An interview with Volker Großmann, CTO of Actico, about the use of Decision Management Systems (DMS) as an enabler of better and faster decision-making within organizations.
Thanks to its Autumn report, “Should Your Project Use a Decision Management Suite?”, world-renowned IT consultancy Gartner has put the topic of DMS firmly on the radar of IT decision-makers. In the report, Gartner identified the specific challenges and opportunities associated with these valuable business tools.
Actico is a pioneer in the field of Decision Management Systems and is known for using artificial intelligence and machine learning to support organizations with particularly complex end-to-end decision-making requirements – i.e. in those cases for which Gartner recommends the use of DMS.
We had the opportunity to talk to Actico Co-founder and Chief Technology Officer Volker Großmann about the Gartner report and the use of Decision Management Systems within organizations and projects.
What is a Decision Management System anyway?
Let’s take companies from the financial sector (i.e. banks, insurance companies or fintechs) as an example. These organizations have to make risk decisions every day: Do we approve a specific loan or insurance policy? The answer depends on various criteria. Alternatively: Can we offer a customer a particular product within the context of the relevant regulatory and legal rules? Are there any suspicions around market abuse or money laundering in this instance?
These are all complex issues that require a lot of expertise, a lot of information and, above all, a lot of industry-specific knowledge. Capturing, mapping and automating this kind of decision-making process from start to finish – that is the “raison d’être” of a Decision Management (end-to-end decision-making) System.
What are the advantages of a DMS over experienced employees?
Depending on where it is applied, a DMS can improve the quality of decisions, help build a personalized approach to customers, or strengthen security for the company and its customers, e.g. through automated decision-making systems that detect cybercrime or fraud.
Of course, greater automation can also deliver savings, especially when a process involves multiple decisions. Examples include document processing and application reviews, especially within public sector authorities or in the manufacturing industry when monitoring machines.
In general, a DMS helps to automate decisions, improve their quality and create consistency, thereby supporting tactical and strategic decisions within organizations.
Sounds like every company should implement a DMS immediately. But does that always make sense? When is it worth using a Decision Management System?
A DMS is overkill for simple decisions that only depend on a few conditions and which don’t need to be made very often.
As Gartner rightly puts it, DMS make sense when
- A lot of complicated decisions need to be made every day, all of which depend on multiple rules and data sources.
- The decision-making logic changes frequently because legal requirements or market conditions are in flux.
- Individual business units or departments need to be able to change the decision-making rules themselves without IT having to reprogram anything.
- High transparency and traceability of decisions is needed (perhaps due to regulatory requirements).
- A high level of consistency in decisions is important. For example, customers should always receive the same information, regardless of whether they go via a call center, an employee or the website to ask about a service.
A DMS also makes a company more agile, because business units can adapt the decision-making logic themselves. This eliminates the need for a dedicated IT project and the associated complex approval processes, each time a change needs to be made.
“In particular, a DMS makes sense when there are a large number of rules, specifications and data involved that also change frequently.”
Many companies already use other standard solutions to support their decisions. Should they switch now?
There are special use cases for which standard solutions can be useful, such as marketing software that decides which offers are promoted to a particular customer. The same applies to product configurators, pricing engines and ERP systems for production planning. In cases where the decision problem can be standardized, the use of standard software makes sense, because it also adds a lot of knowledge.
But that doesn’t work for many other use cases because they are more complex – and that applies to most corporate decisions. There is often a lot of data to consider. Companies are usually well-established structures that often can’t be mapped accurately within standard solutions. Nevertheless, these organizations still want and need to digitize and automate their processes and decisions.
That is why Decision Management Systems are so exciting today, because they are suitable for making all kinds of decisions. And such systems can be used for all of these use cases – even if the starting point for their use is a very specific business problem.
We have many customers who start with one use case and then identify many more that are suitable for automating decisions. A standard solution for a specific application can clearly never do that, but a Decision Management System can.
How does a decision-making system work?
To make automated decisions, I need to know the rules on which the decision should be based. That’s why many decision-making systems are fundamentally rules-based. Humans work in exactly the same way: they analyze the facts, draw conclusions and decide on one of the options open to them. The decision-making mechanism can be thought of as a collection of tests in an “if-then” format, within the context of a process that defines the sequence of the tests.
The whole system defines any complex end-to-end decision-making process in a very systematic and understandable way. Some systems show the rules and the sequence of events in graphical form, some use rules in pure text form, and some only use decision tables. We find the graphical representation of processes, but also decision tables, i.e. a mix of all these options, to be the most useful.
The range of possible decisions are generated taking into account numerous rules and then – if there are several options – the second step is to look for the optimal decision. These optimizations are a secondary approach used within decision-making systems, in addition to the fundamental implementation of rules. If there are several options, one looks for the decision that generates the lowest possible costs, or that delivers the best result for the customer.
Gartner sees great potential for artificial intelligence here too…
Yes, there are always cases where the usefulness of a set of rules reaches its limits. For example, where a complex data set effectively makes it impossible to set up rules, or where even experienced experts can no longer explain why they make a particular decision.
Artificial intelligence and machine learning can help to make the right decisions in these instances. Machine learning means that computers can automatically build up knowledge by analyzing existing data, gain experience themselves and ultimately use that knowledge to predict the likelihood of future occurrences.
This technology offers an enormous advantage when making complex and important decisions, and can help reduce costs significantly. The opportunity to learn, continuously develop and automate the learning process itself – at least to a certain extent – is very exciting.
Will artificial intelligence automate all decision making in the near future?
