We shape and drive AI transformations by building AI solutions. We develop your most promising AI use cases while also building the required technology capabilities. We build AI production platforms to operationalize and manage algorithms at scale. And we ensure business adoption of algorithmic decision making by implementing applications and process change.
During the development stage, a new algorithm is developed to optimize a certain decision process. During this step, we structure the business problem and identify the value drivers. We analyze data patterns to arrive at an analytical understanding of the output (decision) to be predicted, the input (data) to be used and the corresponding machine learning technique to use. Our team of specialized data scientists subsequently develops a first version (MVP) prediction model to be put into production.
Having developed an algorithm with adequate prediction power to create value in a business process, this algorithm needs to be operationalized or “brought into production”. We establish Model Engines as fully standardized components to train and score algorithms in fully automated AI pipelines. Our Model Engine allows us to run and maintain thousands of algorithms in parallel. The Engine orchestrates periodic retraining of algorithms to create truly self-learning AI systems. We use a containerized architecture with on-demand scalability of computation power to make this possible.
Once algorithms run in production, they need proper performance management and quality control. This involves monitoring and continuously improving prediction power. Understanding prediction performance in operational situations is different from laboratory settings. Also, it creates insights into when performance is generally good, or generally challenging. Vital input for users to decide when to trust AI, or when to add human judgment. We manage performance by closely tracking KPIs on prediction power, and comparing them to various benchmarks: historic performance, targets, and different algorithms running in shadow mode.
We support our clients in assessing their Data Analytics Technology stack on the ability to execute the highest impact use cases at scale. To validate use case execution readiness, this assessment will result in a Data Technology gap analysis and Data Technology improvement plan on five layers: data storage layer, data integration layer, algorithm layer, application layer and data governance layer. Depending on the outcomes, we support our client in building components that are critical for use case execution
Once a use case MVP has successfully launched it requires a scalable Technology stack. Our Data Platform integrates different layers of technology to allow for speed, reliability, control and scalability in driving use cases. Our standardized components are developed across industries and according to the Infrastructure as Code principles. We work either within existing architecture and technology, or we help set up new platforms. We are cloud-agnostic, and our certified engineers work with all major cloud vendors (AWS, Azure and Google).
One of the critical success factors for creating lasting use case impact is a scalable data architecture that feeds use case applications and BI reporting. We support our clients in designing and implementing a data architecture that allows for fast and flexible data extraction, creates one integration layer with uniform definitions and is the fundament for continuous use case innovation. We are flexible to work across most common data technologies and do not have any vendor lock-in. We help our clients expanding their existing data landscape or work with choices already made.
Algorithms need a vehicle to interact with business processes. Without it, they risk to remain isolated phenomena as experiments or mathematical prowess. We integrate output of algorithms to existing systems or build new AI-powered applications.
AIR-Pro: “AI for Retail Promotion Optimization"
Promotions form a significant share of retail sales in most categories. And this share has steadily grown over the past decade, now comprising up to 40% of all sales in some cases. Optimizing promotion ROI is therefore vital for driving both growth and profitability. This is a notoriously difficult challenge for most retailers. Getting a firm grip on all promotion effects to establish sound post promotion ROI is already a big plus. Let alone taking an automated and predictive approach to designing new promotions. Our AI for Retail Promotion Optimization app (AIR-Pro) solves this challenge by deploying a range of AI algorithms to measure and predict 11 promotion effects on customer lifetime value. The Model Management functionality further enables promotion managers to work in a symbiotic relationship with the app, combining human experience and machine learning in optima forma.
Smart ROCE: “Optimize returns of Telco network investments”
Network investments are increasingly important for Telco’s, in order to be able to deliver required bandwidth to increasingly demanding customers. However, these investments are costly and labor-intensive, so prioritization decisions need to be made, and this is a hard task since it requires upfront knowledge on value creation and costs per area and technology choice.
Smart ROCE, MIcompany’s new AI application, helps Telco’s overcome these challenges by using AI to model, per area, value creation, costs, and the ratio between them, called ROCE (Return on Capital Employed), so prioritization decisions can be made fact-based. Moreover, the application facilitates the monitoring of value creation after investment, enabling Telco’s to determine initiatives to stimulate investment monetization in areas where value creation is not meeting its target.
“Change as the new constant”
One of the most challenging barriers to impact is the human factor. Sure, the technical aspect of developing a high performing AI solution is not trivial and requires specialist expertise. But getting humans to change their way of working with new tools is just as tough. Marc Andreessen famously said that “any new technology tends to go through a 25-year adoption cycle”. This is true for most new technology adoptions, but AI presents some unique challenges. Our approach is designed to explicitly address these, and shave some years off the 25-year benchmark.
- Algorithms are not humans. Algorithms are not aware of any contextual information, other than captured by the data. Experience is only present as far as training data is concerned. Moreover, they are only right “on average”. Predictions can be off in specific instances, even if the overall performance is far above the human average (consider a self-driving car). It takes time and deliberate collaboration to trust someone — or rather something — “on average”.
- Algorithms do have human traits: they are not flawless. Their performance cannot be measured in discrete terms. Tooling which works or does not work is easy to deal with. If it works, we can gradually get used to it and adopt it as a new reality. If it works in some cases, but not so well in others, an algorithm may display human traits to some extent: not being equally good in all situations. This characteristic increases the need for understanding and collaboration, before full adoption can take hold.
- AI solutions take time to mature and require human-machine collaboration to improve. New solutions need to be integrated with existing systems. New data is added over time. Algorithms are improved through feedback loops. Many reasons why the MVP is not yet the next big bang, although it can be over time.
- Required changes in processes are often not binary, or step changes, but evolutionary. As AI solutions evolve, so will the process they are part of. This means that “change” will be the new constant in many cases of AI applications. A call center of a Telecom Operator may first automate the handling of customer complaints. But further evolution can make a call center redundant altogether, if issues are predicted and preemptively addressed.
Our transformation approach
We partner with our clients to identify their highest-impact use cases, align them with strategic objectives and jointly develop a roadmap. This roadmap stipulates how the various use cases will be developed over time, and how to simultaneously build the supporting capabilities. We act as change agents by working in joint multidisciplinary teams, getting the organization to adopt and drive new AI-powered solutions and ways of working. To get from the first idea to full impact, we work with our clients on developing AI use cases along all stages of maturity. We help incubate, operate, scale and improve. This lifecycle approach has been honed through practical experience with dozens of use cases for leading organizations over the past 13 years. It is designed to address the most critical barriers to impact and sets our clients up for long term success.
“Buy one, get one free… Don’t miss this special offer: click here for 2 months premium membership.” Promotions. Retail is hard to imagine without them. Spending on promotions in Europe has doubled over the past 10 years, with nearly 30% of purchases being on promotion.