Incedo LighthouseTM for operational decision automation
                     s
Harness the power of 
Data using Incedo 
LighthouseTM for 
Operational decision 
Automation
WHITEPAPER
Whitepaper  |  Harness the Power of Data Using Incedo LighthouseTM for Operational Decision Automation
If you are a business l Am I focusing on the right problems and 
solving issues that will make a material 
executive in charge of difference to business performance?
making complex 
l Am I tracking and monitoring the impact these 
operational decisions - actions are creating on key business metrics, 
am I getting systematic feedback to reinforce 
be it in marketing, actions?
operations, customer 
l Am I  evaluating mult iple corrective / 
service or risk corroborative action alternatives using data-
management - a driven insights?
significant part of your 
In today’s VUCA (Volatile, Uncertain, Complex and 
mindshare is taken up Ambiguous) business environment, the above 
by the questions: questions recur with ever-increasing frequency 
and depth.
Every critical-thinking business executive knows 
that it is not enough to rely on pre-defined rules to 
make decisions in today’s world. Consider this - 
cross-sell and up-sell can become cornerstones of 
a bank’s growth strategy given that it could be 
more than 5 times cheaper to sell to an existing 
customer than onboard a new customer. Also, 
deeper penetration into customers means lower 
attrition as well. Well-run, mid-sized as well as large 
banks have sold an average 6 products per 
customer. So, if you are a bank whose cross-sell 
ratio is lower than this, then you need to dig deep 
into the wealth of demographic, behavioral and 
transactional data to identify pockets of business 
that need to pull up their socks. And this you would 
need to do on a frequent basis e.g. every month, as 
business moves fast. So, rather than a one-time 
effort of identifying leads/ targets using static 
rules, you need to think of an intelligent system 
that provides far more effective campaign design 
inputs dynamically.
Decision makers are increasingly required to make 
decisions in a dynamic environment under the 
following circumstances: 
02
Uncertainty:
The situation evolves with time and it is difficult to define business rules a priori. For example, 
Covid-19, the dynamic nature in which the pandemic spread and the varying response 
strategies at state/country/city levels have created unprecedented uncertainty in consumer 
behavior for banks.
Contextual:
Increasingly, external factors have a significant impact on how decision makers need to respond 
to situations. For example, interventions by the Federal government (Covid-19 relief loans) and 
Federal Reserve (reduction in interest rates) drive consumer loan repayment behavior and in 
turn, impact focus areas and actions taken by Collections teams in Retail Operations for banks.
Iterative:
Decisions are no longer limited to linear cause-and-effect scenarios. Instead, decisions have to 
be made in an iterative manner in a ‘test and learn’ mindset. For example, as fintechs continue to 
disintermediate the banking business, consumer banking teams are being forced to switch to 
experimenting aggressively with product/service variants (read bundles, pricing strategies etc.).
There is clearly an opportunity to provide Decision Automation Solutions that can leverage 
AI/ML technologies to augment the human decision-making process. We believe that Data and 
AI/ML can be leveraged to develop the next generation of Decision Automation Solutions. 
Whitepaper  |  Harness the Power of Data Using Incedo LighthouseTM for Operational Decision Automation 03
Whitepaper  |  Harness the Power of Data Using Incedo LighthouseTM for Operational Decision Automation
Our platform Incedo LighthouseTM has been designed and 
developed from the ground up with the following 
objectives:
Distill signal from noise: 
Identify the right problem areas to focus on by organizing the metrics (KPIs) as a hierarchy from 
lagging to leading metrics and then applying the Autonomous Monitoring and Issue Detection 
algorithms to identify anomalies that need to be addressed in a targeted manner. Thereby, 
identifying efficiently crucial problem areas, the business should focus its energy on 
voluminous datasets that are updated at short intervals (typically daily).
Leverage context: 
Apply Intelligent Root Cause Analysis algorithms to identify the underlying behavioral factors 
through specific micro-cohorts. This enables action recommendations that are tailored to 
specific cohorts as opposed to generic actions.