No. Because regardless of the technical feasibility, decisions are not just data-driven. There are laws, regulations and industry standards that must be observed. This also applies to business strategies set by humans. For these kind of use cases, rules will always be needed to create a framework within which the AI can then act.
In addition to mapping expert knowledge, rules are also the perfect means of providing effective control for AI methods. In this context, AI can make the most of its advantages. Conversely, it’s very difficult, if not impossible, to map the data knowledge that AI develops in the form of rules. That’s why we believe the two approaches complement each other perfectly.
Does a Decision Management System also often deliver decisions that surprise experienced decision-makers, or are they usually within the bounds of what is expected?
We experience both. It depends on how much work the customer has already done. Whether you can use AI depends on what data is available, because machine learning looks for patterns in order to recognize relationships. Often there are a lot of expected results that match the intuition of the subject matter experts – which also proves how well machine learning works.
“AI can also discover decision cases in which mistakes have been made in the past.”
But there are also surprises. In one instance, we were reviewing bank transactions to identify money laundering. The AI identified the cases where the assessment of the human transaction processors was simply not credible.
This has happened several times and has also generated trust in machine learning as a result. In most cases, the operator made a different decision to the machine because he knew more, e.g. that the recipient had since passed away. This demonstrated clearly to those involved that additional data sources are always needed, and that they must be included in the decision-making process.
What becomes clear here is that using machine learning as a quality assurance measure can help ensure that decisions remain comprehensible and consistent.
What about the traceability of decisions made by AI technologies such as machine learning? Can the system explain why it made this decision?
Yes, this can happen and we refer to it as local explainability. It can explain why the AI decides in a particular way within the context of a specific case. We use the Shapley algorithm for this. It compares the current case against the average, varies the input data, and can then show which criteria the machine learning system paid particular attention to in this case, and what ultimately made the difference.
Do you get this data formulated as text?
The results are delivered as a list of percentages or in the form of bar charts. We also output the top 3 influences on a decision in text form. So you receive an explanation of what the AI reacted to with particular sensitivity. In dubious cases it turns out, for example, that the person making the assessment knew more than the machine and therefore made a different decision.
How should I think about the implementation of a DMS? Is this usually a plug-in for my existing Business Management Systems, or stand-alone software that I use in addition to them?
There are different levels of deployment. If you only use the pure decision engine, it can of course be integrated into other systems. We use common interfaces, such as connecting them via web services, to achieve this.
But there are also use cases in which the Actico platform acts as a central, standalone software solution for managing the forms and workflows across the entire decision-making process. We see a diverse range of applications in our projects. The software is often integrated into existing systems or combined with them.
Is this a cloud solution that I can easily book and scale, or does it require on-premise installation in my data center?
You can do either. At banks and insurance companies, the systems are usually operated “on-premise” in the local data center. In other industries, cloud solutions are quite common. In either case, it’s important that all the separate parts work together perfectly. We call this “end-to-end”.
How much time does it take to plan and implement a Decision Management System from Actico? What are the individual steps required?
From the technical point of view, it’s easy: the installation and integration of the technology is completed in a week. Most of the work, which of course takes time, is the mapping of the rules that govern how the system will work. Since the use of a Decision Management System is particularly valuable when there are many rules that haven’t already been mapped, this is the scenario that takes the longest. However, Excel files containing decision matrices often exist that need to be adopted and adapted. This process usually takes a few months. But you can also start small and then gradually expand the system.
One reason why our clients implement a DMS is to speed up the maintenance of the rules. Users can do this themselves rather than having to get IT involved and explain to a programmer what to do. This saves a lot of time in the long run.
What distinguishes Actico’s solutions from those of other providers?
The nice thing about our system is that you don’t have to map the rules again in your own terminology with your own vocabulary. When the data sources are connected and the system is installed, users can start building the rules straight away. We use a graphical approach that is very intuitive to use. When we demonstrate this for the first time, most business units immediately say: “Yes, I understand that and I can see myself in a position to make changes to it.” The user can then create his first test cases straightaway and immediately check whether the rule is working as it should be.
In principle, a complex hierarchical decision tree is generated from the rule input, which one goes through from left to right and from top to bottom. And an AI can be integrated at any node within the tree. The relevant data is transferred into it, and the engine then returns a forecast or assessment, which is subsequently used to make a rule-based decision. So, the decision is made by the rule and not by the AI.
How does the IT department benefit from using a DMS solution?
A DMS enables IT to delegate more responsibility to the business units, which can then make changes more quickly. This removes a burden of dealing with technical issues from the IT department. This can represent a source of tremendous value that IT delivers to the company as a whole. There is less friction and fewer misunderstandings, creating a win-win scenario for both sides.
We do sometimes see that there are reservations about allowing individual departments to run these systems themselves. It’s a question of processes, responsibility and culture. But even in these cases, the simple and intuitive interface provides a common basis that both partners can use to decide whether this is the solution that the department really wanted.
“DMS systems can lower the workload on IT and accelerate projects, because you move the management of the decision system to the business unit.”
In many cases, companies employ a technically experienced person who looks after the system and is expert from both a technical and content perspective. This usually works so well that companies gradually expand the application of the DMS to more use cases.
Featuring Forrester Analyst Mike Gualtieri: 10 Must Haves for AI-powered Digital Decisioning Platforms
Guest speaker Mike Gualtieri gives you some guidance from his latest research and customer insights.
Volker Großmann shows some use cases and explain how to combine best of humans, AI and rules for automated decisions.