the customer journey to decision optimization customer journ… · the decision improvement...
TRANSCRIPT
The Customer Journey to Decision Optimization
1 © 2020 Decision Management Solutions
The Customer Journey to Decision Optimization Introduction
The balance of profitability and risk is central to success in lending.
Decision Optimization can accelerate success by helping organizations
develop decision-making approaches that better meet their goals and
objectives. Let’s look at a use case that illustrates the effectiveness of
Decision Optimization at a bank that serves the consumer market.
As new lenders and new ways to borrow started to proliferate in the consumer
loan market, the head of retail credit risk at this bank knew his organization
needed to price its loans more competitively in order to reach its core market. But
the bank could not just cut prices—it had to balance competitive pricing against
risk exposure and profitability. They had to balance the whole loan portfolio and
still make loan offers attractive to individual customers. To succeed in a
competitive environment, the bank had to replace its standard policy for consumer
loan amounts and prices with something more sophisticated. Instead of just
offering better prices for better credit, they needed to find a way to consider
other factors, such as price sensitivity among prospective customers.
The retail credit risk department hired a new head of analytics who was
immediately clear on the need to innovate. He did not want to spend a lot of time
fine-tuning existing policy only to get incremental improvement. Job one was to
understand the data and processes around loan approval. This was foundational for
improvement. In particular, he focused in on analyzing decision outcomes. He
needed to see what happens when a particular kind of loan is made to a particular
type of consumer.
By James Taylor CONTENTS
Introduction
The Decision
Improvement
Lifecycle
The Beginning: Codify
Your Current Practice
The Journey: Improve
Your Decision
Strategy
The End Game:
Decision Optimization
The Value of Decision
Optimization
Decision
Optimization: Critical
Success Factors
Getting Started
Sponsored by
The Customer Journey to Decision Optimization
2 © 2020 Decision Management Solutions
This also made him realize that he needed data on loans that had not been offered
previously. This data would help him better understand the tradeoffs and
opportunities. He gathered some of this data experimentally and consulted
internal experts to fill in other blanks. Historical data, experimental results, and
expertise were all working together. Trust and transparency were critical to the
success of this process, along with multiple discussions, cross-functional teams,
and establishing overlapping goals and metrics.
Once he gained an understanding of this, he moved on to making predictions about
the customer behaviors, individual loans, and the portfolio as a whole. Now he
could predict not just how people would respond to different loan offers, but also
how those loans would play out over time and impact profit.
He then introduced the use of mathematical optimization. This provided a
framework to bring different predictions and profit calculations together,
supported scenarios to be compared and identified the best loan decisions: who to
accept, how much to lend, and what price to offer.
Initially, the new strategy was rolled out to consumer loans. As the head of
analytics began to deliver the analytic accuracy required, an iterative refinement
process began. With success came broader usage, and the approach was rolled out
to pre-approved loans and credit card applications. More complex decisions were
optimized in real time to offer consumers the best possible deal for debt
consolidation. Since this new approach was put in place, the bank rolled out more
than a dozen iterations of consumer loan strategies. And the bank also developed
four credit card strategies over four years. Each time, the bank analyzed data, ran
optimization scenarios, developed a strategy, ran it for a while to gather more
data, observed its performance, refined it, and started the process all over again.
This process of change and refinement has become embedded in the company
culture and is critical for long-term success.
What are the results of this new strategic approach? Loan size and approval rates
are up and so is acceptance. Tens of millions of dollars in new revenue were
realized in the first year, and the profitability of the portfolio has risen by over
25%.
This bank, like other organizations that successfully optimize their approach to
decision making, treats decision optimization as a journey with a repetitive
pattern. Every organization experiments, learns, and adapts to systematically
improve their decision making. Smart organizations never stop learning and
recognize that an approach that appeared to be perfect at one time may have to
be revised when market conditions change or new competitors enter the scene.