Impact feedback loop:
Evaluate action alternatives with controlled experiments to determine the most effective 
actions - and use that learning to iteratively improve outcomes from the decisions.
04
Autonomous Monitoring and Issue 
Detection
Metric monitoring with KPI trees ML 
models to flag anomalous behavior
Incedo
LighthouseTM
Operationalize Actions and Intelligent Root Cause Analysis 
Monitor Impact and Action Recommendations
l Evaluate action alternatives with l ML models to identify behavioral 
controlled experiments drivers
l Integrate recommendations with l Integrate ML based targeted AI 
operating workflows recommendations
Whitepaper  |  Harness the Power of Data Using Incedo LighthouseTM for Operational Decision Automation 05
Whitepaper  |  Harness the Power of Data Using Incedo LighthouseTM for Operational Decision Automation
Autonomous Monitoring & Issue 
Detection
l Start with strategic objectives and a set of CXO 
level business KPIs that measure the health of a 
business. 
l Plot inter-relations among KPIs in the form of a 
hierarchical tree structure, and have a way to 
auto-calculate the values at every data refresh. 
Typically, the KPIs at / towards the root of the 
tree are lagging indicators of the business. 
l Identify ‘nodes of concern’ in the KPI tree 
autonomously, leveraging anomaly detection 
algorithms that run on the data underlying the 
KPIs.
Intelligent Root Cause Analysis and 
Action Recommendations
l Determine root causes and identify drivers that 
explain the movement and shifts in KPIs using 
rule-based and ML-based algorithms.
l Finalize improvement actions at individual 
level (account, customer, product) that would 
materially improve target KPIs by evaluating 
potential options using Experimentation 
implementing testing approaches such A/B, 
Test / Control and Pre / Post in an iterative 
manner.
Operationalize Actions & Monitor 
Impact
l Push selected actions to the Systems of 
Execution (SOE) at client’s end that would 
implement actions on the ground.
l Continually receive feedback from SOE to track 
the success of implemented actions via 
autonomous monitoring, again leveraging 
anomaly detection algorithms, and course-
correct on the go.
06
Incedo LighthouseTM implements this virtuous cycle using a 6-step workflow that encodes the 
Data -> Insights -> Actions journey:
Overall Business Objective & Problem Statement
KPI Decomposition
Execute, Monitor & Track Identification of the Most 
Performance of Actions Important Problem Areas 
Define Experiments to Test Root Cause Analysis 
Action Recommendations to Establish Cause
Define Targeted 
Recommended Actions
Whitepaper  |  Harness the Power of Data Using Incedo LighthouseTM for Operational Decision Automation 07
Whitepaper  |  Harness the Power of Data Using Incedo LighthouseTM for Operational Decision Automation
Step 1: KPI Decomposition
In this step, the overall business objective is converted into a specific problem statement 
which articulates Key Performance Indicators (KPIs) that are tracked at the CXO level. These 
uber level KPIs are typically lagging (i.e. resulting or outcome) indicators. These are useful to 
express the business problem at hand, have senior leadership level visibility and are monitored 
/ tracked closely by business analysts. But, these are not enough to help operational level 
managers to know the status at their span level in closer to real time. For that the lagging 
indicators have to be broken successively into their component level indicators / factors till we 
reach a level that is not possible to be broken further or we reach a level at which a significant 
number of operational-level decisions are made. 
The result of this iterative process, of breaking the higher level KPIs down to their component 
level KPIs successively, leads to the formulation of KPI trees. The lower nodes of the KPI trees 
typically consist of leading (i.e. operational/process level or input) indicators that can be 
directly influenced through better operational decisions. The KPI trees formulated this way 
help address the original business objective.
