HBP# IM1282

IMD-7-2405
14.11.2022

RIMAC: HOW A PERUVIAN INSURANCE

COMPANY IS SCALING AI

Professor Amit Joshi and Ivy Buche prepared this case as a basis for class discussion
rather than to illustrate either effective or ineffective handling of a business situation.

Copyright © 2022 by IMD – International Institute for Management Development, Lausanne, Switzerland
(www.imd.org). No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form
or by any means without the prior written permission of IMD.

This document is authorized for use only in CAM GENSOLLÉN, CÉSAR ROGELIO's TRANSFORMACIÓN DIGITAL EN LA ORGANIZACIÓN 2024-I at Universidad de Lima
from Apr 2024 to Sep 2024.
 IMD-7-2405
RIMAC: HOW A PERUVIAN INSURANCE COMPANY IS SCALING AI

In 2022, Fernando Rios,CEO of RIMAC Seguros y Reaseguros (RIMAC), the largest

insurance company in Peru, set a new direction for the 125-year-old company: a strategic
expansion into well-being. This was in line with RIMAC’s new corporate purpose – “We
protect your world; we promote your well-being.” As Fernando drove this fundamental
shift in the company’s business, he faced the classic challenge – How could RIMAC
continue to bolster its position in the protection business while moving faster than ever
into the well-being space, where the company faced stiff competition not only from
insurance peers but also from other sectors, including fintech, banking, pharma, etc.?
The well-being trend had strengthened considerably in the wake of the Covid-19
pandemic as people’s priorities changed. To drive the new agenda, Fernando knew that
data and digital capabilities were key.

But did RIMAC have the data, algorithms, processes and tools to support this change?
Since 2018, Fernando had been driving RIMAC’s digital transformation, modernizing its
IT infrastructure and support applications across critical business units (BUs). The
objective was to deliver a superior customer experience through faster response times,
enhanced security and improved business continuity and resilience. Fernando believed
that the company was at the point where it had enough data and analytics capabilities
to start scaling up its transformation to deliver on its well-being strategy. Fernando
stated:

In financial terms, we are in two businesses – managing risk and making

investments – as we hold long-term capital. Both areas are under pressure
because of increasing risk premiums, interest rates and inflation. We need
to create a third source of income, because in the future I see significantly
more risk and different kinds of risk worldwide. I need to prepare RIMAC to
be resilient and pioneering in this new environment.

In mid-2022, Fernando was preparing for a meeting with his management team. He
pondered:

• How could RIMAC accelerate the democratization of the benefits of artificial

intelligence (AI) and analytics company-wide?

• How could the company’s AI capabilities drive its well-being agenda in the coming
years?

© 2022 by IMD 2
This document is authorized for use only in CAM GENSOLLÉN, CÉSAR ROGELIO's TRANSFORMACIÓN DIGITAL EN LA ORGANIZACIÓN 2024-I at Universidad de Lima
from Apr 2024 to Sep 2024.
 IMD-7-2405
RIMAC: HOW A PERUVIAN INSURANCE COMPANY IS SCALING AI

RIMAC OVERVIEW
RIMAC was owned by Grupo Breca, a diversified business conglomerate in Peru. With
a customer base of about 3 million and $3.1 billion in assets under management, RIMAC
provided all classes of insurance and reinsurance. Its key businesses included personal
insurance (life, health, auto) and property insurance (fire, hull, aviation, etc.). The
company was a majority shareholder in Clinica Internacional, Peru’s largest private
healthcare network, and RIMAC Internacional EPS, an HMO 1 health insurance
company. RIMAC was headquartered in Lima and had several offices in the main
provinces of Peru (see Exhibit 1 for company overview).

