AI for new product development has applications across multiple industries including healthcare and industrial. In the healthcare vertical, pharma considers a combination of compounds to produce drugs of a particular activity, safety, and delivery while meeting production cost and volume considerations. In the industrial vertical, metallurgy considers a combination of different metals to produce alloys of particular tensile strength, corrosion resistance, and fracture toughness.
New product development is a process, infamous for how time-consuming it can be, noted by some to sink up to 50% of development time.
In early phases of the process, several potential product opportunities are established, while in later phases, commitments are made for allocation of resources to those potential products – potential future failures are therefore quite costly. The lack of transparency and foresight during this phase of product development is painful in both time and resource. In industries, where speed and flexibility in product development is a strong competitive advantage, where new product development can account for a large portion of a company’s profits, and in the case of the pharmaceutical industry, where survival depends on new drug development, new product development is of paramount importance to a company.
How artificial intelligence can help
Artificial Intelligence (AI) allows for robust digital testing and predictions of prototypes before a company sinks time and resources into physical product trials. Often, potential products vary on a range of traits. However, it is difficult to synthesize the impact of the different traits into a single “go” or “no go” decision. Is the shelf life going to be as important as the bioavailability for a certain drug to price and sell, and what shelf-life with what rate of bioavailability is going to be optimal? AI can offer recommendations on what would be most successful for pre-selected end points (likely to produce certain side effects, likely to be toxic to cells, likely to produce certain bioavailability).
Furthermore, AI can be used not just to predict traits of a product, but the traits of the product given constraints such as the conditions of a company’s suppliers and the existing distribution channels, thereby incorporating the evaluation of a product’s go or no go within the constraints of the real world and the multi-parameters of an actual product launch.
Emphasis on technological rigor
Recognizing the importance of new product development, Petuum provides an AI service of technical rigor and enterprise readiness.
We start with ensemble modeling, which creates a virtual technology counsel for the decision itself, allowing different types of models to vote. Ensemble models are famously known to outcompete single classifiers and are noted to reduce bias and variance. Common base models that Petuum uses range from both simple time-series, SVR, decision tree, and linear models to complex n-dimensional deep neural networks. The diversity of such models allows the development of an ensemble model that has strong generalizability. Currently, many companies deploy “classical” rules-based or simple fuzzy logic based solutions which have limited ability to be self-learning. Petuum’s distributed deep-learning capabilities provide the ability for these models to be continuously self-learning.
At Petuum, relationships are also discovered organically at the level of traits and do not require pre-populated or labeled understandings, providing a new way of looking at the data and opens up new avenues of product development.
Additionally, model inputs can span not just a range of traits but also a range of structured and unstructured data types (text, csv, streaming data). This flexibility of inputs allows an organization to consider a finer representation of what defines their potential product. Our models will also account for changes in initial traits and for changes in the relationship of inputs and outputs in non-static systems, building a model that is more adaptable and relevant to the needs of the product development process.
By operationalizing such models, Petuum can help companies be faster to market and improve their competitive position. Today, screening of potential products can take months to years and involves a lot of manual labor and guesswork. An intelligent assist can reduce the time it takes to find that next product and reduce the risky gambles made in the nascent stages of product development.
Feyzioglu, Orhan, and Gülçin Büyüközkan. “Evaluation of new product development projects using artificial intelligence and fuzzy logic.” World Academy of Science, Engineering and Technology 11 (2006): 183-189.
Paul, Steven M., et al. “How to improve R&D productivity: the pharmaceutical industry’s grand challenge.” Nature reviews Drug discovery 9.3 (2010): 203.
Rao, S. Subba, et al. “Artificial intelligence and expert systems applications in new product development—a survey.” Journal of Intelligent Manufacturing 10.3-4 (1999): 231-244.
Takeuchi, Hirotaka, and Ikujiro Nonaka. “The new new product development game.” Harvard business review 64.1 (1986): 137-146.