Data analytics provides a way to use data to support actionable business decisions. Many executives believe their companies have a significant data issue, whereas, in fact, the real issue is a lack of actionable decision-making capabilities.
Investing millions of dollars in software solutions without connecting them to any operational decisions will result in wasted capital. Data only has value when it is used to influence behavior, daily processes, or finances.
What Can be Gained by Taking Structured Decisions
Many analytics initiatives fail due to the large disparity between insights gathered digitally and actions taken by executives. Executives must stop purchasing expensive dashboards that are seldom utilized by their managers and connect each data point with a measurable financial impact.
Revenue Growth Transformations - An Analysis of Corporate Case Studies
An analysis of corporate case studies shows that when a company implements rigorous process updates to incorporate intelligence into decision-making, they can expect to see an increase in revenue and decrease in cost of 20% and 12%, respectively, and have a 30% faster decision-making cycle.
Shifting E-Commerce Operations to Big Data
A mid-market retailer that adopted big data architecture experienced a 12% increase in revenue, a 15% increase in customer satisfaction, and an 8% decrease in operational cost.
Productivity Metrics
Benchmark data from various industries demonstrates an average productivity increase of between 5 to 6% for teams that successfully utilize data versus teams that only discuss data at meetings.
The Need to Break Down Silos through Technology Infrastructure
If you do not have a solid technical foundation, your data analysts are simply guessing based on mathematical concepts. Successful system integration is crucial for sustainability, as it is costly, time-consuming, and extremely challenging to complete.
Centralized Data Systems and Integration Expenses
Legacy systems are a barrier to getting peak productivity from your business because they compartmentalize all business information into different divisions. A centralized system gathers dispersed information but utilizes extensive extract, load, and transform (ELT) processes.
This technical workload cannot compare to the massive amount of human labor required to smooth and match the source data when it comes from legacy systems and malfunctioning API connections. Often, teams in the data engineering space take months repairing corrupted databases before any visualization is produced.
Additionally, the technology architecture of a solution includes cloud-based data warehouses from which the business intelligence layers extract information, as well as machine learning models and resources that relate and interact with those two systems. Missing the extreme volume of technological development required to build such an architecture will invariably result in significant budget overruns.
You cannot create management confidence with inaccurate data.If there are inconsistencies in the core sales metrics between departments, management will stop using the dashboards altogether.
Evaluating cloud-based warehouses and machine learning solutions, you will want to strictly measure data integration costs against expected business value. Today, cloud computing systems typically consist of a cloud warehouse that streams to the business intelligence vendors and predictive analytic tool vendors.
As you acquire these types of solutions, you will want to create clear parameters that define the cost tiers associated, daily compute costs, and cold storage cost. If the compute costs are expensive, this will destroy your profit margin very quickly unless your team's database query operations are optimized through the use of senior-level engineering expertise.
For the next three years, keep your focus on the total cost of ownership. Often, vendors will present very low initial software license costs to get your attention; however, soon thereafter, they will present you with significant fees that will be incurred once you begin to operate the more complex queries.
Stop making procurement and tool decisions based on aggressive vendor placement.Make all of your procurement decisions regarding technical items based solely on clear, quantifiable operational use-cases.
Utilising Data Analytics for Strategic Business Decision-Making Over Time
Timelines are critical when planning for success in business. The vendor's promise of "instant value" can be very deceptive. Value transformation takes time, and the success of an organization will depend on an appropriate plan for the technical deployment of systems and the adoption of processes by all stakeholders.
The First 90 Days of Phase 1 - Descriptive Analytics
For the first 90 days of Phase 1, focus your efforts strictly on descriptive analytics, establishing a reliable baseline of "truth." This stage answers the most basic question: "What is happening RIGHT NOW?"
The primary deliverable for this stage is a central dashboard displaying major key performance indicators (KPIs), such as weekly revenue, cost per order, and active user retention rates. Engineering should focus on integrating the easiest, highest-value data sources initially, so that the proof-of-concept can be demonstrated.