The Customer Journey to Decision Optimization
3 © 2020 Decision Management Solutions
The Decision Improvement Lifecycle
The decision improvement lifecycle is central to delivering the full value of decision
management. Being able to rapidly and effectively execute a decision strategy,
monitor its effectiveness, learn what works (and what does not), and then improve
your decision making is what makes all the difference. Each iteration moves you
toward increasingly optimal decision making and, ultimately, to true decision
optimization. Figure 1 shows a typical process:
1. Decision Performance Review: Production data about decision outcomes is
reviewed, and the results of ongoing experimentation are considered. What
metrics are unsatisfactory? What are underlying drivers?
2. Opportunities Identification: What changes can be made in the data, decision
strategy or supporting analytic models?
3. Experimentation: Updates to the strategy are proposed and simulated using
production data to observe effects of the change.
4. Promote Changes: If the results look promising and do not have unintended,
negative side effects, an approval process pushes those changes into
production.
5. Business-as-Usual Operations: Transactions and decisions are in production.
The new strategy runs for a while, often only impacting a subset of customers
and running in parallel with the old one. This creates new data about decision
outcomes that can be reviewed to start the cycle again.
Decision Management is a set of techniques and business capabilities that enable you to apply artificial
intelligence (AI) technologies—such as business rules, predictive analytics, machine learning, prescriptive
analytics, and mathematical optimization—to automate and manage the day-to-day operational decisions
that are at the heart of your business.
The Customer Journey to Decision Optimization
4 © 2020 Decision Management Solutions
Figure 1. Decision Improvement Lifecycle
The Customer Journey to Decision Optimization
5 © 2020 Decision Management Solutions
The Beginning: Codify Your Current Practice
The most common way most organizations begin this journey is by analyzing and
codifying existing practices. This starts with understanding what is done today,
what works, and what does not. A resulting “best practice” decision strategy is
developed.
Done well, this kind of analysis is still an essential first step. Codifying an existing
approach captures where you are now. Experts like to say that you must know
where you are in order to start designing something better. One of the most
effective techniques for capturing current practices is to build a decision model of
the decision you wish to improve, such as the assignment of an initial marketing
offer to a prospect or a collections strategy. Building a decision model allows for
the effective engagement of domain experts and the graphical representation of a
shared understanding of the current approach.
No matter how well modeled current practices are, they tend to have several
specific limitations.
Current practices often rely on static historical segmentations and
categorizations: Part of this strategy is to assign customers or transactions to
specific segments and categories before using these to select an action. The
static nature of these approaches makes them fragile and likely to be overtaken
by events. For example, many of the policy rules set up to exclude customers
from specific products are not regularly reviewed or challenged.
Decisions involve tradeoffs: Some decision strategies completely ignore the
issue of competing objectives and the resulting tradeoffs, focusing instead on
maximizing one metric (examples: offers accepted or overdue payments
collected). Even when tradeoffs are considered, the mechanisms used are often
not sufficient.
It’s difficult to assess the effectiveness of the current practice: Organizations
often lack complete data, making it hard to assess how the current approach
A Decision Strategy is an integrated set of decisioning artifacts—such as business rules, decision trees,
optimization, and predictive/machine learning models—that assign a specific treatment or approach
to each customer or customer segment.
A Decision Model is graphical representation of a repeatable decision that shows how that decision is
made. It represents the relationships between its sub-decisions, the data it requires to make those
decisions, and the decision-making knowledge required to make them effectively.
Decision Model and Notation (DMN) is an industry standard way to represent decision models.
The Customer Journey to Decision Optimization
6 © 2020 Decision Management Solutions
would apply to all customers, let alone what might happen if changes were
made. Biased data is also common, especially due to selection bias. Most
companies lack data about people who were not selected for consideration and
about the relative performance and results of different decisions or approaches.
A clear understanding of your current practice is essential—but it is just the
beginning.
