Authors:

Claire Thomas Gaggiotti & Matteo Aragone

 

2021 was a challenging year all round, with the economic and social implications of a continuing pandemic and intense climate change summits. A new year is an opportunity to look forward with optimism to new trends and initiatives.

In this article we’ll present what we feel are the six main trends and buzzwords for 2022 in digital technologies and AI.

1. Data-driven culture
2. Responsible AI
3. Explainable AI
4. AI for Sustainability
5. Citizen data science
6. Data quality
   
   

1. Data-driven culture

Practically all organizations aspire to becoming data-driven, but according to a survey conducted by Forbes contributor, Randy Bean, only a quarter of executives feel they have succeeded in bringing about this shift, stating that the problem is cultural, not technological.

Apart from the commitment of all levels of management, and sufficient budget and resources, it all comes down to a question of trust. Many managers are used to taking decisions based on gut feelings, which have probably mostly served them well so far, and are afraid of losing control to geeky data scientists. Many are also wary of the upheaval of implementing new systems and reviewing all current processes.

So it’s vital to choose the right tools, which will provide quality insight, while promoting collaboration and cooperation, such as between business and IT experts. When decision makers realize they are receiving data-driven support, and not losing their prowess to indecipherable algorithms, many more organizations will be able to successfully embrace a data-driven culture.

The process is not instantaneous: it requires investment in training – be it technical or not – in people, and, above all, in the necessary time to onboard all stakeholders. But it delivers results in terms of quality of work and improvements to process management.

2. Responsible AI

As AI permeates our everyday life, recommending films to watch, products to buy, and books to read, AI tends to be seen as a fun technology, which can improve the quality of our free time.

However, AI is also used in far more serious scenarios, such as deciding whether you’re likely to pay back a loan, or make legitimate insurance claims! Responsible or ethical AI deals with those cases where AI is used to define the moral behavior of people.

It’s not a new concept, in 2019 Europe had already produced guidelines on ethics in Artificial Intelligence, but as more cases of unfair bias come to light, including machine bias against blacks in US court sentencing and racial bias in healthcare, and hot topics of diversity and inclusion continue to demand attention, responsible AI is a major buzzword.

Although many industries are already putting ethical AI policies in place (89% of industrial manufacturers in 2021), implementing policies is a real challenge.

The initial data used to train algorithms must be free of bias, and “data scientists lack the training, experience, and business needs to determine which of the incompatible metrics for fairness are appropriate,” (Reid Blackman, CEO Virtue).

The rules produced by AI algorithms must then be closely analyzed to remove bias before being applied to new data, which means they must be clear and legible, and this brings us on to the next buzzword and big trend of 2022: Explainable AI.

3. Explainable AI

The general concept of AI is that it is extremely complicated and intrinsically inexplicable. This was certainly true in the not so-distant past, when AI algorithms were mostly black-box, producing indecipherable mathematical equations. But as we have already discussed, responsible AI means providing transparent explanations of fair play, which is often a legal requirement, so the need for transparency goes hand-in-hand with the need to ethically mitigate bias.

According to Forrester, 20% of enterprises will already rely on explainable AI in 2022, and this percentage is destined to increase vertiginously. Are there explainable AI solutions on the market ready to meet these needs, where performance is not downgraded, and additional layers of software are not required to make their results comprehensible?

Luckily the answer is yes. One of the main players in this field of innately transparent algorithms is Rulex’s Logic Learning Machine (LLM), first developed in 2014 by expert mathematician Marco Muselli. LLM produces results as totally legible if-then rules, and yet has the computational speed and accuracy of a black box solution.

4. AI for Sustainability

It is a great relief to add sustainability to the list of buzzwords for 2022, as the health and wellbeing of future generations depend on it.

In 2022 corporate sustainability is very much focused on the concrete actions which companies can take to combat climate change by reducing their carbon footprint. Over 200 of the world’s largest companies have already taken The Climate Pledge, with the aim to reduce carbon emissions to net zero by 2040.

But what is the role of digital technologies in this pledge?

According to The Royal Society report “Harnessing computing to achieve new zero”, digital technologies, such as machine learning and AI, could deliver nearly one third of the carbon emission reductions required by 2030.

