Currently, a staggering 86% of corporate entities are now likely to rely on AI for data-related requests, while almost three-quarters of these businesses will also make big decisions based on bogus results. As such, the biggest challenge facing corporate analytics is not so much extracting reports from their various products as it is to effectively control the method by which companies compute their metrics across the board.
The current guide of the new "data-first" warehouse approach to analytics discusses the governance framework for subscription-based metrics, as well as the specific platforms needed to successfully scale the reporting capabilities of an organization.
SaaS Business Intelligence Market Trends
The overall cloud analytics space is growing at a brisk pace, but many enterprise companies are still failing to leverage data to drive decision-making. There are numerous estimates for the future of the analytics space, with some predicting the market will amount to over $40B by the year 2033, while others estimate the space at $81B.
As of today, cloud-based products account for more than two-thirds of all analytical revenue, with traditional on-premise products mainly remaining within the confines of regulated industries. Despite the rapid adoption of cloud-based solutions, end-users are still struggling to fully embrace them.
According to several sources, approximately 60%-70% of business intelligence projects will produce subpar results within the first two years of implementation. The failure to meet expectations for these projects typically has nothing to do with the applications being implemented. Instead, the root causes for failure are due to a lack of data readiness, poor definitions for metrics, and insufficient support from analysts.
One of the most profound mistakes that an organization can make is to invest in costly presentations before a solid data infrastructure is developed.
When developing an analytics portfolio, the total cost to develop a SaaS business intelligence stack will always exceed the cost of the software license. Therefore, during the transition phase, organizations should budget two to four times the cost of the software license to account for costs associated with data engineering, tool integration, and user training. The expensive licenses that you purchased will collect dust on the shelf without this investment.
The Warehouse-First Data Architecture
A primary consideration when developing your data analysis stack is how the company will organize the collection and storage of data. When using the warehouse-first architecture, it is essential to have a single central repository for all company data prior to analyzing it.
The first part of the warehouse-first architecture is data ingestion and storage. A company has all of its sales data in Salesforce or HubSpot; all payment-related information must reside with Stripe; and every marketing-related interaction is collected by Google Analytics.
Applications such as Fivetran or Airbyte allow users to pull these various forms of information from the various applications and load them to a single data warehouse. A data warehouse is a database created specifically to conduct analytic queries.
The most widely used cloud data warehouses by expanding software companies include Snowflake, Google BigQuery and Databricks. The data warehouse serves as the single version of the truth for your business. If you do not see a metric included in the data warehouse, you cannot rely on that metric as having been calculated by your company.
Once the data is loaded into the data warehouse, it must be cleaned and formatted into clear business concepts through the transformation of data. This is accomplished through tools such as dbt, which allow companies to format data and arrange it into logical tables.
The semantic layer can be considered the dictionary of your data warehouse. Without a semantic layer, it is likely you will have three different versions of your monthly recurring revenue (MRR). The semantic layer will guarantee that all users of all tools are using the same calculations to determine your metrics, thus eliminating the potential for miscalculating your outcomes.
Rules to Define Analyst Capacity
One of the most significant beliefs of software buyers is that the elimination of data analysts will occur through self-service analytics solutions. This is incorrect in theory, not just in practicality.
Traditionally, an industry standard is 1 dedicated data analyst to every 50-150 business users. If you exceed this ratio, your dashboards will fail.
When users are unable to gain access to help from an analyst, their trust in the SaaS business intelligence data will decrease dramatically. Therefore, their freshness scores will drop significantly. As a result, all credibility associated with an entire data project will diminish. In order to be successful, you must employ analysts in a balance relative to your user base.
Key Metrics for SaaS Business Intelligence
A SaaS company has a very comprehensive list of financial and operational metrics. Traditional metrics for subscription revenue are difficult to evaluate accurately with standard dashboards. A SaaS business must be able to accurately monitor and evaluate how many upgrades, downgrades, and annual prepaid contracts are acquired by the business.
Fundamental Revenue Metrics
Monthly Recurring Revenue (MRR) and Annual Recurring Revenue (ARR) make up the core metrics for SaaS businesses. However, tracking only total MRR is insufficient. A SaaS business must monitor the movement of MRR over time.
SaaS businesses must ensure that their data model architecture must provide the ability to separate and identify new MRR, expansion MRR, contraction MRR, and churned MRR. The individual movement of MRR in these categories provides different insights into the health of a business.
Net Retention Rate (NRR) and Gross Retention Rate (GRR) are used to quantify how well a company retains its customers. The GRR metric identifies the percentage of revenue that has been retained since the start of your business, disregarding any new upgrades to customers. The NRR metric demonstrates whether the company continues to grow in value or contract among its existing customers.