The Journey: Improve Your Decision Strategy
Every organization’s journey to decision optimization is different. These journeys,
however, have several common elements. Each element can be adopted separately,
and there is a wide range of possible sequences. Some organizations adopt one at a
time, while others tackle several in parallel. Each of the four elements listed below
represents a concrete way to improve your decision strategy:
Decision performance monitoring
Data-driven segmentation
Experimentation framework
Predictive analytics and machine learning
Decision performance tracking
One of the advantages of formalizing your decision strategy is that it improves your
ability to track how you made decisions and how each decision worked out for you.
For instance, if you develop a decision model for your decision strategy, then you
should be able to record how each decision was made each time the model was
applied. Every customer treatment decision can be matched to decisions about risk,
segmentation, lifetime value, and so on.
You already track business performance data. The overall outcome of the decision
strategy—the customer treatment selected—is recorded for each customer, every
time. Now this data can be enhanced with an understanding of why the decision
was made the way it was made. This decision strategy detail can be linked to
business outcomes, creating a continuous improvement feedback loop.
By taking all the customers that were treated a certain way and then segregating
them into a subset based on any of the sub-decisions in your model or strategy, you
can focus on very specific subsets of customers. Seeing how well the treatment
worked for those subsets often identifies obvious improvements. New, more fine-
grained approaches commonly result from this analysis.
Data-driven segmentation
Most decision strategies involve categorizing or segmenting customers so that a
suitable treatment can be selected for customers in that segment. Most
The Customer Journey to Decision Optimization
7 © 2020 Decision Management Solutions
organizations have some rule-of-thumb categories and historical segments that are
baked into the first version of their decision strategy. One of the first steps you can
take to improve your decision strategy is to use your data to improve this
segmentation.
Each element in your strategy can be assessed. Taking the logic in these
segmentation decisions, review your historical data. How alike are customers in the
segment? Is there a clear and distinct difference between the behavior of
customers in one segment versus another? These assessments can be done using
standard business intelligence and guided analytic tools. It’s even more effective to
use analytical techniques to develop data-driven decision trees.
The purpose of the segmentation needs to be clear. For example, in a lending
strategy, one segment could consist of customers who have good credit but are
reluctant to use it. A data-mining or data-driven decision tree tool can be used to
analyze your data and create a set of precise criteria that better differentiates
customers. This might match your expectations, but it is likely to both refine and
occasionally challenge your approach.
Before you integrate this new segmentation into your decision strategy, you can
simulate the impact of the change. Build a version of your decision strategy using
the new, data-driven approach, and run it against a large historical data set.
Compare what you did then with what the new strategy would suggest and analyze
which customers would receive a different treatment. Use the differences to see
how much more value the new approach might have. Invest the time to ensure that
everyone is comfortable with the new approach and agrees that it is more
effective. If it is not completely clear that the new approach is better, consider
using an experimental comparison (see below).
Because your strategy is now based on analysis of data, you need to establish a
rhythm of updates. Data continues to be gathered, so the patterns in your data will
change continually. Regular reviews of your data and an assessment of your data-
driven segmentation is essential to ensure that you do not get out of sync with your
data and, therefore, with the actual behavior of your customers.
Experimentation framework
There are often advantages in assessing different decision-making approaches on
the same subset of customers, so you can identify the relative impact of changes
and which approach works best. Static analysis of the different approaches can
highlight the differences between them. Simulation with historical or synthetic data
can show the likely difference in outcomes. But there is no substitute for real-world
experience. The only way to get good data about what people will do when treated
a certain way is to treat them that way and observe their response.
Running one experiment after another for a period and collecting data about how
that works out works to some extent. In practice, the delay between different
experiments often takes too long and runs the risk of being overtaken by events,
The Customer Journey to Decision Optimization
8 © 2020 Decision Management Solutions
such as a change in market conditions which could render any performance
comparison useless. Creating an experimentation framework to run several
experiments in parallel on random samples is much more effective, and easier to
measure.