BUILDING A DATA-DRIVEN COMPANY

Fernando joined RIMAC in January 2017 as executive vice president of individual
insurance and marketing. Prior to that, he had gained nearly two decades of experience in
the banking industry, where he had worked with data analytics and created significant
value by transforming the credit card business, changing the client acquisition process,
evolving the value proposition and revamping the customer experience. He found key
differences between the banking and insurance businesses. Whereas customers always
needed a bank, it was difficult to demonstrate the appeal of insurance due to its negative
association with unforeseen life events. Also, banks typically held lots of data and moved
fast to leverage it for the purpose of better pricing, cross-selling, up-selling the next best
product and so on. By contrast, insurance companies appeared to have less data and
were also careful to avoid risks associated with how it was harvested and used. Fernando
carefully studied the Peruvian market to understand customers’ needs. He was certain that
the insurance industry as a whole needed to change. Two attributes were critical – speed
and personalization. Fernando believed that RIMAC could truly differentiate itself from the
competition by delivering customization at scale. The question was how? Part of the
answer lay in digitalization and building a data-driven approach to develop new value
propositions and a different perspective toward clients. According to Fernando:

No aspect of our lives has the potential to be more personalized than well-
being. To provide well-being products and services, we definitely needed
to start the journey to customize our offer using data, AI, algorithms and a
strong analytics platform.

1 An HMO, or health maintenance organization, is an insurance structure that provides coverage

through a network of physicians. HMO plans typically have lower monthly premiums than a PPO
(Preferred Provider Organization) plan.

© 2022 by IMD 3
This document is authorized for use only in CAM GENSOLLÉN, CÉSAR ROGELIO's TRANSFORMACIÓN DIGITAL EN LA ORGANIZACIÓN 2024-I at Universidad de Lima
from Apr 2024 to Sep 2024.
 IMD-7-2405
RIMAC: HOW A PERUVIAN INSURANCE COMPANY IS SCALING AI

Fernando’s first challenge was to convince the board that data and analytics needed to
be used differently, and significant investments would be required to build new
capabilities. Fernando explained:

Becoming an analytics-driven organization is expensive. If you invest the

amount of money that is necessary to shift your organization, you will
impact the net profits. And it’s not only a matter of money. You have to
initiate a huge organizational change because analytics alone is cost, but
analytics connected with your organization and clients is value.

After several discussions the board agreed, and the budget to invest in advanced
analytics was released.

The initial approach to advanced analytics: Project-based

experimentation
The process started with small steps. In 2018, Fernando invited Miguel Parades to head
up a small, newly created team. Miguel held a PhD in applied machine learning and data
science from MIT. He had been working with RIMAC since 2015, as part of efforts driven
by Breca’s Center for Advanced Analytics, which he had built. Prior to that, he had
12 years of experience in the energy, aviation, consulting, academic and government
sectors. His objective was to demonstrate the ability of data analytics to deliver better
outcomes and capture untapped value by addressing specific business problems. The
10-member team was comprised of six data scientists (from the Breca group) and four
engineers from RIMAC’s existing business intelligence (BI) group.

Miguel found that the company already had a small data warehouse, but that the data
was only being used in traditional ways – mainly for reporting, creating predefined
dashboards and providing BI that enabled (some) decision making. Given its small size,
the BI group could only serve a third of the organization and aspects like data governance
and data management were not mature. There was significant potential to upskill the
team and operate on a more advanced level. The business side was also not
accustomed to using data in different ways – they did not know where to find it, what it
meant and how to use structured query language (SQL) or BI queries to extract
information from it.

Given this starting point, Miguel realized that, in the short and medium term, value was
not going to come from advanced analytics and AI, but rather from getting the data in
order. Initially, the data scientists delivered on a five-in-one role (data collection and
cleaning, as well as model creation, deployment and monitoring). They even struggled
to get enough data to build training datasets. They had to reach out to several BUs,
extract data on their own and figure out whether or not the data was consistent (e.g., if
there were two datasets, which one was closer to the ground truth).

© 2022 by IMD 4
This document is authorized for use only in CAM GENSOLLÉN, CÉSAR ROGELIO's TRANSFORMACIÓN DIGITAL EN LA ORGANIZACIÓN 2024-I at Universidad de Lima
from Apr 2024 to Sep 2024.
 IMD-7-2405
RIMAC: HOW A PERUVIAN INSURANCE COMPANY IS SCALING AI

In the first year, the key areas of focus were data governance (including quality and
remediation), data democratization and availability, and the creation of a data platform.
There were many obstacles to collecting and aggregating data from the core IT systems
in order to create a central data lake. It required new processes and new architecture for
capturing data and standardizing the various streams, which involved complete migration
and re-factorization. It also called for new skillsets – cloud engineering and modern data
architecture – that were not easily available in the Peruvian and Latin American markets.
Only then could the data scientists access the data in order to build and deploy AI
models. Miguel emphasized:

All these big, important things were not sexy, but they needed to be put in
place for the company to be data-driven so that people could look at
reports and trust the numbers. The hardest thing to do was to make sure
that the company had the necessary guardrails and systems in place,
which in turn would allow for superior value creation with AI and analytics.