Customer Acquisition and Health
Customer Acquisition Cost (CAC) payback period is the primary key efficiency metric for SaaS businesses. The amount of time it takes for a company to recover its marketing costs is determined using time to recover (T2R), expressed in months.
For start-up businesses, customer lifetime value (CLV) is an unreliable metric since these businesses do not have access to historical data. Therefore, it is best to focus on activation rates for new customers.
An activation rate is defined as the percentage of new customers who perform a key action using your service. When customers complete a core action within your service, they tend to stay on longer and generate larger order values (LOVs). To determine how successful you are in retaining customers, your analytics should correlate your marketing costs with customer activation.
Software Tools and Platforms
There are numerous software companies that have created an all-in-one package that looks appealing to customers, yet the technology behind an all-in-one solution can be very different from one vendor to another. Each vendor uses completely different methodologies to solve problems and support their clients.
Enterprise Discovery Platforms
EDPs are large software products that provide extensive exploration and analytical capabilities. Popular EDP options include Microsoft Power BI, Tableau and Looker. All EDPs connect to a data warehouse.
Microsoft Power BI is commonly used by companies that utilize the Microsoft ecosystem. Its pricing starts at approximately $14 per month for the Pro version and $24 per month for the Premium version.
Tableau provides a user-friendly interface and a visual data exploration process. However, significant amounts of training are required to use Tableau effectively. A Tableau Creator license typically costs approximately $75 per month.
Looker is heavily integrated with the Google Cloud platform and offers the benefit of managing governance and consistency via LookML. Due to the increased complexity in utilizing Looker, it is recommended for technical analysts to be responsible for its use. Looker also has plans starting at approximately $5,000 per month.
Lightweight SQL-Driven Tools
Many companies do not require an enterprise-level solution. Mid-size businesses typically prefer lighter-weight and faster tools that enable easier implementation and flexibility. Metabase, Mode and Lightdash fall into this category of software applications.
Metabase allows for the non-technical user to ask questions without having to write any code. A Metabase Starter plan will typically cost a SaaS company around $85 per month for five users, making it very affordable for SaaS companies that are less than one year old.
ThoughtSpot is at the forefront of this shift to search-based analytics. Customers can enter their questions in plain English, but it is expensive and often starts at $95/user/month.
Product Analytics and Event Tracking
Tools that use a data warehouse as their foundation are poor at capturing current user clicks on the software interface. Specialized product analytics software is required to effectively track events in real time.
Product analytics companies such as Amplitude, Mixpanel, Heap, and PostHog collect a great deal of data related to user behaviors in their products and are excellent for creating funnels to convert users into customers and for retention curve measurements.
Typically, product-led SaaS companies run 2 different systems simultaneously: Amplitude or Mixpanel are used to track product events, while report-quality marketing and sales data is typically aggregated by Databox or Improvado.
Financial Subscription Reporting
General-purpose data processing platforms have major problems calculating complex subscription proration automatically. However, it can be calculated instantaneously using a revenue analytics platform.
ChartMogul and Baremetrics connect directly to Stripe or other payment processors. These platforms provide exact figures for monthly recurring revenue (MRR), churn rates, and cash flow without a data engineer.
Most companies will likely begin using ChartMogul, but as they grow, they will need to combine billing data with usage data. Hence they will transition to a custom SaaS business intelligence warehouse solution to do this.
The Embedded Analytics Market
The embedded analytics marketplace is a distinct category of software solution. This category consists of products designed for inclusion in a company's customer-facing software to provide customers with data-driven dashboards.
Multi-Tenant Architecture Security
As customer-facing dashboards require a significant level of security, they must be built with the capability to segregate users using tenants from one another. Company A's data cannot be inadvertently delivered to Company B.
In order to accomplish this, row-level security (RLS), as well as a rigorous multi-tenant implementation must be enforced via the use of integrated enterprise software applications such as Holistics, Sigma, Qrvey, and Sisense. By using these tools, a company can design written analytics code once and distribute it securely to thousands of clients.
Most enterprise applications are not suitable for embedded analytics because they rely heavily on coded visual templates rather than a coded model. An embedded reporting system that can scale needs to have a Git-type code-first development methodology.
Avoiding the Viewer Fee Trap
Embedded analytics pricing schemes can be detrimental to the profitability of a SaaS company. There are three major schemes that will impact profit margins: per-user, by-capacity, and per-viewer schemes.
