spending related to the analytics stack can be decomposed into five submarkets from the perspective of the client assembling the stack. Each represents different levels of commitment, opportunity, risk, and accountability for vendors and providers. 

The Five Kinds of Analytics Vendors: Know What You Want and Who You Are Dealing With

a) Analytics Talent as a Service: Provide skilled data engineers, data scientists, and other data related talent who can supplement existing client teams to address the profound gap in demand and supply. Integrators providing these services are typically specialized by specific tools and platforms (as they should be). Clients need to be very specific on the skills they lack and in evaluating how the vendor can complement them.

b) Prepackaged Analytics / AI Solutions: These are configured and bundled using third party data and AI platforms (like Watson), and delivered to address specific use cases. The data collection and wrangling will indeed be custom to every client, but will use pre-defined data models, templates, application logic, and UX. This can also be at the horizontal level (ex: pre-packaged connectors to common apps, making integration easier for analytics projects). This is still a people-based model, but more value-added with potentially higher margins, and benefiting from referrals through the software vendor. Perficient relationship with IBM Watson is a good example.

c) Analytics Software: This represents data analytics/AI business applications for specific domains. The difference from b) is that the business model is selling and delivering a licensed product or SaaS, and not an accelerator or an amorphous solution. (A good example might be Ayasdi which sells both a platform and a set of unique business apps; Opera is similar.) Most of the traditional software market is outside the reach of services companies, but the pie is large and there are indeed several areas not addressed well and emerging.

d) Custom Intelligence: This is the modern equivalent of “Application Development and Maintenance”, but in the context of Analytics and AI. This involves delivering custom intelligence solutions for enterprises using a combination of business consulting, software (owned or/ and 3rd party), and custom development. Deployments are usually soup-to-nuts transcending data engineering, data science, decisioning, and integrating the intelligence into end-user transactional applications. The difference from b) and c) is that these projects are larger in scale and scope, more custom, require more direct accountability for business outcomes, and involve multi-year contracts spanning development and maintenance. (A good example is Palantir)

e) BPO for Data-Centric Functions: Becoming a trusted analytics partner who can take on larger scale business process outsourcing in data intensive domains. (Ex: Loyalty and Personalized Offers in Retail, Claims Fraud in P&C Insurance, Medicare Fraud in Healthcare, Intelligent up-sell/cross-sell in Banking, etc.). These are massive deals negotiated at the board level, and there are several industry specific analytics BPO firms doing this in Retail, Healthcare, Insurance and Media. These projects typically go beyond data analytics and extend into every activity in the business process being outsourced, including harvesting and selling data to upstream suppliers.

There are certainly some overlaps across the five categories, but the profile of the buyers and the deal characteristics (deal size, deal cycle, margins, risks) are fairly distinct by category.

In a future post, I will discuss challenges related to the new analytics stack in terms of team composition, hiring, and retention, and key strategies for addressing them.