Miguel’s team interacted with the business to understand the pain points and build a
solution. This was not easy since the team was new to insurance and had limited domain
knowledge. They had to make several presentations and iterations to ensure that the
solution worked for the user, and that the right business stakeholders were in the room
to provide feedback. They started with basic classification models (such as regression)
and gradient boosting techniques. Running the model each month using new data in
order to provide the BU with up-to-date projects involved a great deal of manual work.
Some early successes included:

• Reducing customer churn by five percentage points in vehicle

insurance. The pilot proved that the reduced churn was equivalent to about
$750,000 in retained policies per year.
• Customer classification to identify propensity to buy. Based on past
transaction data, customers were assigned a score and a color (red, yellow or
green). This seemed simplistic, but it allowed the salesforce to identify which
customers were likely to stay with the company and then focus their efforts
accordingly. Over time, this model increased overall customer acquisition by 10%.

Initially, the fledgling team simply supplied what the BU asked for without exploring the
underlying problem. This created a new issue: Each data scientist, being research-
oriented, used his or her own creativity to build the predictive models. Building the
algorithms took time because the data scientists were often perfectionists and
accustomed to working alone as opposed to co-creating with the business partners. They
did come up with interesting models but were unable to scale them up to become a part
of the business strategy. Also, there was increasing duplication. Putting each model into
production took months and integrating them took additional months.

© 2022 by IMD 5
This document is authorized for use only in CAM GENSOLLÉN, CÉSAR ROGELIO's TRANSFORMACIÓN DIGITAL EN LA ORGANIZACIÓN 2024-I at Universidad de Lima
from Apr 2024 to Sep 2024.
 IMD-7-2405
RIMAC: HOW A PERUVIAN INSURANCE COMPANY IS SCALING AI

Building momentum: Creating a Center of Excellence (CoE) for

Advanced Analytics and AI
In mid-2019, Miguel recruited Rosario Alcedo to help him build the CoE for Advanced
Analytics and AI. The goal was to develop the analytics capabilities and capture the full
value of solutions. The CoE served the BUs (life insurance, health insurance and
enterprise insurance) and coordinated with other functions to deliver value (IT, risk,
actuarial, claims, etc.). Rosario stated:

My challenge to the team was that the business should feel that the solution
is theirs. If the business partner/user cannot explain our algorithm, then we
are not successful. We had to change mindsets and ways of working.

A new role – project lead – was created and three people were appointed to liaise with
the businesses to drive co-creation and validation of the solutions. Communication was
key. The project leads and Rosario spent at least 50% of their time understanding the
business problem or the strategic opportunity to enhance impact. The CoE then moved
on to atypical/unstructured data – such as audio for speech analytics and images for
computer vision – to which they applied more sophisticated algorithms and deep learning
techniques.

Successful use cases

• Faster assessment of vehicle insurance claims using computer vision.
Previously, it could take up to a week to verify vehicle accident/damage claims
because it was a manual process – someone had to go and take pictures while
another person had to verify image by image. The data team decided to focus
on the principal issue – time. An assessment model based on image
verification first classified the damage as either major or minor. During the pilot
phase, verification time was shortened to 10 minutes, with an 80% accuracy
rate for minor damages, which in turn accelerated the approval process. For
major damage scenarios, a person was sent out to view the damage and offer
advice. More and better-quality images as well as human feedback allowed
the model to continuously improve, and a full scale rollout was planned by end
2022, when the accuracy rate was expected to reach 95%.

