In short, decision intelligence flips the traditional data mining process so that the action dictates the data rather than the data recommends the action. This technique moves companies away from the previous data driven model to a decision driven model. The reasoning behind this approach is to stop high fail rates in AI and data driven projects by predetermining the outcome and working toward it rather than trying to milk value from a completed and cost-spent process.
The principles of decision intelligence
The following principles apply to decision intelligence:
- Determine your options.
- Use decision theory methods to analyze the likely outcomes of each decision.
- Compare the expected results and choose the optimal outcome — make the decision, in other words.
- Work backward from the decision to determine the processes needed to make the decision a reality.
- Determine whether a team is needed and, if so, which disciplines are required in the team mix.
- Use decision sciences to form, map, and evaluate the processes to be used.
- Determine the tools needed to implement the processes.
- Determine the data needed to inform and feed the processes.
The decision intelligence thinking process
These steps, with feedback loops and process testing throughout, comprise decision intelligence:
- Understand the problem but ignore the scope. Decision intelligence can be used to solve the simplest of problems or the most complex; therefore, the scope of the problem doesn’t matter at this point. Focus on understanding the substance of the problem and do whatever research you need to grasp it fully.
- Frame the decision. Begin by defining the limits of the decision to be made according to your will, resources, and capabilities. If you will consider only one action, stop here — you have already made your decision.
- Consider options that fit the decision framing. For example, if your company clearly lacks the will or the budget to implement one or more decision options, strike those from your list, because there’s no point in pursuing those options.
- Weigh the options. Determine the likely costs and outcomes of each option. Use decision theory via an appropriate tool, which can be anything in the decision-making toolbox, from a simple pros-versus-cons list or SWOT table to a formula established by one of the decision sciences or a complex algorithm.
- Choose the preferred outcome. The preferred outcome may or may not be optimal or even positive. After all, some decisions are of a rock-versus-hard-place nature, where no truly good outcomes are available to you. It is, however, presumed that the decision-maker is rational and is rationally deciding on the preferred outcome. This is the decision from which all work in the decision intelligence framework will spring.
- Work backward. Once you’ve made a decision, work backward from that decision (the end) to the data (the start) to map the best path to manifestation. After you have established the needed processes, determined the tools you need, and selected the relevant data to feed the processes, you will have effectively mapped the path you need to follow and the actions you need to take to realize the impact you’ve decided on already.
- Test along the way. Make sure you’re on course by testing your DI processes regularly. Adjust as needed to remain on the path to realizing the business value and to avoid scope creep.
AI and non-AI takes
Decision intelligence is emerging as a top solution to high fail rates in AI projects. Investments in AI tend to be significant, so poor results are untenable. When you make a decision first on the specific business impact that’s sought, AI can be trained to produce that specific result.
Because it’s difficult to adjust AI models, retraining is usually necessary to get the AI to adjust to changes or to pursue a new objective. Framing the AI model and objective with a predetermined decision on a desired outcome means that every AI project is linked to a discernable and measurable business value. This is a far more proactive method to control AI resources and success rates.
You should realize that AI is automated decision-making at scale and so decision theory is typically already in use. However, decision intelligence requires that decision theory be applied in a more forward and different manner. The same is true of other, more traditional forms of automated, digital decisioning.
Multiple AIs or associated automations of another form, including robotic process automation (RPA), also benefit from setting the business impact target at the outset, because this strategy ensures that microdecisions and serial decisions are aligned to the end decision, also known as the bigger picture. The goal remains throughout to unfailingly produce measurable business value.
Despite much of the buzz surrounding decision intelligence centering on AI and automation, these are but some of the tools in the larger decision intelligence toolbox. The nature of the decision dictates the tools, so AI isn’t a given in the execution of any single decision. A simple decision may require only a pencil and the back of a napkin to complete. Conversely, a highly specialized decision like developing a COVID-19 vaccine required bioinformatics to implement it, given that AI isn’t sufficiently specialized in biogenetic analyses.
How to counter objections to decision intelligence
Despite the clear benefits of using the decision intelligence approach, resistance to it does exist. Points of resistance include — but are not necessarily limited to — inertia, resistance to change, a strong preference for math-only models, and the lack of company will to implement the approach. To overcome resistance, try one or more of these suggestions:
- Determine whether the organization has the will and resources to implement the optimal decision before proceeding to build the decision intelligence processes. It may necessitate choosing the next optional decision to ensure that the work comes to fruition.
- Ensure that the decision intelligence project team is multidisciplinary rather than single-disciplinary. In other words, you need more than data scientists on your team — you need decision chefs as well.
- Explain the principles of decision intelligence, and make it clear that just about everyone, including members of your audience, use them intuitively and have done so for years. This isn’t a change so much as perfecting something people already do (at least at some level).
- Suggest beginning with decision intelligence on the next AI retraining date on the existing schedule, at the start of a new project (whether AI is involved or not) to avoid resistance to change after work on a current project is already in progress.