The per-viewer scheme can be particularly problematic for a rapidly growing SaaS company. For instance, an embedded tool that charges $12/user/month charges $6,000/month to have 500 customers using their dashboards. If the number of customers grows by another 500, the reporting expense doubles, then doubles again every time it does.
When it comes to embedded analytics pricing, a company should always seek a capacity-based pricing scheme, which focuses on the computing resources paid for rather than the number of individual users. Most buyers want to know exactly how much they are paying and have reported that negotiations with companies typically result in pricing discounts ranging from 25% to 50%.
AI in SaaS Business Intelligence
Artificial intelligence's integration with business intelligence is changing how end users perform their roles. Nevertheless, depending exclusively on the language capabilities of AI for performing basic mathematical functions can have severe consequences.
AI Trust Deficit in Reporting
It has been documented that 91% of regular AI users use it at least once every day for business analytics applications. However, due to high rates of error with large language models, there is often a lack of trust in the data produced by these tools, as they merely predict the next word in a sentence rather than accurately finding a solution to the equation they are attempting to solve. Users of traditional analytic tools have learned not to trust their AI results.
When AI users find one erroneous number in a dashboard generated by an analytic application, they label the entire dashboard as untrustworthy. As a result, users are now looking for analytic tools with query-sourced and governed accuracy.
Metric Verification
Metric verification is possible with good modern analytic tools. All good modern analytic tools enable users to trace back through the computing history of the number returned by the AI to ensure accuracy. This includes the SQL statement that the AI generated to return the metric.
A good modern analytic application provides users with the exact definition of the metric used to calculate it, which is important. The AI simply takes the English language definitions provided by the end user and converts them into SQL. Actual calculation of numbers is performed within the data warehouse's query engine.
You should never purchase an analytic tool that claims it can remove your analytics team and replace them with AI. Metrics must still be defined by human analysts. Then, the AI provides a means for non-analysts to query those SaaS business intelligence metrics.
Architecting a Unified Data Source for Enterprise Analytics
Strategic resolution requires the creation of a unified data source. Simply buying analytic presentation software will not solve the issue of incorrect data logic. The purpose of an enterprise analytic program is to create a definitive, one-source-of-truth for the organization. Thus, the first step in building your analytic strategy should be to consolidate every business occurrence into a single managed data warehouse.
After the data has been moved into either Snowflake or BigQuery, you must then establish a strict semantic layer governing every business critical metric from net revenue retention to customer acquisition cost and assign a titled ownership position to each one.
If you do not have data governance in place while you start to scale the way that you report on your data, then all you are doing is creating new opportunities for confusion. In fact, the way different departments will come to meetings with different sets of numbers is going to completely destroy any momentum that might exist for you and your executives.
Therefore, if you focus on building the data engineering, data semantics, and proper ratios between the number of analytical resources you employ versus the amount of output that you create, you will ensure that you are producing accurate reports.
The net result is a company that is able to operate much more quickly and has the confidence in its own data to make better informed decisions and scale its operations intelligently without wasting money on software licenses that do not produce answers.
FAQ
How do we manage attribution conflicts across reporting platforms?
Attribution conflicts occur when advertising or marketing platforms function separately from financial systems. Typically, advertising platforms want to take credit for every conversion they have generated, whereas financial systems only record the final payment transaction. To resolve this problem, advertisers must extract and bring together all raw dynamic marketing data from various marketing channels and technologies.
Using a tool such as Fivetran, this data must be moved into a central location (data warehouse) so that data from multiple systems can be analyzed. Once this data has been collected, you can create a data model that enforces a consistent method for determining the attribution of the data to the business (a company-wide defined attribution model).
This ensures the information from the executive report is based on actual cash flow versus estimated cash flow by the marketing department.
When should a growing SaaS business stop relying on built-in reporting and begin using a dedicated data warehouse?
If you have business questions that require you to combine information that is collected by three different systems or more, then you have outgrown the built-in reporting capabilities. For example, if you need to know how something that was sent in an e-mail marketing campaign impacted the long-term use of your product and ultimately determined the need for upgrades in your billing plan, built-in reporting tools will not suffice.
When the time required to manually export and combine reports using spreadsheets exceeds the cost of hiring a junior level data engineer, then you must have a data warehouse.
What are the major warning signs that a data analytics implementation has been developed inadequately?
The first and most obvious warning sign is purchasing a data visualization tool before the development of a semantic layer or the establishment of a data warehouse. A second indicator is not having specific individuals assigned as owners of key metric definitions.
Finally, if the ratio of business users to data analysts is greater than 150:1, the system will suffer from sync errors, broken dashboards, and, eventually, total loss of confidence from the executive team.