Incedo LighthouseTM provides the below features that aid the KPI Decomposition step:
l Self-service: An easy-to-use, self-serve UI lets users define and configure KPI trees by 
themselves, with minimal to no dependence on IT, to structure a business problem into 
the language of KPIs e.g. cross-sell percentage via digital channels.
l KPI repository: Functionality to contribute to and leverage a shareable repository of KPIs 
that are available to several business users from multiple departments that way baking in 
consistency of KPI definitions and adding efficiency to the process e.g. recovery rate 
(collections).
08
l Automation of calculations: Data pipelines that establish data connectivity with the data 
sources and automate KPI tree refresh process e.g. daily or periodic refresh of full set of KPI 
trees to provide updated insights to business users to aid in operational decision-making.
Step 2: Identification of the Most Important Problem Areas
The KPI tree created in the previous step consists of several nodes that are connected logically 
in an inverted tree-like structure. The root of the tree is a single node that then breaks down 
into multiple nodes which are its children and each child breaks down into its own children. 
The nodes at the end of the structure are leaf nodes. Typically, the leaf nodes of the tree extract 
the data from the analytical data warehouse or other sources using SQL queries. And the 
parent nodes utilize the built-in, aggregation logic to calculate the values at their level.
While the KPI tree provides an overall numbers-driven picture of the performance of the 
business that could be as latest as yesterday (or even intra-day, data refresh cycle permitting), 
it still lacks critical  identification of specific areas of performance dip that need to be 
addressed immediately. 
Incedo LighthouseTM employs anomaly detection algorithms for that purpose, whereby the 
timeseries data of each of the metrics at each of the nodes of the KPI trees are analyzed in 
order to understand if the performance dip observed, especially in the last data update, is a 
significant deviation from the trend - and therefore a cause of worry, or it is a temporary blip 
and should only be observed further in the future.
Whitepaper  |  Harness the Power of Data Using Incedo LighthouseTM for Operational Decision Automation 09
Whitepaper  |  Harness the Power of Data Using Incedo LighthouseTM for Operational Decision Automation
Incedo LighthouseTM provides the below features 
that aid in identifying most important problem 
areas in the KPI trees:
l Timeseries Anomaly Detection: Machine 
L ea r n i n g  a l g o r i t h m s  to  m o n i to r  K PI 
movements across t imelines and flag 
exceptions e.g. if yesterday's outbound call rate 
is anomalously low or just a one-off blip.
l Non-t imeseries Anomaly Detection: 
Machine learning algorithms to identify the 
accounts, transactions or products that stand-
out from other clusters in a significant manner, 
when compared on key dimensions or 
features, so as to design a custom treatment 
for them e.g. the accounts / customers that 
have significantly reduced their purchase or 
transaction volumes when compared to their 
peers (same level of affluence, same number of 
years with the bank etc.)
l RAG status of Nodes: Depending on severity, 
frequency and recency of anomalies in its 
metrics, the individual nodes are indicated in 
R/A/G to draw immediate attention. 
Step 3: Root Cause Analysis to Establish 
Cause
Once the nodes that are critical in nature from a 
performance standpoint are identified, the next 
step is to identify the root causes of the dip in 
performance. If we were to do a new analysis 
everytime i.e. every day or every period for this then 
it would be highly inefficient and by the time we 
are done, the business and associated problems 
would have moved on. We, therefore, need to 
‘package’ this root cause analysis into a set of pre-
defined and pre-configured model sets. Also, this 
is to be done while keeping in mind the multi-
faceted nature of the problems at hand.
Incedo LighthouseTM accomplishes this using the 
below pre-packaged configurable model sets, the 
output of which is presented in a format that is 
c o n d u c i v e  f o r  t h e  n e x t  s t e p  a c t i o n 
recommendations.
