Data-driven PE: value creation with repeatable analytics and ML
DragonFly is a unique set of technology and tools that powers our investment strategy at WovenLight, from sourcing and diligence through to value creation under our ownership. Here, we take a look at one of the key challenges in developing DragonFly, building repeatable data analytics and machine learning solutions.
For an investment firm, applying data analytics and machine learning (ML) can unlock new value, growing capabilities in portfolio companies and within our own. But as highlighted by Casado and Bornstein, developing these approaches brings challenges, in part because business problems and data inputs vary across companies — unlike traditional software development in which problems and inputs are often more easily defined.
This adds up to increasing cost and development time to maintain analytics, reducing their adoption and robustness, and increasing the time to value. Ultimately, we see a decline in return of investment, compared with traditional software where there is typically a low marginal cost (and high margin) once products are developed. As Casado and Bornstein note, the reduction in margins gets worse given additional cloud costs and requirements for on-going support and maintenance. (Although there are product companies that develop analytics products for more specific or narrow use cases, these products still require configuration and customisation where data or problems change slightly.)
Feasible, viable and valuable
Building products generally requires defining a target problem or use case, trading-off between specificity and flexibility (generality). But this can risk building a product that is specific to a use case and doesn’t generalise well to other contexts (and hence risks not being used); or developing products that are too general to be useful. The challenge of product development lies in finding a balance between the two that is feasible, viable and valuable.
The challenges with analytics productisation have led to a proliferation of consultancies offering services for the development of data science and analytics solutions. In this context, where consultancies are brought in to run projects for clients, it’s often necessary to address novel use cases and build infrastructure that must integrate with existing custom client systems. These requirements increase the project cost and development time, and significantly lower the ROI.
By contrast, in an investment context, more influence can be placed on system architecture and infrastructure direction, potentially reducing costs and development time and allowing for standardisation. While large enterprises may be able to develop in-house, or hire consultancies, these additional costs make it more difficult for small and medium size companies to realise the value of analytics.
In this context, data analytics needs to work at different scales: across companies; across product lines; across geographies.
Building repeatable solutions requires a standardised, yet highly flexible, approach including:
— Templates for analytics use cases to provide jumping-off points;
— A common analytics framework and project structure for higher quality and to enable collaboration;
— A technology infrastructure for agile, scalable development and deployment of solutions to improve the development cycle and reduce time to feedback;
— An open and collaborative culture combined with strong technical practices to empower analytics professionals.
Let’s examine each in a little more detail.
As discussed above, productising analytics use cases is hard as edge-cases will persist. However, there is often commonality and overlap between use cases, such that starting from scratch would be slow, costly and unnecessary.
Developing templates as starting points that are both guides and a collection of data repositories, features, and hypotheses with analytics libraries. Templates that are based on modular, reusable components reduces customisation and development time, thus accelerating the journey to value. Iterating on the templates after delivering use cases leads to a feedback loop of improvement.
Delivering for analytics use cases can involve many moving parts: from data and pipelines; to configuration and models. Additionally, there are a plethora of tools and systems available for use. Over time, a lack of standardisation and fragmentation introduces additional technical debt and raises barriers to collaboration, increasing maintenance and slowing down development.
Standardising project structure and building on a common, robust analytics framework, while codifying best practices and design patterns, leading to higher quality and more maintainable projects, with smoother collaboration.
In an investment context, developing analytics solutions where there is imperfect, constantly evolving data, varying operational and technical requirements, and need to transfer capability to portfolio companies, poses several technical challenges. It demands a platform that rapidly provisions and configures secure cloud and analytics infrastructure, facilitating the agile development and deployment of solutions.
Developing analytics requires managing complex data and code dependencies: data; pipelines; features; models; and outputs, leading to slow development cycles and feedback loops, and increasing friction when collaborating. Data and compute requirements can vary considerably across projects and project stages. There are a myriad of tools and platforms across public cloud providers to support the development lifecycle, but their use can lead to fragmentation and lock-in.
Deploying analytics solutions on a standard, cloud-based infrastructure supporting the end-to-end development lifecycle — enabling agile, scalable development and deployment of analytics use cases. Infrastructure delivering the components of an operational analytics workflow, such as: development environments; CI/CD; experiment tracking; feature stores; and model monitoring. Ultimately, this means enabling quicker development cycles; reducing friction to collaboration; increasing impact; and reducing time to value.
Together with the templates and analytics framework, enabling faster deployment and value capture for companies using data and analytics.
Delivering successful analytics in the investment environment, across portfolio companies with operational and infrastructure requirements, requires cross-functional expertise and teamwork. Without strong technical practices, this can lead to knowledge silos and ‘blocking’ dependencies or handoffs, increasing friction to collaboration and slowing the development cycle, ultimately reducing the impact. We mitigate this by empowering teams with end-to-end ownership of the analytics value stream, for a quicker development cycle, faster feedback; and enabling continuous improvement.
Through building a technical culture around DevOps/MLOps practices for fast flow, and continuous feedback and improvement, we create healthy, growing teams. And we supplement this with clear on-boarding processes and best practices further reducing technical fragmentation, strengthening cross-team collaboration, and building quality in from day one.
At WovenLight, we aim to combine a Playbook of analytics use case templates built on a decade of experience deploying similar use cases; with an evolving Protocol that defines our repeatable methods; with our exceptional People; all underpinned by our technology, Dragonfly, used to accelerate deploying analytics across portfolio companies.
— A set of analytics use cases underpinned by templates of re-usable analytics components to enable faster deployments.
— Expertise assisting in where to look for the right data, what to build, and assisting in building it.
— An evolving, repeatable method for the end-to-end analytics lifecycle, driving consistency, efficiency and quality.
— A consistent process for the on-boarding and training of team members to our analytics approach, accelerating to value, and increasing productivity.
— Codifying technical best practices and technology choices reducing development fragmentation and technical debt.
— Open, collaborative team culture to reduce knowledge silos and increase productivity.
— End-to-end ownership of the analytics value stream empowering analytics professionals.
— Practices for continuous feedback and improvement to support team development, improve productivity, and encourage a growth mindset.
For WovenLight and our teams, the goal is to kickstart the flywheel, building capabilities that compound: improving sourcing and diligence, driving repeated value creation, and better ownership.
All of the above is driven by an open and collaborative team culture, to solve the most important and impactful problems, and reduce your responsibilities and obligations. The outcome is that your business should feel easier to sustain year after year.