Within a decision strategy, you can identify specific elements as the subject of
experiments. Several different ways of making those sub-decisions can then be
defined and used to support various experimental designs. Customers are randomly
allocated to each of several approaches.
Running multiple experiments within the decision strategy itself allows experiments
to be created, tested, evaluated, updated, and retired—without changes to any of
the company’s other systems, as the experiment is completely encapsulated within
the decision-making component.
Predictive analytics and machine learning
Analyzing data to create data-driven segmentation is a powerful tool for improving
decision strategies. Historical data can also be analyzed to make predictions about
likely behavior or the likely impact of a decision. Using predictive analytic
techniques and machine learning, data about past behavior can be turned into
insights that you can use to improve decision making.
For instance, business experts might see that they could change their decision-
making approach if they knew which customers would use the credit they were
offered, buy additional products, default on an existing debt, or promise to pay but
then not stick to it. While machine learning cannot give definitive answers to these
kinds of questions, it can provide probabilities. Often, the decision strategy can
integrate these probabilities into decision making, changing the approach based on
the likelihood of a risk or opportunity.
A decision model or other representation of a decision strategy can be used to
discuss the potential value of such predictions. Business experts can evaluate which
Companies are often reluctant to experiment too widely. Champion/Challenger testing is used so that
most customers get the preferred, or Champion approach, whilst smaller randomly assigned groups are
experimented on—Challengers. The random allocation of customers to different approaches is managed to
ensure statistically significant distributions to all the experiments. Once the experiment has been run
long enough to collect useful data about outcomes, you can compare the differences in outcomes.
Experimental Design is a set of principles and techniques to ensure that the setup and execution of
experiments meets its objectives. Good experimental design allows multiple hypotheses to be tested at
once, helps ensure proper randomization, and supports replication. The experiments might be somewhat
random to fill in gaps in data required for analysis, or they can be more systematic, comparing the
effectiveness of well-defined approaches.
The Customer Journey to Decision Optimization
9 © 2020 Decision Management Solutions
aspects of the decision strategy they might change based on a specific prediction
and then offer guidance on the required level of accuracy. Data scientists can then
analyze available internal data—and perhaps data from outside the company—to
build a predictive model that will integrate into the decision-making.
It is generally best to evaluate new strategies utilizing predictions as experiments
first and compare the results achieved to decision strategies that do not use them.
Data bias and selection bias need to be considered. Companies have data about the
kinds of customers they have historically selected and not about those they have
historically rejected. This needs to be considered when analyzing data to make
better, more informed predictions that impact decisions.
Useful Tool: Influence Diagrams
No matter which of the above techniques you utilize, a good starting point is to build an influence
diagram focused on the decision in question, to see how the various elements fit together. An influence
diagram is an intuitive visual display of the key elements of a decision problem, illustrating the influences
among them as arrows. Often undertaken as a whiteboard exercise, just drawing out an influence diagram
of your business problem can be a great help in better understanding the key profit drivers and how they
are connected. If you can predict or measure the outcome of these drivers for different potential
decisions, you can start to make better decisions.
Figure 2 shows that three types of data (application, credit bureau, and customer data) are used to make
three key decisions: who to accept, how much to loan them, and what rate to charge them. How those
decisions are made affects some critical measures, like take-up rate (Are the people who are lent money
spending it?), charge-off (Do they fail to pay it back?) and early pay-off rates.
Figure 2. Influence Diagram
These elements can materially
improve your decision strategy.
Eventually your strategy will have
multiple drivers of profitability,
several key predictions about
customer behavior or risk, and
multiple experiments running in
parallel. The trade-offs between
these may not be clear—and
manually choosing among them will
lead to sub-optimal results. At this
point, it’s time to think about
optimizing your decision strategy
mathematically.