According to the United States Environmental Protection Agency, the transportation sector is the main culprit. In 2019 it was responsible for 29% of greenhouse gas emissions, and 83% of these emissions came from small and large trucks.

There are long-term plans to reduce emissions through the adoption of electric and hydrogen trucks and biofuel, but in the short-term, digital technologies can already do a great deal by simply reducing the number of trucks on the road. This can be achieved by identifying the best routes and travel times and maximizing truck loading.

Rulex Axellerate is an innovative example of how transportation can be optimized, as it is able to consider both current and future orders, to not only identify the best shipment time, but also combine deliveries, consequently filling all trucks to their maximum capacity, and reducing the number of trucks and their carbon emissions.

5. Citizen data science

Processing small amounts of data on two or three spreadsheets is within most people’s reach, but aggregating large quantities of data, processing with predictive modelling, and process optimization involves bringing in highly specialized data scientists and data analysts. These profiles use traditionally complex tools, often based on programming languages like SQL, R, Python, and C++, which go far beyond the average business expert’s remit.

As a result, a series of no-code tools (including the Rulex Platform) have appeared in recent years, enabling advanced operations (data pre-processing, machine learning, optimization…) using visuals such as workflows. These tools are simple and intuitive enough to be used by business profiles (also called citizen data scientists or citizen developers), but are powerful enough to be appreciated by data scientists, data analysts and IT personnel.

They enable different company departments to work together, with significant improvements in terms of results and time savings – amounting to 4.6X productivity gain over traditional programming according to research by No-Code Census.

And companies have taken note: the low code market is growing annually by more than 20% , according to estimates by Gartner.

6. Data quality

We’ve covered many different aspects so far: corporate culture, processes, technologies, tools… but one obvious and utterly essential prerequisite is missing… What’s needed are data, specifically, clean and reliable data.

Two years of Covid have taught us the importance of clean and consistent data. We’ve all read data-driven conflicting predictions by scientists, physicians and research institutes. There are many reasons for this: data may be inconsistent (based on different assumptions, or samples), or “dirty” (incomplete, with input errors) and therefore difficult to read. The lower its quality, the more data is open to different interpretations, and consequent incorrect predictions.

A similar scenario applies to corporate data. Bad data can lead to inefficient operations, lost profits and mistakes in strategic decision making. It’s estimated that large US enterprises lose 15 million dollars a year because of bad data. So it’s essential to prioritize the data quality process, from data creation to storage, guaranteeing accuracy, accessibility, completeness, consistency, validity, and uniqueness. The road ahead is clear: according to Gartner, by the end of 2022 an impressive 70% of companies will be closely monitoring the quality of their data.

One of the most innovative aspects is without doubt augmented data quality, which makes it possible to suggest automatic adjustments and correct data. Rulex RDC, for example, tracks down data inconsistencies even where correlations seem impossible to find, and even suggests data corrections that business experts can accept, refuse or improve.

We all know that data-driven decision-making processes can bring huge benefits to companies, such as greater confidence in taking decisions, more transparency, and a higher chance of improving in the long term.

Gartner estimates that more than 33% of large organizations will have analysts practicing decision intelligence by 2023, and 75% of enterprises will shift from piloting to operationalizing AI, driving a 5X increase in streaming data and analytics infrastructures by the end of 2024.

But at the same time, businesses often have doubts on how to deliver data-driven decision-making projects successfully.

In a previous article, we talked about how important it is to choose the right project when implementing a data-driven decision-making process. We said that businesses should think about the goals they want to achieve, consider the available data, and the actionable solutions that might be implemented thanks to their data-driven knowledge.

Let’s now assume we have identified the project we want to put in production.

HOW TO STRUCTURE DATA-DRIVEN DECISION-MAKING PROJECTS

Once the project team has been created – it should be balanced, well defined, and with all the stakeholders involved – we advise you to include 3 phases:

• Project Scope (<1 week)
• Pilot (between 4 and 6 iterations, each iteration lasts for one week)
• Production deployment (between 2 and 4 weeks)

The phases can be tailored according to your needs but, in principle, we recommend you follow them as closely as possible.

Let’s imagine your data-driven decision-making project is highly complex and requires many months of work to get to the final results.

