Historical clinical trial data frequently hide extremely valuable insights on alternative, or more targeted applications of known active substances. A leading pharmaceutical company asked us to develop a new tool to systematically investigate historical data.
To unleash the potential of historical datasets, we tested, tailored and implemented a variety of machine learning techniques such as PRIM (Patient Rule Induction Method). We also developed a new, proprietary tool with a web-based interface.
The new tool is now in daily use in the research facilities of the client. It has already indicated a number of important leads, being considered for new clinical trials.
A leading international technology company was planning expansion in a new European region. The planned promotion needed to combine digital and traditional channels, and to reach well-defined, measurable targets in a short time for a limited budget.
In addition to in-depth understanding of available market data, we implemented a proprietary digital market research tool, in collaboration with a partner company. The promotion approach was then optimized by machine learning, by applying micro-testing and removing bias in collected data through state-of-the-art statistical modeling.
The proposed promotion action plan, including market segments, value proposition to be communicated, channels and budget, is being rolled out by the client exactly as recommended.
A Fortune 500 client had a challenge of forecasting yearly spend of a portfolio of 600+ projects with annual budget in excess of 300 million USD. We were hired to understand the root-causes, and to fix the problem.
We first analyzed the client operations, and in particular budgeting, financial reporting and project management practices and procedures. Based on this, we developed a new statistical approach and tool to forecast spend of individual projects, based on a number of predictors. In addition, we proposed new reports, tools, and changes to the existing processes in finance and operations.
The tool and the process changes have been fully implemented. As a result, the client now plans, approves and delivers 20% more projects for the same budget, with at least proportional increase in delivered value.
The hospitality industry is undergoing a dramatic change, with rapidly increasing importance and market strength of the new distribution on-line sites such as booking.com . To achieve optimal, or even reasonable occupancy and price levels, existing manual processes are thus often neither agile nor accurate enough. A leading hospitality company hired us improve it to partially automate pricing and capacity managenent.
We first mapped in detail existing pricing and capacity management processes, and understood opportunities by studying historical data. We then developed an automated revenue management tool, with key levers: forecasting, optimizing and managing overbooking; allocating capacity accross room types and distribution channels; and optimizing prices, all in real time.
The tool is now operational, managing the entire accomodation capacity of the client. It partially automates revenue management decisions in real time. The added net present value of the project is estimated to exceed value of developing two entirely new hotels.
Here is a collection of articles from McKinsey & Quarterly on various strategic aspects of big data, as well as on intriguing case examples.
McKinsey Quarterly reports on a machine-learning approach to venture capital. The model recommending the deals already contributes to the decision making. The main break-through seems to be in collecting sufficient amount of quality historical data enabling modelling.
While forecasting of fashion trends by machine learning is still in its infancy, The Economist reports that the industry of fashion forecasting is approaching the "tipping point" driven by AI advances. The winners leveraging it will be both on the retail side, by correctly forecasting trends, as well on the supply side, by more effectively managing supply chains.
Harvard Business Review in its cover story reviews the current applications of machine learning in business, and predicts some future trends. It is a great introduction for business leaders with no practical AI experience, but with awareness of its likely importance.
A 2014 paper in McKinsey & Quarterly gives a C-level overview of applications of machine learning techniques in manufacturing. There is a particularly insightful matrix of case examples and relevant data analysis techniques across industries.
A Harley-Davidson dealership has improved sales by measuring and optimizing marketing campaigns. Harvard Business Review tells a story how they rely on Albert, an AI tool developed by Algorithms. The key seems to be using the tool incrementally, to drive experimentation and quantify outcomes.
Modern machine learning methods can give you statistically useful predictions in many business situations, even when the source data is relatively scarce and unstructured. Here is a winning example from a Kaggle competition on predicting house prices.
More on the gradient boosting algorithm used there by a leading data practitioner.
McKinsey survey of senior executives indicates that, while they have high hopes of their data and analytics programs, they report only mixed success so far. Some insights on the key success factors can be found here.