I Need Causal AI- Now What? Build vs Buy

Executive Summary

For Data Leaders, deciding whether to build or buy Causal AI tooling will have a significant impact on your decision-making capabilities.

As AI-driven decisions are being demanded across the enterprise for business critical use cases, minimising time-to-value is essential. With decisionOS by causaLens, you can create solutions which optimise business KPIs in days rather than months, while minimising risk:

• decisionOS simplifies and enhances the highly fragmented Causal AI open-source ecosystem, reducing the time to value for Causal AI solutions
• We have built and refined decisionOS based on our unique experience of applying Causal AI to your key uses cases
• causaLens’ enterprise features mitigate risk in production ensuring workloads run stress free
• Track decision success using our decisionOps framework, clearly understand the impact of decisions on KPIs

For Data Scientists, causaLens offers the only unified Causal AI workflow taking you from data to decisions.

Ultimately, Data Scientists can focus on building models, experimenting, deploying models into production, and driving business value without worrying about the underlying infrastructure. decisionOS is aimed at high-performing Data Science teams offering a code-first experience in a familiar Jupyter Notebook environment. It allows Data Scientists to:

• Access the latest techniques through infrastructure that works seamlessly
• Leverage enhanced open-source capabilities
• Utilize unique Causal AI frameworks

Building a solution using open source solutions is an alternative to decisionOS.

We’ve found that organisations which take this approach to causality tend to struggle to deploy solutions for value driving use cases. There are a number of key reasons for this:

• Maintaining the infrastructure required to keep causal models in production is labour intensive and requires personnel across a range of different functions
• The open source causality ecosystem is still developing with most solutions focussing on a single aspect of the causal modelling workflow.
  A significant amount of extra time and resource is required just to ensure that the underlying packages work well together and are maintained. This is before any experimentation/ modelling work takes place
• Without appropriate tooling, building up the expertise to apply the new science of causality to business use cases, such that it drives value, takes time

Considerations for Data Leaders

For Data Leaders, the build vs buy dilemma is a familiar one. Why spend money on a vendor solution when you already have a high-performing Data Science team who could build this themselves?

In this case, we believe the build vs buy distinction is a false one. Open-source solutions can be compelling and should be utilized. At the same time, a solution based on open-source will take time to build and will introduce risk into the decision-making processes within your organization. This is particularly true in the world of causality where the open source ecosystem is highly fragmented, and covering the full causal workflow is a significant challenge.

decisionOS simplifies and enhances the highly fragmented Causal AI open source ecosystem, reducing the time to value for Causal AI solutions.

decisionOS neatly packages up years of causaLens’ cutting-edge R&D work, with open source packages in a single, easy-to-use platform. We continually develop the platform to include the latest developments in the field, and we maintain all of our proprietary code and open source code to ensure it functions as expected.

Conversely, open source solution packages are developed in a piecemeal fashion. They tend to provide algorithms for small parts of the journey, but do not cover the full causal workflow. No one else knits this all together to allow you to go from data to an application running in production powered by Causal AI.

Developing capabilities based upon the open source would mean building, testing, and refining interfaces to cover the full causal workflow, increasing time to value. Additionally, there are the usual risks associated with open source software:

• Software quality and reliability may be low
• The long-term sustainability and maintenance of packages is not guaranteed
• Licensing arrangements can be complex
• There is often a lack of production system reliability

Finally, building an entirely new infrastructure capability from the ground up is unlikely to be a valuable investment for your Engineering and Data Science team that is expected to deliver value from analytics. Estimates suggest 60% of code for a machine learning project is related to infrastructure, when starting from scratch, with only 40% used for modeling work.

Even if you decide to make this investment, there is then the risk of job satisfaction decline amongst valuable Data Scientists who would rather spend their time on exciting modeling work.

We have built and refined decisionOS based on our unique experience of applying Causal AI to your key use cases.

causaLens has years of experience using Causal AI to solve your specific challenges across different industries and has collected feedback from a range of companies. Our first-hand knowledge of how to apply the variety of techniques is reflected in the product’s capabilities.

Your Data Scientists can use this knowledge to focus their efforts on valuable use cases related to your core business, rather than spending time and effort building out and maintaining a technology platform.

causaLens’ enterprise features mitigate risk in production ensuring workloads run stress free with 99.9% uptime.