The Customer Journey to Decision Optimization
10 © 2020 Decision Management Solutions
The End Game: Decision Optimization
Decision Optimization applies mathematical processes to identify the best decision
strategy, given an organization’s data, constraints, and objectives. Most
organizations have multiple constraints on how they make decisions based on
regulatory frameworks, budgets or policies. They also have competing objectives,
such as profitability, attrition, and bad debt. By working through the incremental
improvements described above, you can generate a wide range of data about your
customers and the impact of decisions. Decision Optimization consumes all this and
finds the best possible decision strategy.
Data
To optimize a decision, you ideally need to have data about all the possibilities.
The first step is to ensure that the journey so far has filled in the data you need.
Mathematical Optimization uses mathematical formulas to identify the best or most plausible solution to
a complex problem. Optimization generally maximizes or minimizes something, such as profit, while
considering various inputs and constraints on the possible solutions. The intent is to find an optimal
answer to a well-defined problem.
The Customer Journey to Decision Optimization
11 © 2020 Decision Management Solutions
You may need to run experiments to deliberately offer products or prices to those
not normally eligible to see what happens. You will almost certainly need to rely on
experts to extrapolate and fill in some of your blanks. Tracking the confidence of
your data in various areas will help you focus optimization where the data is robust
enough to get a good outcome.
Decision Impact Models
Your understanding of the elements of the decision (shown in Figure 2) now needs
to be modeled more formally. To get started building such a model, your influence
diagram can be extended into a more detailed and precise model, detailing the
inputs, decisions, component models, calculations and the influences between each
element. FICO, for instance, use Decision Impact Models to provide the framework
for a model specification, such as that shown in Figure 5.
It’s important to ensure that the predicted outcomes of each decision are as
accurate as possible. You may need to build prescriptive, causal or action-effect
models that consider the impact of different decisions. Such models make
predictions about the likely effect of each action by analyzing your historical data
to find the things that determine who does what in response. For instance, you
need to predict who will take up to a loan offer and drawdown the loan. These
models are a mathematical representation of your understanding of how customers
respond to your decisions, as illustrated in Figure 3.
Some of the actions customers take are variable, and the variability also needs to
be predicted. For example, you can predict that a customer will stop paying for a
loan before it is paid off. But to calculate profitability, you will need to know how
Figure 3. Action Effect Models
The Customer Journey to Decision Optimization
12 © 2020 Decision Management Solutions
many payments they are likely to make before they stop. These additional
predictive analytic models predict key measures or metrics about a customer or
account in specific circumstances and are combined with the action-effect models
that predict how likely those circumstances are.
You also need to ensure you have considered all the different outcomes in the
decision problem. For example, the potential outcomes to a Loan Origination
decision are shown in Figure 4.
When the customer applies, you can accept or reject them. For accepted
applications there are three possible outcomes:
1. Customer accepts the loan but does not use it – there is no take-up.
2. Customer use the loan but then defaults – they fail to make all the scheduled
payments.
3. Customer uses the loan and then pays if off completely.
Each outcome needs to be considered. The probability of each outcome can be
calculated using the model in Figure 4.
1. The probability of Non-Take up is 1-a, the inverse of the probability of the
customer taking up the loan.
2. The probability that the customer will take up the loan and then default is a * b
– the likelihood of take-up multiplied by the probability of default.
3. The probability of the loan being taken up and then being paid off completely is
a*(1-b).
Because these represent all the possible outcomes, the sum of these probabilities –
(1-a) + (a*b) + a*(1-b) should be 1.
Models such as FICO’s
Decision Impact
Model tie all this
together with the
calculations of your
metrics and
objectives. The
resulting combined
model shows how
likely the possible
customer responses
are for a given
decision and what
the result of those
responses will be on
metrics and
Figure 4. Decision Outcomes
The Customer Journey to Decision Optimization
13 © 2020 Decision Management Solutions
objectives. Fixed costs can be included too, so that profit and loss can be
accurately modeled. A mathematical model has been developed to predict the
value of each possible decision.