10
Whitepaper  |  Harness the Power of Data Using Incedo LighthouseTM for Operational Decision Automation
l Cohort Analyzer: Cohort Analyzer attempts to establish the sections of customers, 
accounts or transactions that are highly likely to be responsible for the change in a key 
metric represented on the nodes of the said KPI tree. This is done by letting the user define 
various cohort archetypes - rule based, decision tree based or activity path based. Each 
cohort archetype can have multiple cohort types e.g. for a banking use case related to 
identifying responsible customer groups that are pulling down collection rates, the cohort 
archetype of a decision tree could have two cohort types such as ‘demographics based’ and 
‘transactions based’. Each of them could then contain multiple cohorts which are nothing 
but the customer groups with each group responsible partially for the change in key metric 
in a decreasing order of contribution. By automating the identification of cohorts (micro-
segments) that drive KPI movements, it provides significant insights into the root causes of 
performance  issues.
l Clustering and Segmentation: Further to uncovering the cohorts, the functionality is 
provided to create custom clustering and segmentation models that provide additional 
insights that indicate natural groupings in the population (accounts, customers, 
transactions). This further describes the cohorts on behavioral dimensions that are arrived 
at using feature engineering, with the advantage of a ready workflow setup in the platform 
for hosting and using all the analytical artifacts together in the Data Science Workbench of 
the platform.
Step 4: Define Targeted, Recommended Actions
If the workflow stops at only delivering the insights using anomaly detection and root cause 
analyzer, however sophisticated the algorithms are, it would still be a lost cause if the executives 
are not supported on taking the corrective, preventive or corroborative actions based on the 
insights delivered.
11
Therefore, Incedo LighthouseTM incorporates the Action Recommendation module that 
enables the actions to be created at each individual cohort level for a targeted corrective or 
improvement treatment based on its individual nuances. 
The Action Recommendation module helps define and answer for each cohort:
l What: What is the right action to implement? E.g. define & deploy a targeted contact 
strategy, define & deploy a loss-mitigation strategy, define & deploy a marketing strategy.
l Who: Who needs to be targeted? E.g. specific cohort or a group of cohorts.
l When: How long should the action be executed? E.g. defining a period to run the targeted 
high frequency contact strategy to increase collection rate as Jan 1 to Jun 30.
l Goal: What is the expected outcome? E.g. define the focus KPI (from the KPI tree) and 
target improvement desired in that KPI from the action undertaken.
The actions can be designed & defined by executives based on their expert knowledge of the 
domain and the context or those can be recommended by the system leveraging the 
historical data by employing machine learning for pattern recognition, identifying similarities 
between the under-consideration cohort characteristics and the historical cohorts-actions 
data inclusive of  success rates. The actions run for a specific period and can be made exclusive 
if needed, in which case, the same customer or account would not be targeted by different 
actions during the same period.
Step 5: Define Experiments for Each Recommended Action
Before implementing actions on the ground, it is important to test them on a smaller scale so 
as to be able to select a particular treatment, that is, a version of that action that is likely to 
produce the highest impact i.e. highest change in the primary metric. This is known as the 
Whitepaper  |  Harness the Power of Data Using Incedo LighthouseTM for Operational Decision Automation 12
Whitepaper  |  Harness the Power of Data Using Incedo LighthouseTM for Operational Decision Automation
Experimentation step which provides data-driven evidence to business executives based on 
which they can confidently implement the most significant action treatment at full scale. That 
said, experimentation is an optional step, especially in cases where executives have significant 
trust in their expertise and judgment about success of the said action and / or where the 
particular action is inevitable or inexpensive.
Incedo LighthouseTM has an Experimentation engine that supports business executives to 
make an informed decision on actions to be undertaken. Some of the key features of the 
module are:
l Determine the experiment type from the Options: 
¢ Digital A/B: For testing all the parallel options or treatments available under the 
  action. In this case there is no control group, so the executives have made up their 
  minds to change from the status quo. An example - if the action is about passing a 
  reduction in interest rate (and it is decided that keeping the rates same is not an 
  option), then the action treatments or options could be about 3 levels of interest 
  rate discounts.