Avoid trying to take on complicated machine learning challenges during this phase. Establish basic governance around your data by determining who has ownership of it, who has access to it, and who is responsible for securing it. The time required to get value out of this particular phase should always be less than 90 days.
The first phase is purely the demonstration of technical competence.At the end of this stage, you should have produced one clear, unequivocal success that secures continued funding from the executive team.
The Second Phase - 90 Days to 270 Days, Predictive Modelling
Once you have established a reliable baseline, you will transition from descriptive analytics to diagnostic and predictive analytics for a window of 6 to 9 months. During this time, the business will now want to start determining "why" things happened and "what" may happen next.
A mid-market retailer trying to reduce stockouts will require predictive inventory models in order to make improvements. The development of predictive algorithms that utilize historical purchases and seasonal trends will start shortly.
The data lake will have to be significantly improved, and all variables will need to be run through sophisticated machine learning systems. Expect a strong backlash from most mid-level managers, who are unlikely to want their supply chain to be directed by an algorithm.
Getting human buy-in is much more difficult than managing the database.You need to create an environment where your employees completely trust the results of the calculations.
Phase 3 – Days 270-540, Prescriptive Automation
This is the last operational phase, which requires approximately 18 months of continuous development. Prescriptive modeling not only generates a prediction of what is likely to occur, but it mathematically identifies the precise action to take in order to increase profit and/or reduce loss.
For example, hospital providers must reduce the labor costs for clinical services while maintaining clinical outcomes; prescriptive modeling can help hospitals assign staff schedules based on estimated patient volumes for each day. The operation of the data pipelines in this phase must occur in near real-time, with no interruption of service.
A data-literate culture is essential for the prescriptive modeling phase to function properly and be successful. The ROI during this phase is overwhelmingly positive; however, the failure rate is also relatively high due to the necessity for a complete overhaul of the policies and procedures regarding daily operations.
You are no longer just reporting the numbers; you are automating all of the operational decisions.
Creating a Governance Structure for Decisions and Measuring Performance
Failure to comply with the governance policy results in absolute chaos in the corporation. The governance of data is not an IT issue; it is a critical part of a company's business strategy. Analytics outputs should be connected to C-level operations.
The way you measure your marketing spend can have a huge impact on how likely your customers are to churn. Therefore, to create a dashboard for a CMO to evaluate vendors based on this information, you must have every aspect of that dashboard directly tied to a business cycle.
For every project you undertake, you will create a strict written measurement plan that outlines the exact metrics, the source of those metrics, the expected benchmark range, and the name of the individual who will be held accountable for the outcome.
If any of the metrics are red, you will know exactly who in the organization is responsible for fixing it. If you cannot answer that question, remove the metric from your dashboard. C-level executives do not want to see 50 different charts displaying random things; they want to know what five charts will dictate where they allocate their resources each week.
Vanity metrics are detrimental to the focus of your business and create inefficiency through wasted time. Only measure those metrics which you will actively manage, modify, and improve.
Managing Limitations Around Data Privacy and Data Literacy Constraints
Globally, data privacy is one of the primary limitations on enterprise-wide analytics programs. You must have a formal written process for sign-off on access to data throughout an organization.
Through strong governance of decision-making, it should be clearly defined who will have access to raw data on customers and who will only have access to aggregated market data. In addition, you should be investing heavily in providing basic data literacy for all workers in the organization who use data.
A highly advanced, sophisticated machine learning model will not provide any value to the sales team if they cannot interpret the basic probability scores generated by it. Lastly, the operations teams should create standard service level agreements for all data requests so that highly compensated analysts are not constantly used as report pullers by lazy managers.
Your company's data is an extremely important regulated financial asset.
Regularly audit your access logs on a weekly basis and continually train your staff.
