InsuranceCo seemed to have all the ingredients needed to achieve its data analytics vision: good-quality data, strong investment in technology, internal talent and a commitment to success. But at the end of the day the tangible results were not progressing fast enough.
Perhaps that shouldn’t have been a surprise, because, while the company had assembled some important building blocks, it had also succumbed to some common pitfalls. For one, it had not systematically identified, evaluated and prioritized the opportunities; it delivered on small projects but didn’t tackle the richest opportunities. Its analytics team was not adequately connected to business-unit and functional experts, and its operating model was unclear. Business owners didn’t understand or trust the data team, regarding them as junior employees who lacked a full understanding of the business. And, despite InsuranceCo’s investment, the analytics team still lacked critical talent and technology.
Working with Bain, InsuranceCo sought to make good on its analytics goals. To make that happen, Bain worked closely with InsuranceCo on a five-part plan.
- The project began with the creation of a joint Bain-InsuranceCo team and the selection of a high-value opportunity to prove that a fresh approach to analytics could produce meaningful results. The joint team comprised a sponsor level and a joint operational team of a dozen people (dedicated full time, all with coding knowledge). For their first effort the team focused on the significant potential in the company’s cross-selling efforts.
- That effort required a careful analysis of the current situation. Leveraging internal and Bain benchmark data, the team assessed InsuranceCo’s current portfolio and cross-selling performance, created customer cross-sell profiles, explored the events (personal and contractual) that triggered the sale of additional products, and mapped the time and sequence dynamics that affected product ownership over a 10-year span. Bain also tapped the power of artificial intelligence, by applying machine-learning algorithms against the data to continually refine predictions about which product a customer was likely to buy next. To support that analysis, the team created a comprehensive data lake, integrating 20 databases into a system that contained a rich 10-year history of client and external data.
- Key to the project’s success was to co-locate the team in a single room and use Agile development methodologies to break the project down into discrete sprints that spanned every core task, from data prep and loading to, at the end, test and implementation plans and knowledge transfer.
- Another important facet of the project was to continually engage top management and challenge accepted beliefs, as a way to go beyond the status quo. A series of weekly meetings ensured that roadblocks were quickly addressed and that all levers affecting cross-selling were surfaced and effectively explored.
- Finally, the team put a premium on knowledge transfer at every step, including the provision of systematic documentation in a shared library with version control. The Bain team tackled much of the most complex coding, helping the client team refine their skills along the way, through both formal and on-the-job training.
By the end of the project the team had developed a detailed roadmap comprising a dozen levers anticipated to produce an approximate 25% in additional revenue potential through cross-selling. The marketing insights gleaned through new machine-learning capabilities promised a 10x gain in marketing performance. The team developed a number of different machine-learning models, of increased complexity, for two different products, and benchmarked them against standard cross-selling methods to determine the financial impact.
InsuranceCo is now positioned to apply advanced analytics to ever-larger data sets, across more initiatives, with in-house resources that have been carefully trained in both analytics and Agile methods.
* We take our clients' confidentiality seriously. While we've changed their names, the results are real.