 ¢ Test/Control: For testing all the parallel options or treatments available with the 
  control group for which no action is taken. In this case, the comparison is among 
  the treatments or options as well as with the control group. An example - same as 
  above with a change that there is a control group which sees no reduction in the 
  interest rates; so the comparison is across the 3 levels of rate reduction and the 
  control group.
13
Whitepaper  |  Harness the Power of Data Using Incedo LighthouseTM for Operational Decision Automation
 ¢ Pre/Post: This option is to test across the 
time horizon, the same population - once 
 before the action is taken and then post the 
action period completion. It is to be 
 noted that the pre-action period and post-
action period has to be the same for a 
 fair comparison. In this case, the sample 
chosen for the experiment does not 
 undergo further division.
l Finalize the target sample size to undergo 
the experiment based on the original 
population size and confidence interval 
desired. In addition, in case of A/B and 
Test/Control, further division of the sample 
population into the treatment groups is 
usually kept to be equalized for a fair 
comparison in the end.
l Identify success metrics and define targets 
to help finalize the specific action treatment to 
be chosen for full scale implementation post 
the experimentation stage. The treatment or 
the option that achieves or goes closer to the 
target is typically chosen for implementation.
Step 6: Execute, Monitor and Track 
Performance of Actions
Post the experimentation stage, comes the final 
step of executing the finalized action treatment / 
option on the full scale of the target population. 
This requires a two-way, solid handshake between 
Incedo LighthouseTM and the System of Execution 
(SOE) that is used as an operations management 
system at the client. 
Incedo LighthouseT M covers the following 
activities in this step:
l Execute Actions: Finalized action data 
including action metadata (name, duration 
etc.) and a target population list is sent to the 
SOEs (e.g. Workflow systems; Experimentation 
platforms) for operational execution.
l Monitor KPIs: How is the KPI responding to the 
14
Whitepaper  |  Harness the Power of Data Using Incedo LighthouseTM for Operational Decision Automation
actions is tracked using the actions-outcomes data sent over back to Incedo LighthouseTM 
by the SOE. Further analysis is done using the KPI tree structure - change at a node with 
drill-down to individual cohorts. 
l Experiment Summary: How are the actions performing over a period of time i.e. historical 
analysis of actions is done to update the recommended actions list for the future.
In summary, Incedo LighthouseTM accomplishes:
l Autonomous identification of interrelated business metrics that are witnessing anomalies 
or out-of-order behavior based on analyzing TBs of data that is refreshed on a daily basis 
using machine learning technology. This is achieved by various state-of-the-art anomaly 
detection algorithms working on timeseries and non-timeseries data.
l Automated identification of the root causes as to why the performance dip or uptick 
identified in the previous step are occurring. This is done by utilizing various machine 
learning algorithms such as segmentation, clustering and random forests that identify 
cohorts e.g., specific groups of accounts, customers or transactions that are responsible for 
the movement of the said KPI.
l Analyzing effectiveness of the improvement actions at the cohort level to improve the 
business performance. This is done to assess effectiveness of these actions by subjecting 
those to controlled experimentation before final recommendation and implementation.
l Continually tracks the improvement of business performance on the ground due to 
implemented actions by processing feedback data from the client’s workflow systems 
using KPI tree construct
15
Canada
Gurgaon
Iselin
Pune Hyderabad
Santa Clara Chennai
Mexico Bangalore
About Incedo
Incedo is a digital transformation expert empowering companies to realize sustainable business impact from their 
digital investments. Our integrated services and platforms that connect strategy and execution, are built on the 
foundation of Design, AI, Data, and strong engineering capabilities blended with our deep domain expertise from 
digital natives.
With over 3,000 professionals in the US, Canada, Latin America, and India and a large, diverse portfolio of long term, 
Fortune 500 and fast-growing clients worldwide, we work across financial services, telecom, product engineering, 
and life sciences industries. Visit our website to learn more about how we help clients transform today.
[email protected]
www.incedoinc.com