Decision Impact Models can be executed against your customer portfolio. Decisions
can be compared, applying two different decision strategies to each customer and
then calculating the value of each decision-making approach. More interestingly,
though, the models can be used to optimize the decision strategy using a solver.
Optimization
The possible decision strategies are fed into the model, along with portfolio-level
constraints, such as total amount of credit available or bad debt tolerance. The
constraints limit the available strategies at both the customer and portfolio level.
The objectives are used to give the solver an objective function. The solver
compares possible outcomes for each customer to find the best possible decision for
each customer, given your objectives and your constraints. These results can be
used directly to treat each customer or can be analyzed to come up with an
overarching decision strategy that selects the best of the available decisions for
each customer.
To support mathematical optimization, the relationships shown in a Decision Impact
Model are expressed as formulas that can be calculated and variables available to
the solver. The example in Figure 6 below shows a section of the profit calculation
for the example and includes formulas for probabilities, costs, losses etc. The
Figure 5. Decision Impact Model for Loan Origination
The Customer Journey to Decision Optimization
14 © 2020 Decision Management Solutions
formulas can be used in multiple scenarios. These can be simulated with different
constraints or decision strategies. These scenarios can be combined to find the best
way to manage within your current constraints and show the value and tradeoffs of
changing those constraints.
It should be noted that getting value from Decision Optimization does not require
an immediate jump to a fully realized solution. Several simplifications can be made
to get started quickly while still benefitting from decision optimization. You can
begin at the segment level. Because you’re not trying to assess behavior at the
individual level, there’s less need for action-effect models at this stage. You can
also consider just the initial decision and have generic calculations for subsequent
predictions and outcomes like offer acceptance and usage. A simpler decision
impact model design can be used, such as that in Figure 7 below.
Moving on to consumer-level assessments, you can minimize complexity by defining
less sophisticated action-effect models and by considering fewer drivers of
profitability or success. Being consumer-focused and not segment-focused will
improve accuracy, even if you do not add all the possible details to action-effect
models.
These additional models and the use of optimization enhance but do not change the
decision improvement lifecycle. As Figure 8 below shows, the added sophistication
and precision enhances how you improve decisions in your decision improvement
lifecycle.
Figure 6. Profit Calculation
Solvers are mathematical software tools that take problem descriptions in a mathematical form and
calculate the best and most practical solution to that problem. Business problems are turned into a
mathematical representation defining objectives or business goals, variables, and constraints. Solvers use
various techniques to find the best solution out of a potentially large number of possible solutions.
The Customer Journey to Decision Optimization
15 © 2020 Decision Management Solutions
Specifically:
Influence diagrams enhance stage
two—opportunity identification—by
making it clearer what improvements
will be beneficial.
Action Effect models can also bring a
lot more focus and accuracy to the
Impact Analysis in stage 2.
Decision Impact Models and
Optimization can be added to stage
three, finding the best outcome
among those being considered.
Optimization also allows many more
simulation attempts to be
considered, mathematically picking
between all the simulated scenarios.
Figure 7. Basic Loan Origination Decision Impact Model
Figure 8. Decision Optimization and Improvement Lifecycle
The Customer Journey to Decision Optimization
16 © 2020 Decision Management Solutions
The Value of Decision Optimization
Adopting Decision Optimization can bring many business benefits. In summary, key
areas highlighted by users include:
Focuses on the outcomes of different decisions, the things that truly impact
business success.
Identifies decisions that meet your goals and constraints – whether its
meeting a regulation our making the most out of scarce resources.
Clarifies the business problem as being able to quickly compare scenarios
identifies the key profit drivers, shows how to respond to potential threats, and
reveals opportunities.
Generates a significant return on Investment. Published case studies show
annual profit improvements of 5% to 30% or more, with ROIs of over 10:1. these
are the kinds of numbers that can transform a business.
Decision Optimization: Critical Success Factors
Decision Optimization is a powerful technique that builds on your work in decision
management, experimentation, machine learning, and data quality. Here are some
critical success factors shared by customers who have had positive experiences.
