For most of us, making data-driven business decisions is a four-step process. First, you collect the data. Next, you “mine” it, which just means some combination of tools and data scientists look for patterns and correlations between different kinds of data. Third, those discoveries get pumped into the dashboards and visualizations that managers get to see. From there, it’s up to the manager to interpret what the dashboard is telling them and make their decision.
The problem there is that the data you’ve collected, and the patterns your tools and data scientists have discovered, now define the decisions you can make. A simplified example: Say PCMag collects loads of data on which articles have performed the best in terms of how many clicks a particular article or group of articles has received. Then our database engines rumble to life, groups the best articles together, and builds pretty visualizations so we can understand what was found. What we’re looking at lets us see the most successful articles up to that point. We can then work to replicate that success in the future by writing more such articles on a given data pivot, like the topic, the type of article, or even the author. So what we’re doing is using our data to replicate our past successes. Certainly an effective practice.
But what if we turned that around?
Instead of limiting ourselves by the data we’ve collected, what if we simply started by asking the question we really want answered: What kinds of articles are going to do the best for us in the future? If we start there, we need a process to not only discover the questions we need to ask to get the answer, but also the data we’ll need to gather to support those queries. But what we’d get is a much more valuable set of answers with which to make our editorial decisions.
That’s one of the more exciting new methodologies emerging in the next generation of business analytics tools, and it’s called “decision intelligence” (DI). Below, we describe DI in more detail and discuss what you’ll need to know to make it work for your organization.
What Is Decision Intelligence?
Cassie Kozyrkov, chief decision scientist at Google, describes DI as a way to augment data science with social science, decision theory, and managerial science. This combination is more effective at helping people actually use BI data to make better decisions. She describes the difference between data science and DI as the difference between those who make microwave ovens and the cooks who use them.
DI grew out of software engineering efforts to build improved best-practice decisions and do so on a large scale. And according to experts, it’s matured enough that it should start impacting even small to midsize businesses (SMBs) in the next iteration of popular cloud BI tools, like Microsoft PowerBI or Tableau.
“Decision intelligence connects AI and human decision-making to form more intelligent conclusions, which lead to more favorable outcomes,” says Jack Zmudzinski, a senior associate at Future Processing, a custom software development company. “So, rather than a decision made by a human or a decision made by a computer, it’s the best of both worlds.”
Decision intelligence upends what businesses are typically doing with their data. In a big data approach, the analysis tools and the queries are typically chosen to fit the data. With DI, it’s the decision being sought that takes first priority; the query is then constructed, and the data selected by its relevance to the question. So the data takes a supporting role rather than the starring role when making data-driven decisions.
Experts define decision intelligence as a methodology, but it’s not one that has a single umbrella process. How you go about DI will be depend on your business, the data you’re gathering, and also the capabilities of your analytics tool set. However, the basic idea will always be the same: using a visual approach that begins with the required decision and then works backwards to determine what data is required and how to go about getting it.
Why You Need DI
If you think this all sounds like a big business or enterprise problem, think again. Even small businesses and “solopreneurs” will soon wield these technologies, and they’ll be able to manage them with very little effort, at a reasonable cost, and with success based on their own knowledge and talents. Bottom line: even small business leaders these days use sophisticated cloud databases that hold plenty of data and solid analytics. What they lack is any real guidance on how to use that information to make real-world decisions. That’s especially true for small companies, and it’s precisely what DI is addressing.
“Algorithms and data are good at telling us ‘Here are the observations or data and what can be concluded.’ They’re not good at telling us what decision needs to happen,” said Gopi Vikranth, an associate principal with ZS Associates, a global consulting and professional services firm. Before joining ZS in 2019, he held roles as vice president of big data and marketing analytics for Melco Resorts and Entertainment, and vice president of marketing analytics at Caesars Entertainment.
“Decision intelligence, on the other hand, answers [the question] ‘If you were to take action X, what will be the outcome in the real world?’ This is critical to businesses as there is rarely a situation with perfect information,” said Vikranth.
He describes a midsize business as a typical example. Say this company has a customer loyalty program. An AI is tasked to improve business profit, so it could mathematically find ways to change or remove customer perks or raise prices to optimize that profit. But while those conclusions are objectively “correct,” such a decision could well cause a costly backlash with customers and influencers, ultimately creating a long-term loss of loyalty and therefore revenue.
Jason Cotrell, CEO of software studio Myplanet, cites the following as potential use cases for decision intelligence:
- Personalizing software’s front-end components (adaptive UI).
- Product recommendations.
- Customer churn prevention.
- Price optimization for transaction-heavy businesses, such as airlines or pharmaceuticals.
“Instead of saying ‘What data do I need to make this decision?’ say ‘How do I make this decision? Which pieces require data analysis and which pieces can I automate?’ That way, you’ll better leverage your analytics and automation,” explains James Taylor, author of the book “Digital Decisioning: Using Decision Management to Deliver Business Impact From AI” and CEO of Decision Management Solutions.
Coming Soon: Decision Intelligence Tools
If you’re a smaller business dipping a toe into big data, your current platform may already have the tools you need to get started with DI. Experts say you may not even need that.