• Improved payment collection across BUs. Three analytics models were

developed to segment customers based on past payment data. The payment
behavior model, based on a one-month window, scored the portfolio at the
beginning of the month to see if customers would pay or not. A complementary
payment projection model, also based on a one-month window, identified
whether a person who was delayed would at least pay some amount. Finally,
a delinquent model identified if a customer would pay on time, pay late or be
a complete loss. The collections BU suggested new variables to retrain the

© 2022 by IMD 6
This document is authorized for use only in CAM GENSOLLÉN, CÉSAR ROGELIO's TRANSFORMACIÓN DIGITAL EN LA ORGANIZACIÓN 2024-I at Universidad de Lima
from Apr 2024 to Sep 2024.
 IMD-7-2405
RIMAC: HOW A PERUVIAN INSURANCE COMPANY IS SCALING AI

model, ultimately achieving an improvement in collections from 65% to 80% in

two years. Carlos Seminario, head of collections, noted:

Capturing value from the AI models is faster in some divisions like ours
because we can contact the right customers straight away and start
making recovery more efficiently (as compared to new customer
acquisition which takes longer). With the correct models, we can
personalize the best day, time and channel to follow up on collections.

• Better customer sentiment analysis using speech analytics. Contact

center managers did not know if the people handling calls had done a good
job and followed all the procedures. In addition, it was unclear how they could
be empowered to solve customer complaints. Every month, managers had to
listen to the audio recordings to check. Despite this, they did not know whether
the customers were happy with the service provided because not everyone
responded to follow-up surveys. The analytics team built a solution in five
months. Audio recordings from the call center were analyzed using speech
analytics algorithms to recognize the customer’s sentiment, and the data was
displayed on a dashboard showing positive, negative and neutral sentiments.
Cluster analysis was then used to improve the customer experience.
Exchanges with call center people were also assessed for training purposes.
After implementing speech analytics and taking relevant action, the positive
sentiment increased from 34% to 68% in one year. Giselle Larco, vice
president of contact center, observed:

Simply by using the data we understood what our responders needed

to solve the problem instantly. The number of complaints fell from 4,000
to 600 per month in three years. Our next goal was to quickly respond
to those 600 complaints. Three years ago, we answered 2% of the
complaints in 24 hours. Now it is 60% in 24 hours and 85% in 7 days.
I would love to get to the stage where we already know what the
customer is calling us about before we ask, “How may I help you?”

The success of these use cases was measured by two types of metrics – business-
related KPIs (such as new sales, bottom line improvement, etc.) and analytics-related
KPIs (such as accuracy, precision, effort required and correlation between predictive
power and business KPI). In two years, RIMAC’s net promoter score (NPS) increased
from 20 to 53. The biggest improvement was in the vehicle insurance business which
saw a jump in NPS from 18 to 70, largely attributed to the positive impact of the above-
mentioned analytics models in handling claims and improving customer assistance.

Prioritization
As the CoE gained momentum, it started experiencing greater pull from the businesses.
Rosario stated:

© 2022 by IMD 7
This document is authorized for use only in CAM GENSOLLÉN, CÉSAR ROGELIO's TRANSFORMACIÓN DIGITAL EN LA ORGANIZACIÓN 2024-I at Universidad de Lima
from Apr 2024 to Sep 2024.
 IMD-7-2405
RIMAC: HOW A PERUVIAN INSURANCE COMPANY IS SCALING AI

When we began, we were going to the businesses. Subsequently, the

businesses started coming to us. The demand was higher than our ability
to support. We had to prioritize the initiatives.

To make go/no-go decisions, the CoE evaluated initiatives using three criteria – feasibility
based on data availability, execution readiness of the BUs and potential impact (measured
as the discounted present value of incremental revenues or reduced costs). For example,
the customer service unit wanted the CoE to build an “agenda” model so that the
salespeople could prioritize which leads to follow-up. While it was clear that the potential
impact would be significant, the data was not available. Miguel explained:

Prioritization is challenging in data science because there's so much

uncertainty. You do your best to do the value assessment and establish
the value potential of projects with the BUs. But you don't necessarily
understand the implementation costs or complexity. Sometimes things
take longer, or the data is not as good as expected, or the model might
have great accuracy, but the interventions might not work. This is not
because of a lack of technical rigor, but because of poor execution on
model insights and predictions.