We suggest you reduce the initial scope to give you a first deliverable in 2 or 3 months.

This has two major advantages:

• You focus your efforts on something immediately usable
• You build trust in all stakeholders, who see results from the outset.

One of the advantages of Rulex is that it is extremely flexible and interactive thanks to its drag and drop functionality and eXplainable AI. As a result, it delivers value from the very first iterations.

The graph below sums up the life cycle of every Rulex project: at an early stage, little effort is involved and the return in value is already greater than the investment.

Over time, the management effort grows slowly, while the value of the solution grows exponentially.

HOW TO DEFINE YOUR PROJECT SCOPE

The scope phase is crucial to the success of any data-driven decision-making project.

If you make a mistake in this phase, for instance, setting an incorrect goal or misunderstanding your data, the project might be seriously compromised.

Typical activities during the assessment phase are:

• Goal definition: set out the goal in detail, identifying performance measures and expected KPIs
• Data definition and audit: identify the available data, evaluating data quality and suitability for analysis
• Deployment in production: plan how the new decision-flow will be deployed in production, e.g., how it will take place within the daily work routine and how it will interact with all company processes
• Assessment: evaluate whether objectives can be met or if corrective actions should be taken

If the assessment is positive, project sponsors should sign-off so you can move to the pilot phase.

DEVELOP YOUR PILOT, FOCUSING ON VALUE CREATION

In the pilot phase, the team should provide proof of concept, and clients should see the first results. With conventional methods, customers expect these steps to take several months, requiring a large number of billable hours from data scientists, and delivering little or no directly actionable insights.

With Rulex, customers will quickly receive tangible business value in the form of understandable, actionable insights.

In this phase, typical activities are data preparation, model creation and refining, and dashboard creation.

DEPLOY IN PRODUCTION FAST AND SUCCESSFULLY

In the final stage of the project, the proof of concept will be put into production, integrating the solution into the customer’s IT infrastructure.

The scenario for deploying the proof of concept strongly influences implementation activities. Among other factors, two main aspects are critical to determining the complexity of this phase:

• How different the input / output format is compared to the format used in the pilot phase
• The solution’s level of autonomy with respect to client IT systems

The good news is that with Rulex, production deployment is much easier than traditional methods, thanks to its no-code approach and numerous project monitoring features.

Data-driven decision-making (DDDM) is revolutionizing almost all industries and departments. Experts agree that those who now embed DDDM into their organization will reap significant benefits and competitive advantages in the future.

This brief article looks into the topic while providing some useful tips on how to choose the DDDM project that is best for your business.

WHAT IS DATA-DRIVEN DECISION-MAKING?

Data-driven decision-making is the process of making decisions by analyzing the data at your disposal. DDDM processes may be characterized by fully automated data-driven decisions, generally driven by artificial intelligence and optimization algorithms. We are faced with this type of decision every day, for example, when we use a search engine to find something on the internet or when we look for the best way to get from A to B on Google Maps.

However, we aren’t always faced with this type of decision. When a decision involves granting a loan based on the applicant’s risk profile or launching a new product based on a market study, we cannot rely exclusively on an algorithm.

Data analysis can certainly help decision-makers take a more informed decision, but it is they who have the final say.

WHY DATA-DRIVEN DECISION-MAKING?

Data-driven decisions bring numerous practical benefits.

Frist of all, when driven by data, the decision-making process relies on concrete evidence rather than personal opinions (which can sometimes lead to long discussions, if they are not shared by others). Furthermore, data-driven decisions are transparent and reliable, and they can be continuously improved by new data.

Data-driven decisions enabled by emerging technologies are having a major impact on almost all industries (financial services, healthcare, manufacturing) and all departments (marketing, supply chain), bringing important benefits in terms of cost savings, revenue growth, and margin optimization. Many repetitive decisions can be partially or fully automated, leaving the more complex and strategic tasks to humans.

Recent studies confirm these trends. According to Gartner, more than 33% of large organizations will have analysts practicing decision intelligence by 2023, and 75% of enterprises will shift from piloting to operationalizing AI by the end of 2024.

So far so good! Fast automated decisions, lower costs, more revenue… isn’t that wonderful?

Yes, but it’s not so easy.

Many data-driven projects fail. Why?