“You can use a DI methodology with just pencil and paper, or lately I use the Lucidspark app to collaboratively draw action-to-outcome diagrams (CDDs),” says Lorien Pratt, the inventor of transfer learning for machines, a decision intelligence pioneer, and chief scientist and cofounder at Quantellia, a machine learning and decision intelligence company. He believes that DI is maturing rapidly and that soon this type of data modeling will be available to any size business.
According to Pratt, upcoming iterations of common business intelligence platforms will support DI. “At the next level of sophistication, you’ll be able to embed DI models inside existing tools, such as [Microsoft] Excel or PowerBI,” he says. However, for less sophisticated tools, this process will likely be limited, since those users won’t be able to change their models on the fly.
But while small businesses can look forward to simplified DI platforms in the cloud, enterprises will need a lot more firepower.
“The biggest issue is an investment approach,” says Pratt, which means large businesses need to put decisions front and center. That can get complicated in large organizations. You’ll need to figure out not only what kinds of decisions need to be made, but also how to capture those requirements.
“For some decisions, the existing BI stack will be enough,” says Taylor. “But for others, [enterprises] will likely find they need to invest in more advanced technologies, like predictive analytics and machine learning tools.” Additionally, he advises that for the decisions companies need to make often or rapidly, deploying a business rules management system, like those from Agiloft or IBM, can automate the process and better leverage your machine learning algorithms.
For smaller businesses, that’s probably overkill. Especially since next-generation cloud analytics services should provide all the DI muscle SMBs need, just with fewer deployment and learning headaches. But if you still want to go DIY on DI, here’s what Taylor says a typical enterprise will need:
- Decision modeling software to do requirements gathering and data modeling.
- Business rule management software to develop your decision rules (unless you’re making relatively few decisions with allowances for long result times).
- Some kind of machine learning stack with which to develop the algorithms you’ll need.
- A data platform that’ll let you both create your algorithms and also manage transactional data delivery, preferably in real time.
- A data visualization tool, especially if your final decisions will still have a significant human element.
Patterns vs. Intuition
Machines see problems and patterns as distinctly defined: black vs white. People, on the other hand, see nuances, potential alternative meanings, options, and bridges to other thoughts. Clarity on any issue is often a function of intuitive intelligence rather than academic training. Humans can use both, and DI aims to scale those abilities.
“Decision intelligence is based on this idea of trying to incorporate realistic approaches that mimic human-like decisions,” says Ervin Sejdic, associate professor of electrical and computer engineering and intelligent systems at the University of Pittsburgh Swanson School of Engineering. “Whereas in typical AI, you basically just set up the rules for machine learning: if it’s red, it’s this, and if it’s blue, it’s that.”
Sejdic cites a car purchase as an example. “If you’re buying a car, and you set certain criteria like miles per gallon or a certain make, an algorithm would find a car for you. But we test drive the car and see how it feels and drives, and those are the soft inputs that are difficult to hard code,” he explains. “For those reasons, decision intelligence tries to encode the softer decisions we make, and those things are different from your typical AI.”
Sejdic notes that if it’s successful, DI can be applied to anything. The way to decide where to use it is to find areas where you want to know what action would be best for you to take.
“Most of analytics and business intelligence is descriptive. You record what happens and then you plot the numbers in a chart. It tells you what just happened so that you can try to understand. Other forms are predictive. They’re like weather forecasting. They tell you what’s about to happen.” explained Pathmind’s Nicholson.
“There is a quantitative and qualitative side to decision intelligence, and businesses need to take both into account,” says ZS Associates’s Vikranth. On the quantitative side, he explains, is data collection, triangulation and engineering, AI and data science infrastructure, and programming talent. The key on this side is accurate data. Businesses of any size will need to invest time and effort to define performance indicators and ensure that they’re collecting the right data and that it’s stored in a way that’s usable by their DI stack.
On the qualitative side, Vikranth says businesses need the right talent to convert insights, conclusions, or outputs into decisions and actions. These need to be contextualized for their business. What-if and test-and-learn tools can greatly help this process.
“Both sides combined is decision intelligence,” says Vikranth. “This way, AI doesn’t take humans out of the equation.” Vikranth believes this is important since mathematical algorithms, especially if they’re working on imperfect data, can’t come to any kind of optimal decision.
The Data Scientist Evolution
Database administrators of old gave rise to business intelligence analysts and traditional data science. DI is helping evolve that discipline into a new and much more effective role, namely that of data science translators. According to experts like Vikranth, these people will use DI to take a business’ what-if scenarios and work them through an AI stack to understand what actions the business needs to take and the kind of results those actions can provide.
Decision intelligence lets technology and people each do what they do best. Technologies, like analytics and AI, rapidly find the connections and patterns in huge pools of data, but that’s only half the journey. DI can use that information and help you apply the more intangible human factors, like intuitive intelligence, creativity, experience, and the ability to successfully navigate through nuances. That makes DI a powerful new hybrid analysis model even for small organizations, and one that’s especially effective at scale.