Obstacles to Implementing Analytical Solutions
To develop a modern data culture, one must work to break down corporate traditions. The barrier to implementing solutions is often not the technology itself but rather the resistance of the people using it throughout the process.
The Knowledge Gap in Operational Teams
Many organizations seriously underestimate the technical knowledge gap of their operational teams. Deploying the latest and greatest predictive analytics platform in the world will not achieve success if the marketing manager can only read a simple spreadsheet.
Closing the knowledge gap must occur through ongoing training and education, rather than one-day training programs. Organizations must also hire dedicated analytics translators who have a solid understanding of advanced predictive analytics techniques, as well as the ability to communicate clearly in business language.
These translators will help connect the analytical, predictive, and data science teams to the executive level and facilitate the transfer of actionable data to the people who make business decisions based on the analytical results. Without dedicated analytics translators, data engineers will build products to answer the wrong questions.
Your data scientist should never be expected to teach you about business strategy.You must hire someone to provide this bridge.
Legacy Applications as a Barrier to Modern Data Solution Adoption
Legacy applications are a direct impediment to adopting a modern data solution. Many established mid-market businesses continue to run their core operations from outdated systems that were created 10 to 15 years ago.
The outdated legacy systems currently in use do not have an interface with the cloud lake, so your data engineering staff will have to create new code to gather and export sales data every day. This results in significant delays to implement the new system and will quickly deplete the project's overall budget.
Therefore, only brutally honest assessments of the existing systems should be provided so management can make it known to the executive staff that real-time analytics can no longer wait. Many times, it may be cheaper to purchase a new cloud-based solution to replace an outdated sales system rather than trying to interface a new machine-learning platform.
Never build a long-term analytics house on a frail or crumbling base. Your company should kill your legacy systems before legacy systems destroy your data strategy.
Evaluation of Procurement and Tooling Spending Within Parameters
Enterprise software vendors tend to sell products as if they will change the world. However, evaluating and purchasing enterprise-level software is an extremely painful exercise in managing the overall cost.
Evaluating Time to Value Within a Current, Up-To-Date Stack
The only measurement that will give the most accurate picture of your organization’s ROI will be the measure of elapsed time to see any value from the purchase of any external analytics vendor. The time-to-value metric is the time from project initiation until measurable financial results can be observed by the business utilizing the purchased goods.
In a well-designed, well-executed dashboarding solution, this number should range between three and six months. For advanced prescriptive and predictive analytics machines, the time frame is often nine to eighteen months.
Software vendors also tend to hide the length of the integration period in between when they tell customers the sales pitch and then demonstrate an absolutely pure and pristine user experience. The vendors do not want to take the time to walk you through all of the messy data integration and project setup; they often just want you to look at a perfectly sanitized customer demonstration of the program.
Never sign a major contract based on a demonstration that provides theoretical proof of value rather than a proven project that you develop yourself using your own messy data samples.
Hidden Realities of Platform Pricing
The initial cost to license a given analytics platform is only a small piece of the overall total cost of ownership. Most modern cloud products are charged on a straight basis on total computing power and total storage volume used.
Given that once a business adopts a data-driven strategy, their usage of queries will typically increase exponentially. If an organization has not perfected the optimized architecture of the database, typical costs incurred for just one daily revenue query could total thousands of dollars per month.
Therefore, procurement staff should develop a clear tier of pricing and hard limits on the usage of any future agreement before signing it. In addition, procurement staff should conduct a detailed analysis of what the hidden costs of training and adoption will be before signing. A product will have zero value if your employees spend the first six weeks of training on filtering out their most basic tables.
User-friendly interfaces increase enterprise adoption.Complex solutions almost always become expensive, abandoned software.
Final Words
Stop blaming your data engineering team whenever your company has a downturn in sales. Hold your company’s executives accountable for their own actions by requiring them to provide clear, measurable output for their business operations using only fact-based numbers.