Scaling up
In 2021, Miguel left RIMAC. Fernando decided to move the CoE under the marketing
department, with Rosario reporting to the chief marketing officer. At the same time, the
data engineering team was moved under the IT department. The rationale for having the
data engineering team under IT was the interconnected nature of the work – processes,
control mechanism, applications, supporting infrastructure, cybersecurity, etc.

The CoE, comprising about 20 people, was divided into three groups (see Exhibit 2 for
the CoE organizational structure):

• A data scientist team consisting of 10 to 12 data scientists was tasked with

designing, implementing and standardizing models using a machine learning
(ML) design and development framework.

• A five-member MLOps team was in charge of getting models to production

and deploying them as a service via APIs. In the process, the team understood
that building an ML model was not about executing a Jupyter notebook or
building traditional software (web or app). Instead, they had to be cognizant of
many dependencies – on the code and algorithms used in model
development, as well as on the data. This team interacted with both the
advanced analytics team and the IT team to ensure successful deployment.
They also interacted with the business when something went wrong. One team
member outlined the challenges:

© 2022 by IMD 8
This document is authorized for use only in CAM GENSOLLÉN, CÉSAR ROGELIO's TRANSFORMACIÓN DIGITAL EN LA ORGANIZACIÓN 2024-I at Universidad de Lima
from Apr 2024 to Sep 2024.
 IMD-7-2405
RIMAC: HOW A PERUVIAN INSURANCE COMPANY IS SCALING AI

There is a huge difference between the test environment and

implementation in the production environment where you have to take
into account response times, traffic, etc. We code and automate the
process of deployment. Sometimes the data scientists give us models
in different formats; the deliverable is not clean. We also have to
integrate with different vendors.

• A three-member AI project lead team acted as strategic partners of the

businesses. The team sought to understand each business’s needs and
assessed the impact and feasibility of use cases.

AI Factory
To standardize processes and scale up model deployment for high-quality solutions, the
CoE established an AI Factory (see Exhibit 3 for AI Factory functions and process flow).
Another goal was to share knowledge and best practices, thereby reducing effort, time and
costs. The planning of the AI factory included all parties – users who benefited from the
models’ insights/predictions (BUs), the data providers (the data engineering team under IT
and the BUs) and those who would put the technology into production alongside other IT
systems (IT department). The AI Factory enabled a shift from providing algorithms and
predictions in piecemeal fashion to working with the P&L owners on business problems
and strategizing how to create value. The AI Factory’s key elements were:

• Feature Store. The Feature Store was a repository of clean, standardized

data and a system for transforming, storing and serving model features to data
scientists. The goal was to reduce the time required to build solutions because
the features were already in place to train the model. The next step was to
automate as many processes as possible in order to generate analytics
solutions with reduced human intervention and a fast response time.

• AI Platform. The AI platform featured a set of tools, guidelines and templates

so that data scientists didn’t have start from scratch when building algorithms.
The data scientists uploaded the algorithms ready for deployment on to the
platform. New data scientists could develop ML models that were similar to
those created by their predecessors, thus avoiding re-processing and
duplicating efforts and ensuring reutilization of features.

• MLOps Platform. This platform transformed the algorithms built by the data
scientists into consumable solutions via API channels. About 20 solutions
were being productized in various business areas.

• Model governance. Real-time monitoring of the performance of AI solutions

and level of use helped during subsequent scale-up. It was essential to obtain
feedback from the clients to understand if value was being created, and to

© 2022 by IMD 9
This document is authorized for use only in CAM GENSOLLÉN, CÉSAR ROGELIO's TRANSFORMACIÓN DIGITAL EN LA ORGANIZACIÓN 2024-I at Universidad de Lima
from Apr 2024 to Sep 2024.
 IMD-7-2405
RIMAC: HOW A PERUVIAN INSURANCE COMPANY IS SCALING AI

improve by retraining/recalibrating the ML models – not in an isolated manner

but in collaboration with the business. By mid-2022, this was a work in progress.