Be willing to experiment: You will need to experiment to fill the gaps in your data. Your optimal decision strategies may be quite different from your current approach. A willingness to experiment and run experiments in production will help people feel comfortable that the new approach is both better and safe.
A mixed team brings together a wide range of knowledge and skills: Besides data science and technical skills, you will need domain expertise. Domain experts are essential to creating your initial strategies, designing meaningful experiments, and reviewing business impacts.
Decision optimization requires flexible and powerful software: You will need tools that handle rules, machine learning, and optimization in a coherent way. These tools need to be integrated with the software you currently use to deploy decision services into production.
Use a proven methodology, ideally one based on models, not requirements documents: Decision models built using standard notations like DMN and decision-impact models that show how the pieces fit together are easier to validate and for business owners to accept.
Finally, if you are inexperienced in this realm, get advice: Seek some expert guidance and support, especially for experimental design and some of the more advanced analytic models.
The Customer Journey to Decision Optimization
17 © 2020 Decision Management Solutions
Decision Optimization: Where Next?
Many approaches and techniques for decision optimization have been tried and tested over recent years.
Several recent developments offer the potential to significantly improve the development, management,
and improvement of decision strategies.
Enhanced machine learning and artificial intelligence models
Machine learning and artificial intelligence techniques are evolving to deliver more accurate predictions
more rapidly. These predictions can be used to improve decision optimization, though risks around data
bias, overfitting, and bad model design must be mitigated. An increasing number of these developments
focus on causal inference. These techniques answer the question of what would happen if we did X or Y or
Z? Predicting this is fundamental when calculating the outcomes of different potential decisions. Such
techniques may reduce the need for controlled experiments and data collection.
Global tree optimization
Decision trees are used extensively in Decision Management. Standard decision tree algorithms assess the
predictive power of different variable splits at each level of the tree, one node at a time. This can lead to
overfitting and locally optimal splits. Research by institutions such as MIT, Microsoft, IBM, and FICO is
looking into ways to create the entire decision tree at once and achieve a globally optimal result. These
will result in smaller, easier-to-use decision trees that have more predictive power and that can be easily
assessed and understood by business reviewers.
Automated business learning
Decision Management generally relies on specialists who understand how to design, develop, configure,
and manage the various business rules, analytic models, and optimization components. User demand and
new technology increasingly allow business domain experts to manage these components. Tools are
evolving that allow domain experts to define their goals and objectives and automatically consider
alternative scenarios for achieving them. These solutions seamlessly combine machine learning,
decisioning, and optimization technologies. A high degree of automation ensures key models and
components are kept updated.
The Customer Journey to Decision Optimization
Getting Started
It’s time to embark on your own journey to decision optimization. These three
steps will get you on the right path.
1. Begin with the decision in mind: Be explicit about the decisions you want to
improve or optimize. Identify, understand, and model your current approach to
making these decisions. Put in place the decision management technology you
need to manage these decisions.
2. Work with your business partners to ask “if only”: Explicitly define these
conditions: “if only we could predict this,” “if only we knew how customers
would respond,” “if only we could balance these conflicting objectives.” This
will help you understand what is preventing you from making better decisions
and will guide the data you collect and where you need to focus your energies.
3. Plan for the journey at the start: If you want to optimize decisions, make that
your objective. Find the help and support you will need for the whole journey,
not just for the first step. You need a goal for your project that is compelling
enough to get people excited.
To help you get started, the following page has a template for designing Decision
Impact Models. List your inputs, decisions, predictions, outcomes, and objectives.
Then think through the connections between these elements so you can develop a
more effective, more optimal decision strategy.
CONTACT US
Decision Management Solutions helps large organizations harness data-driven decisions by applying
decision management, business rules, and advanced analytics to solve their most pressing business
challenges.
www.decisionmanagementsolutions.com Email : [email protected]