Enterprises require support features to ensure solutions meet scalability, availability and security challenges, and to ensure seamless collaboration among Data Science teams. Building these out is a significant undertaking with open source alternatives. decisionOS has a set of enterprise features which cover all of these challenges:

• SSO (Secure single sign-on)
• SLA Guarantees
• Centralized user management
• Scalable computing resources
• SOC2 and HIPAA compliance
• Data Ingestion engines with typical sources, Snowflake, SAP, Oracle, Sagemaker etc
• Enterprise support & debugging for open source packages
• Enterprise connectivity
• decisionOps (beta): Store and quantify decisions made, measure ROI from decisions

Track decision success using our decisionOps framework, clearly understand the impact on KPIs

Don’t simplytrack model accuracy, measure ROI by tracking the success of decisions. With decisionOS, embed Causal AI powered decision intelligence capable of: scenario planning, recourse, fairness, effect estimation, and root cause analysis. Application users can explore and generate recommendations while measuring the impact of decisions which have been taken. The whole process if focussed on business KPIs while at the same time building trust through high levels of interpretability and explainability. 

Track solutions at the business KPI level, rather than the model metric level

Considerations for Data Scientists

There are a number of factors for Data Scientists to consider when deciding whether to develop an in-house solution, based upon open source, or to use an off-the-shelf solution for Causal AI. At causaLens, our overall goal is to help Data Scientists deploy models to production and deliver measurable value to the business as fast as possible.

causaLens integrates the best of open source with our own innovations to provide an enterprise-ready Causal AI experience. With decisionOS, you can:

1. Access the latest techniques through infrastructure that works seamlessly
2. Leverage enhanced open-source capabilities
3. Utilize unique Causal AI frameworks

Access the latest techniques through infrastructure that works seamlessly.

decisionOS was built for Data Scientists by Data Scientists. The decisionOS team are familiar with the day to day struggles of setting up local environments, figuring out how to deploy, track and iterate models, and ultimately ensuring business stakeholders see the value of modelling work. We’ve built decisionOS so this all works seamlessly, out of the box, with no configuration required.

With decisionOS, you will have access to a wide-range of causal discovery methods, model building techniques and decision intelligence engines so you can quickly build applications for data-driven decision- making. At the same time, decisionOS simplifies many of the deployment challenges associated with production ML models, significantly increasing time to value. Reporting, versioning and building custom visualizations are all taken care of within the platform.

Additionally, we’ve optimized workflows to increase scalability. For example, causaLens’ proprietary implementation of the major causal discovery methods allows causal discovery with hundreds of variables. Open source techniques typically do not scale beyond ~10 variables.

Leverage enhanced open-source capabilities

With decisionOS, all open source solutions are integrated in a single place with a standardised interface. There is no need for your Data Science teams to research the best package for a particular application, figure out how it works, ensure the package is reliable and build the necessary interfaces to it.

Additionally, with decisionOS you can be assured that early-stage open source packages, or those designed for academics, have functional code and operate
as expected.

Finally, decisionOS contains our proprietary decision intelligence engines which orchestrate interventions and counterfactuals to transform model predictions to actionable recommendations.

Utilize unique Causal AI frameworks

In addition to cutting edge algorithms, causaLens provides frameworks that accelerate the creation of production-ready solutions:

1. Human-Guided Causal Discovery:

Open source techniques typically rely on either algorithmic causal discovery or hand drawn graphs. causaLens proprietary Human Guided Causal Discovery combines the best of domain expertise and algorithmic causal discovery, optimizing the causal discovery process. This makes the interaction between Data Scientists and knowledge experts seamless

Human-Guided Causal Discovery facilitates seamless interaction between Data Scientists and knowledge experts

2. Enhanced model building:

Building causal models which are based on causal graphs is a key step in the causal modeling workflow. Most open source packages do not cover both the causal discovery and causal modelling steps. Those that do are usually restricted to use of a single model or small set of models. Additionally, based on the training techniques used, these tools often fall into the trap of producing biassed edge relationships.

With decisionOS, we’ve optimized the model building process through our proprietary, versatile and powerful CausalNet model which captures complex functional relationships between variables. CausalNet is trained using causaLens’ DoubleML approach, guaranteeing that edge relationships are debiased from backdoor paths.

3. High-frequency data support:

Our unique computational engine for data streaming supports high frequency data & allows models to be updated efficiently. This technology has been stress-tested in some of the most challenging environments.

4. Dara: causaLens app building framework:

Finally, we make it much easier to ship solutions to the business using our proprietary app-building framework, Dara. Dara is the only app-building framework that is optimized for Causal AI. You will see significant impact from your work as models can be packaged up so that a business stakeholder can seamlessly interact with them, ensuring Causal AI is applied to the business’s most pressing challenges

Conclusion

decisionOS offers a fully-supported, enterprise-ready experience to support your critical decision-making processes.

Building up a Causal AI capability is a significant undertaking which will increase the time it takes for your enterprise to make better decisions, while introducing risk into the process.

Organizations that use decisionOS make data-driven decisions faster while removing much of the risk associated with building out a new capability. Data Science teams are able to seamlessly leverage all of the latest causality research, quickly deploying models to production and significantly reducing the amount of Data Science time allocated to repetitive tasks.