Emergence of a federated model for scaling AI

As the BUs bought into the value of analytics, they had to compete for CoE resources
for new initiatives in their verticals. The perception was that they were in a queue. Some
businesses started feeling the need for distributed capabilities to be able to move faster
and get more value. Taking the long-term view for scaling up capabilities, it was
necessary to map out which aspects would need to be centralized, and which could be
distributed over the next two to three years. After discussions with Rosario, the federated
model took shape in the health insurance business. Roberto Leon, vice president of
health insurance noted:

We have to think about the organization that we want to have. Perhaps a

third of the organization will be centralized and two thirds will be embedded
in the business areas working on dedicated projects on priority.

Roberto hired 10 people to create a dedicated team that worked for more than a year to
organize and structure claims data. It started with complex business intelligence.
Subsequently, the team created customer health profiles to distinguish between chronic,
high-risk and healthy people. They then used analytics to design incentives and programs
so that the customers would work towards maintaining their health. For example, they
created a model to see how people with diabetes would fare over time – with and without
treatment. Sharing this kind of data with customers not only nudged them toward better
health choices but also showed them how much money they could save.

The team also co-created models with the CoE to identify use, abuse and fraudulent
claims. Instances of undesirable use of health insurance policies were revealed. The
business began monitoring such behavior and invited customers to the company’s local
offices to understand why they had made so many appointments or obtained
prescriptions for several drugs. Health-related claims by these high-use customers
dropped from 92% to 75%.

Driving a culture change

Fernando knew that building a data-driven company would require a deep culture
change. It was not enough to change the profile of the employees; working methods had
to evolve to help people achieve the best results in the various areas. The most important
change was to people’s mindsets, which started top-down at RIMAC. Fernando had
several conversations with C-suite executives to explain what this change meant and
what they, as leaders of BUs/functions needed to do to foster it. In the early days, having
Miguel in the C-suite reporting directly to Fernando sent a clear signal about the
intentionality and importance of the change. According to Fernando:

© 2022 by IMD 10
This document is authorized for use only in CAM GENSOLLÉN, CÉSAR ROGELIO's TRANSFORMACIÓN DIGITAL EN LA ORGANIZACIÓN 2024-I at Universidad de Lima
from Apr 2024 to Sep 2024.
 IMD-7-2405
RIMAC: HOW A PERUVIAN INSURANCE COMPANY IS SCALING AI

Not everyone has the right attitude to change. You have to hire people
from outside who share your dreams and your mindset, and who know the
value of this change. It is important to start small to capture value, to
demonstrate the value, to create momentum and then to scale.

Both Miguel and Rosario were change ambassadors. The CoE had learned a great deal
as it had grown, and in particular developed the ability to follow RIMAC’s strategy and
co-create with the businesses. This was easy to say but finding the pathways to work
together in pursuit of quantifiable strategy-related KPIs for every solution being deployed
in the business was hard.

CHALLENGES
In four years, RIMAC had come a long way to become a data-driven company. Initial
skepticism and resistance to change decreased as the business generated value from
the direct impact on customers. Yet by mid-2022, some challenges remained:

• Availability and quality of data. RIMAC needed to know its customers in terms of
qualitative information: their needs, lifestyles and expectations – not just
demographic data. Despite significant progress, there were still gaps. Milagros
Rosell, vice president of customer experience, noted:

We need not just macro but also micro data to personalize the CRM
[Customer Relationship Management] program. For example, we don’t
have lifecycle data such as churn or renewal by channel. We need to work
further on pricing, renewal, churn and retention models that will help us to
prioritize – giving better prices or add-ons to the best customers to improve
retention, reaching out to those customers who have a propensity to churn
or making proactive efforts to engage detractors who might have had to a
bad experience and improve their perception about RIMAC.

• Limited data sharing across BUs. For example, the life insurance business knew
who its best clients were but was unaware if the same customer was a good one for
vehicle and health insurance as well. Makaly Rivera, vice president of life insurance,
stated:

I want to sell a life insurance policy to a new mother. We have that

information because we pay for health insurance claims. But I don’t receive
it. We need all the information of client interactions for our other products.
We need more scores – e.g., the propensity to buy at the product level.
For one year we’ve been trying to cross-sell, but we don’t have a process.
We don’t know how to tell if the score, the implementation (experiments,
A/B testing), or the salesperson is the problem.

