Causal AI: A Deep Dive with causaLens' Director of Product

Introducing Causal AI

In a world where generative AI and machine learning tools are fast proliferating, causaLens stands out as a vanguard of innovation. The company envisions a future where humans can entrust machines with our most pressing challenges across various sectors – be it the economy, society, or healthcare.

causaLens isn’t merely content with artificial intelligence as it's known today; it pioneers Causal AI, a significant leap forward in machine intelligence. This evolution in AI doesn't just predict outcomes; it comprehensively understands the whys and hows, thereby empowering decisions with far-reaching positive impacts. Through its recently launched Causal AI platform – decisionOS – causaLens offers an intuitive interface that caters to a broad spectrum of users, allowing them to make decisions that are superior and trusted.

We had the privilege of delving deep into the journey with Tom Farrand, causaLens’ Director of Product, with whom we discuss the challenges and the future of this groundbreaking product.

The Genesis of decisionOS

Molten Ventures: Where did the idea for decisionOS come from? Was it a natural evolution for causaLens?

Tom Farrand: The idea of using causality for decision-making is not new—far from it. As far back as the 1700s, physician James Lind conducted clinical trials to treat scurvy and decide on the best medication. Similarly, in Victorian London, John Snow employed causal techniques to identify the source of cholera and then determine its treatment.

More recently, pioneers like Guido Imbens have received the Nobel prize for their work in understanding how social policy changes can impact different socioeconomic groups, informing future policies.

This rich history of causality, combined with our work with customers in the field, gave us the confidence to shift our thinking about the product: building atop our collection of data science tools for answering tricky statistical questions to a product our customers can use to inform their most critical decisions.

MV: How does decisionOS differ from causaLens's previous offering, Causal Inference Engine?

TF: For the past seven years, causaLens has been a key player in the AI industry. Over this period, both the field and our company have seen tremendous evolution. We've observed the rise of the data scientist role, often called the "sexiest job of the 21st century" (a title many data scientists might suggest is taken with a grain of scepticism). We've also witnessed numerous trends, with the most recent being LLMs (large language models). In response to these changes, causaLens's offerings have continuously adapted. The previous product focused on one of those trends: AutoML (automated machine learning) where users input their data, and the product automatically analyses it to build a suitable model.

However, this presented challenges:

Trust: Many clients found it hard to trust the results, given their inability to delve into the process.
Scope: Our ambition to influence real-world decision-making showed that solely focusing on data science tools was narrow.

Our latest offering, decisionOS, emphasises bridging the gap between domain experts and data scientists. We prioritise ensuring that decision-makers can easily interact with and comprehend data outputs. Building this trust, especially for business-critical decisions, is paramount.

Building decisionOS

MV: What were some key challenges in developing decisionOS?

TF: One of the most significant challenges has been a mindset shift. Since its inception, causaLens has led in the realm of Causal AI, boasting a highly skilled and experienced data science team. However, the transition from a data science toolkit to a decision-making platform powered by causality has ushered us into unfamiliar territory.

Now, deep collaboration with our customers is essential to truly understand their business intricacies and decision-making processes. Strategising for this, and discovering the best ways to aid our customers, is an ongoing learning experience.

Compared to B2C or product-led growth offerings, we engage with a select group of customers. This means that while we receive feedback in smaller volumes, we gather a richer context alongside it. Understanding individual user goals, the KPIs they aim to achieve, and the datasets they use is crucial. As such, customer interviews are our most effective feedback tool. We further complement this with product analytics to monitor how new features are adopted, and we extensively test new features internally.

MV: What real world applications of decisionOS have you been able to realise since its launch?

TF: One of the joys (and challenges) I find with decisionOS is the vast array of use cases it addresses. It offers immense value to customers, but deciding where to focus and how in-depth to delve into a particular use case poses challenges.

Below are just some of the use cases we have developed this year:

Challenges in the industrial sector: optimising wind turbines on offshore wind farms or identifying root causes of manufacturing line faults and rectifying them.
In the consumer goods industry: discerning the relationship between product prices and demand while accounting for biases introduced by global events and competitor actions.
Financial institutions: setting up their products and partner networks appropriately.

