3 reasons business intelligence is not human centric and how to fix it

The state of the art in Business Intelligence is humans understanding business health by looking at dashboards, often in weekly or monthly business reviews. Tableau, the leader in this category, captures this paradigm perfectly with its tagline: “We help people see and understand their data”.

But this approach is not human centric. Here’s why.

3 reasons why BI is currently not human centric

1. Psst…no one’s looking: Most people are not good at boring and repetitive tasks like looking at dashboards. This is the dirty little secret of BI. Behind closed doors, data teams share how little usage their dashboards get. They’re like unused gym memberships. I welcome every team to track daily/weekly active users for every dashboard they’ve created and create a usefulness-scorecard (perhaps even as a dashboard :-)).
2. Not enough eyes to monitor everything: Even if we could shame people into looking at dashboards every day, this approach just doesn’t scale anymore. When we had thousands of records and tens of columns, this could’ve worked and that’s why Tableau and Looker are massive companies. Now we have billions of records and thousands of columns. Human vision and memory cannot monitor or make sense of this much information. So we attempt to aggregate data to match human capacity. In this aggregation, all actionability and meaningful insight is lost.
3. Dashboards aren’t explanations: We repeatedly hear from business users that what they crave is not the dashboard but the explanation narrative next to the dashboard screenshot in slide decks they look at every week. These explanations are generated manually by analysts.

A human centric approach to business intelligence

Before getting into the solution, here are the principles behind human centric business intelligence.

Principles of human centric BI

1. Optimize for what humans are great at: Executive decision making and storytelling
2. Leverage machines for what they are great: Repetitive tasks, pattern recognition and automation at scale
3. Assume that business decision makers are intelligent and the reason they don’t use data is not because they don’t want to, or they aren’t “data savvy,” but because they are being given data that is either too high level to be useful or too granular and overwhelming

Inspiration

With the shift from packaged software to highly available online services, managing the health of these services underwent a transformation. Instead of engineers and IT teams monitoring health via dashboards and rudimentary alerts, DevOps systems like AppDynamics, Splunk, and Data Dog emerged. They were able to capture, monitor, and detect health issues and notify Ops teams with actionable, specific problems to address. Without these systems, we would be lucky to have 9% uptime as opposed to the 99.9% uptime we have come to expect.

Can you imagine if Amazon.com was only available 9% of the time?

There’s no reason, other than self imposed limitations, why business health shouldn’t undergo the same transformation. This is our inspiration at Falkon, described in more detail below.

Solution

Define business health as good metrics

The foundation of true intelligence is good metrics. In future posts, I’ll share metrics and dimensions teams should be tracking (sales, customer success, marketing, demand generation etc.). The thinking that will guide these recommendations is captured here.

Good metrics:

1. Are granular (computed hourly, daily, not pre-aggregated to monthly). Granularity results in richer analysis and understanding of business health.
2. Have lots of dimensions (i.e. you can slice and dice them in hundreds of ways). By region, SKU, platform, channel, campaign, operating system, customer cohort, deal size, city, zip code, source). If you have 2–3 dimensions on a core metric you look at, you’re short changing on all the hidden insights.
3. Have actionable dimensions. Dimensions like Campaigns, Test IDs, Account Rep ID, Channel, Account Executive ID, Customer Support Rep ID, Product SKU, Promo, all represent activities that we control to encourage good business performance. Without these dimensions available on metrics, we cannot explain which of our activities is causing business health to change.
4. Represent input metrics and output metrics. A lot of companies have output metrics like Churn, Revenue, and NPS but are remarkably scarce on input metrics like user engagement, time to 1st interaction, time to customer ticket resolution etc. With only output metrics, we are always putting out fires about why the output is not what we want. With input metrics, we are in the driver’s seat and good outputs become inevitable.
5. Are properly typed. A count metric (e.g. number of Leads) is analyzed very differently from a rate metric (Lead Conversion to Opportunity). Not understanding the difference between them means our understanding of why they are moving will be incorrect. More on this in a subsequent post.

Automate business health monitoring

With all these wonderful metrics with lots of actionable dimensions, a system needs to be put in place to constantly monitor the performance of these metrics. Banish the human-on-glass version of BI and replace it with machines.

If the monitoring system is rudimentary, it’ll be noisy and cause more trouble than it’s worth. If it’s smart, it will amplify the signal from the noise.

A good business health monitoring system

1. Detects anomalies: These are statistically significant, highly unlikely events. Often these end up being data bugs or service outages/bugs
2. Detects drifts and shifts: These are slow undercurrents in data that are happening over long periods of time and often represent meaningful changes in the business. For instance a channel that used to perform really well and continues to get traffic has also slowly declined in its ability to generate transactions.
3. Identifies noteworthy changes based on smart baselines: Monitoring an account development representative’s (ADR) activity is very different from monitoring conversion rate for every platform and app version. The difference is in the definition of “expected/desired” behavior i.e. the baseline. For conversion rate this may be derived from the historical values over 6 months. For an ADR, the baseline will be inaccurate if it is solely based on that rep’s historical performance. Instead, the baseline needs to factor in other reps’ performance, the overall team’s performance, and the goals set by the manager.
4. Understands seasonality: Many metrics and activities have a seasonal nature where certain changes can be explained by day of the week, day of the month, week of the month, week of the year. For instance Cost-Per-Click for most campaigns will increase for retail companies during the holiday season. This is not indicative of under-performance but rather environmental competitiveness. Similarly remittances will go up around pay day. A good business health monitoring system can identify the unique seasonal patterns hidden in historical data and infer if an increase/decrease is truly unexpected or just a seasonal effect.

Alert the right user on the right channel, at the right time

As important as monitoring is what is done with the insights that are generated. If these are sent to the wrong people, they’re noise. If they’re sent at the wrong frequency, they’re noise. If they’re communicated on the wrong channel, they’re invisible.

A good alerting system:

1. Leverages users’ existing workflow: No one wants another tool to log into, another password to remember, another dashboard. Alerts should just be sent on Slack and Email, two places many of us spend a lot of our waking hours and feel a high level of comfort.
2. Alerts the right users: Sending an alert about an underperforming campaign to a product manager, or an alert about an over-performing sales representative to a customer support manager is noise. However if the alert is sent to the person that’s most empowered to take action, it’s signal. Similarly if five metrics drop but only one is the root cause, alerting on all 5 independently is noise. Alerting on the one that’s actually the cause is signal. A good alerting system is able to route the right message to the right person and de-duplicate alerts to what’s most causal.
3. Alerts with the right urgency: Some alerts require immediate attention. Others might be best served as a “trend report” in a weekly business review. A good health monitoring system is able to act with priority in mind.

Envision a human centric BI solution

With the right solution in place, a business user is proactively informed of the most important business trends that need their attention today, this week, this month. They are provided with enough context and explanation that they can immediately take action. They spend most of their time forging the path ahead, confident in the actions they take, armed with insights (not dashboards).

This is the vision that drives every single day at Falkon.

Want more articles like this? Follow Falkon on LinkedIn.