© 2022 by IMD 11
This document is authorized for use only in CAM GENSOLLÉN, CÉSAR ROGELIO's TRANSFORMACIÓN DIGITAL EN LA ORGANIZACIÓN 2024-I at Universidad de Lima
from Apr 2024 to Sep 2024.
 IMD-7-2405
RIMAC: HOW A PERUVIAN INSURANCE COMPANY IS SCALING AI

• Dynamics between the advanced analytics team (under the marketing

department) and the data engineering team (under IT). Although the two teams
were complementary, there was an inherent tension. The data engineering team was
responsible for the complex tasks in getting the right data, merging it from disparate
systems and cleaning it for use. The analytics team was also under significant
pressure as they were answerable to the business to achieve ambitious targets. They
needed good data – and quickly – driven by business-related urgency, while the data
team struggled to explain why it was so hard and time-consuming to provide clean
data. Carlos Herrera, vice president of IT and Data, noted:

The two teams have two different mindsets, and both are pressured by the
circumstances of the job to be done. The data team is trying to have as
much data as possible – this additional variable or that additional field –
while the analytics team is focused on the customer request. It's quite
difficult to harmonize both groups, and we are still working on the
interaction between the two.

• Addressing increased demand from BUs. Initially, in an effort to deliver quick wins,
the CoE had focused on the auto and health insurance areas. Life insurance,
however, which was one of the largest BUs, had not yet been adequately addressed.
Some 50% of the costs in the life insurance business were the indirect costs of data,
analytics, IT and marketing because cost allocation was based on the number of
customers. But the value was not yet visible. For example, the business had some
scores which allowed salespeople to identify which customers were likely to stay with
the company, but it still lacked scores that could highlight customer preference at the
product level. The businesses were also asking for more predictive models with
greater accuracy based on an integrated view of the customer.

• Alignment of objectives. The CoE had its objectives and key results (OKRs) and
goals. These included generating $9.5 million in incremental gross profits in 2021.
However, the BUs within which this value was to be created did not have these
specific objectives. This meant that they did not necessarily prioritize resources,
bandwidth or efforts to implement (or help create) the models. Going forward, RIMAC
realized that it needed to move to a scheme with shared OKRs between the CoE and
the BUs.

• Value accounting and value validation. A significant challenge was to rigorously

define and validate the value created by the models, and then get the BUs to agree.
The CoE only accounted for marginal value that was generated by the models and
comparing against the past. The concept of value was also evolving as the AI models
became more sophisticated. Giselle Larco, vice president of the contact center,
observed:

So far, we have been doing speech analytics after the fact (using
recordings). Can we do it in real time? Also, the principal channel for

© 2022 by IMD 12
This document is authorized for use only in CAM GENSOLLÉN, CÉSAR ROGELIO's TRANSFORMACIÓN DIGITAL EN LA ORGANIZACIÓN 2024-I at Universidad de Lima
from Apr 2024 to Sep 2024.
 IMD-7-2405
RIMAC: HOW A PERUVIAN INSURANCE COMPANY IS SCALING AI

complaints now is our website; can we answer 20% of complaints with a

bot instead of a human?

• Build long-term thinking. Business leaders tended to take a short-term view of

data, i.e., they needed data for a report or for quick decisions. The challenge was to
change the mindset so that they understood how data could be used for more
sophisticated long-term endeavors. Giancarlo Diaz, deputy manager of vehicle
insurance, observed:

We have to evolve our thinking from products to customer needs to the

ultimate willingness to pay, which can be unlocked through personalization
of multiple aspects – the appropriate product, coverage, channel, payment
preference, etc. The opportunity to grow the business is massive as
insurance penetration is low in Peru. For example, out of two million cars
in the market only 28% have car insurance.

• New priorities for the CoE. It was essential to expand the team, build superior AI
capabilities and forge new collaborations (with external partners) to gain time-to-
market. For example, RIMAC held a hackathon which attracted 500 participants. The
event not only brought new ideas but also created a brand name for RIMAC as a
technologically advanced insurance company. Additionally, to identify areas of
collaboration, Fernando, Rosario and certain business heads visited Israel in early
2022 and interacted with more than 20 startups and venture capital funds. Fernando
stated:

Their advanced use of AI in healthcare was impressive, for example,

creating devices and applications to decentralize diagnostics so that more
people could have access. After reviewing all the new products and
services, I realized that the world is more advanced and moving faster; we
have to accelerate.