Personal Learnings

MV: What are some of the lessons you learned from your experience as the Director of Product for decisionOS?

TF: Prioritisation is probably the single hardest challenge I face. It’s incredibly difficult. This is true of any business, but it's compounded in several ways at causaLens:

We are an enterprise B2B with a handful of customers. This means that almost all the data upon which we rely to make prioritisation decisions is anecdotal.
We are introducing a novel technology in a rapidly evolving space. Therefore, there's limited historical precedent to draw upon, and conditions are constantly changing.

To address these challenges, we have developed several internal techniques:

Maturity stages: Different maturity phases, such as prototype, beta, and production, are defined in our roadmap to guide the amount of refinement and effort required for a feature. The time we aim to spend working on each stage during a quarter is then roughly divided. For instance, if prototyping is allocated 50% of the time for the next quarter, we have a guide on how many ideas can be accommodated, while relaxing our prioritisation standards for those ideas. We've found that this overall time allocation, motivated by strategy, makes prioritising specific activities from the backlog more effective.
Clear incremental goalposts: We aim to incorporate meaningful milestones into each project, giving us opportunities to pause and assess whether we are delivering value. The closer together these goalposts are, the faster we can learn and consequently, deliver value.
No is not an opinion: If a prioritisation decision hinges on opinion, this rule comes into play. It's not sufficient to just disagree with a prioritisation; one must propose an alternative solution. This is crucial to avoid decision deadlock.

While these techniques don't resolve all our prioritisation challenges, they offer a clear path forward and help prevent decision paralysis—most of the time.

MV: What advice would you offer to other Directors of Product developing a key product?

TF: Due to the varied nature of product management roles, giving a one-size-fits-all piece of advice is challenging. To me, product management is fundamentally about people. As a product manager, to accomplish tasks, you must craft a compelling vision of what should be built and why. This typically combines strategic foresight and customer insights. You then need to guide and influence the realisation of that vision amidst competing priorities and distractions.

Throughout both tasks, you'll be shoulder to shoulder with a diverse group of stakeholders. Therefore, it's crucial to understand the motives of different individuals. What makes them tick? What are the three top business challenges for our biggest customer? How is our champion going to get their next promotion? Is our sales team incentivised more to close net new business, or upsell existing accounts? What is the reporting structure of the engineering organisation?

Looking Ahead

MV: What are the future plans for decisionOS?

TF: One of our next ventures is the development of what we term decisionOps. In recent times, there's been a substantial emphasis on transitioning machine learning from the lab to production environments. This has given rise to machine learning operations, or MLOps, and a flourishing ecosystem of tools centred on the effective monitoring, deployment, and management of machine learning or AI models and data.

However, we've observed that while enterprises are now more equipped to set up their infrastructure owing to MLOps, they often can't confidently assert that these endeavours yield genuine business ROI. This results in a credibility gap where data science teams talk in terms of accuracy, recall, and precision, while stakeholders expect KPI impacts such as return on ad spend, factory throughput, and revenue.

Therefore, two primary forces drive decisionOps' development:

The first is the credibility gap. To bridge this chasm between model metrics and business KPIs, there needs to be a causal understanding of how the two relate, coupled with a grasp of the broader business landscape.
The second is that typical AI predictions aren’t actionable. Predictions indicate outcomes but don't guide actions to influence those outcomes. Hence, in practice, a decision-making process is necessary, usually a blend of rules and predictions.

decisionOps is our solution to these issues. It disentangles these processes, enriches them with causal models that surpass mere predictive insights, and ties decision-making effects back to the KPIs our clients value most.

It's undeniable that the world of AI is ever-evolving. With pioneers like causaLens leading the charge, the horizon seems not just hopeful but full of promising, ground-breaking innovations.

decisionOS is a testament to the capabilities of Causal AI and its potential to revolutionise the decision-making landscape across diverse sectors. As we peer into the future, it's evident that causaLens's journey has only just begun.

**causaLens is currently spearheading an early access program for decisionOps. If these challenges resonate with you, please reach out to the team.**