FERNANDO’S FUTURE AGENDA: WELL-BEING

RIMAC needed to address all of the challenges listed above in order to deliver on the
company’s well-being strategy – “the second ring” as Fernando dubbed it. His aspiration
was to make RIMAC an everyday brand for Peruvians with a positive energy around it.
Fernando stressed:

The main objective is to build the second ring by creating well-being-

oriented products and services that Peruvians will want to use every day.
As an example of this, we launched a digital pharmacy in 2021. Imagine
things like nutrition, mental health and physical fitness in the well-being
space. We are starting to offer products and services to meet these needs.

© 2022 by IMD 13
This document is authorized for use only in CAM GENSOLLÉN, CÉSAR ROGELIO's TRANSFORMACIÓN DIGITAL EN LA ORGANIZACIÓN 2024-I at Universidad de Lima
from Apr 2024 to Sep 2024.
 IMD-7-2405
RIMAC: HOW A PERUVIAN INSURANCE COMPANY IS SCALING AI

To execute the strategy, new well-being platforms with data at the core would need to
be built. This major project kicked off in 2022 and it called for greater resources – budget,
talent, capabilities and processes. Conceptually, it was difficult to put the customer at the
center because traditionally the norm had been to think of the business in terms of
products. In contrast to insurance touchpoints that occurred only couple of times in a
year, typically at difficult moments, well-being would shift RIMAC to a more positive and
ongoing relationship. For its new well-being platform called Estar Bien, RIMAC would
need to start combining internal data with big data to create real-time customization. With
Estar Bien up and running in the near future, the data provided by the customer to the
platform could be made available to all the BUs, thereby creating the opportunity for
RIMAC to become a trusted advisor and improve customers’ lives.

With the goal of becoming a protection and a well-being company, Fernando started
penning his agenda for the management team meeting, “Our ambition is to move into
new territory, and well-being will help us become closer to our customers….”

© 2022 by IMD 14
This document is authorized for use only in CAM GENSOLLÉN, CÉSAR ROGELIO's TRANSFORMACIÓN DIGITAL EN LA ORGANIZACIÓN 2024-I at Universidad de Lima
from Apr 2024 to Sep 2024.
 IMD-7-2405
RIMAC: HOW A PERUVIAN INSURANCE COMPANY IS SCALING AI

15
Exhibit 1: RIMAC Overview 2022

Source: Company

© 2022 by IMD

This document is authorized for use only in CAM GENSOLLÉN, CÉSAR ROGELIO's TRANSFORMACIÓN DIGITAL EN LA ORGANIZACIÓN 2024-I at Universidad de Lima
from Apr 2024 to Sep 2024.
 IMD-7-2405
RIMAC: HOW A PERUVIAN INSURANCE COMPANY IS SCALING AI

Exhibit 2: CoE for Advanced Analytics and AI Organization Structure

Source: Company

© 2022 by IMD 16
This document is authorized for use only in CAM GENSOLLÉN, CÉSAR ROGELIO's TRANSFORMACIÓN DIGITAL EN LA ORGANIZACIÓN 2024-I at Universidad de Lima
from Apr 2024 to Sep 2024.
 IMD-7-2405
RIMAC: HOW A PERUVIAN INSURANCE COMPANY IS SCALING AI

17
Exhibit 3: AI Factory Functions and Process Flow

© 2022 by IMD

This document is authorized for use only in CAM GENSOLLÉN, CÉSAR ROGELIO's TRANSFORMACIÓN DIGITAL EN LA ORGANIZACIÓN 2024-I at Universidad de Lima
from Apr 2024 to Sep 2024.
 IMD-7-2405
RIMAC: HOW A PERUVIAN INSURANCE COMPANY IS SCALING AI

18
Exhibit 3 (continued)

Source: Company

© 2022 by IMD

This document is authorized for use only in CAM GENSOLLÉN, CÉSAR ROGELIO's TRANSFORMACIÓN DIGITAL EN LA ORGANIZACIÓN 2024-I at Universidad de Lima
from Apr 2024 to Sep 2024.