There are various  explanations. Sometime projects fail because of a poor choice of technology – for example, a complex tool might create barriers between business and IT experts – and at other times, because of a bad choice of use-case.

Before starting a new project, it is crucial to ask ourselves some questions to avoid making the most common mistakes.

1. IDENTIFYING A CLEAR BUSINESS GOAL

The first ingredient for successful projects is identifying a clear business goal. This should be specific, realistic, with measurable objectives and a limited time frame. You can start “small”, then extend your goal after you have gained more experience.

Let’s think of two examples of well-defined business goals:

• Improving revenues by 3% in the first quarter of the year through dynamic pricing.
• Reducing the daily cost of your supply chain for transporting goods by 10%.

Conversely, increasing sales is not a good business goal since it is hard to measure and poorly defined.

2. UNDERSTANDING IF YOUR DATA FIT THE PROBLEM

Are your data representative of the problem you want to solve?

Imagine this scenario: you sell products in France, and you want to use this experience (and its data) to launch your products in Italy, assuming that European markets are similar. The Italian and French markets are probably alike (it’s up to you to decide), while applying French market data to a very different market is not a good idea.

But data representativeness is not the only issue.

One of the most common and often underestimated problems is data quality.

The expression: “Garbage in, garbage out” has become popular. If you have poor-quality data (tables full of typos, missing values, etc.), it will be hard to get something good out of them. Fortunately, there are many data quality tools, such as Rulex RDC, which achieve  and maintain data accuracy at 100%.

3. DEFINING ACTIONABLE SOLUTIONS

Can data-based predictions lead to actionable solutions, and then to achieving your business goals?  

If the answer is no, you should think again.  In the past, many AI projects were launched out of pure curiosity, without an actionable solution, and often ended up with disappointing results.

Imagine you create an AI solution that tells you the exact number of goods needed in the warehouse day to day, but you have no resources to move goods in and out. So, your information turns out to be completely useless from an operational point of view. The new solution, perhaps ordered by a curious manager, risks putting artificial intelligence in a bad light as something superfluous and time-consuming.

Carefully choosing your new project builds trust in new technologies.

4. DESIGNING A USER-FRIENDLY, INTEGRATED DATA ENVIRONMENT

It is highly advisable to design an integrated technical environment, enabling different professionals (technical experts, business analysts, and IT people) to work seamlessly without having to go through a fragmented set of solutions.

Easy-to-use, no-code tools such as Rulex Factory allow you to manage the entire data value-chain, from integration to machine learning, from optimization to visualization, from deployment in production to monitoring.

5. INVESTING IN TRAINING

Lack of competences can cause misunderstandings and delays in delivering a project.  It is therefore important to invest in training to ensure the project’s success and avoid wasting time. Full training improves communication between corporate teams, as it allows professionals from different fields to understand the project and work seamlessly on it.

Rulex Academy is a good starting point for reinforcing both technical know-how (from data preparation to machine learning, from optimization to dashboarding and maintenance in production) and business know-how (particularly in financial services and the supply chain).

Reducing churn: a real example

In the previous section we focused on theory. Now it’s time for some real examples. Let’s consider a real use-case where we are trying to reduce customer churn – a recurrent problem in many industries such as energy, retail, finance, etc.

• The business issue is very clear: the churn rate is high (12%), causing significant revenue loss compared to the previous year. It is a major problem, since, as explained in the video, attracting new customers is 6-7 times more expensive than keeping existing ones.

• The company has a good dataset which is representative of the problem. It contains customers’ personal information (customer ID, education, number of family members, etc.), customer behavior (total energy consumption, % of energy used in different time slots, number of service calls made), and pricing information.

• The number of rows is high (around 34,000), data quality is good, and no privacy issues have been detected. The company identifies two actionable solutions: offering discounts or modifying queue management. The decision depends on the output of the algorithm.

CONCLUSIONS

Data-driven decision processes can dramatically improve businesses, in all industrial sectors.

There are no more excuses for not adopting them, as they are widely accessible at different levels. However, you do need to embed DDDM into your organization consciously to avoid wasting unnecessary time and resources. Always ask the right questions (what are my business goals?; what is the quality of my data ?; and how can I change my business?).

We all know that data-driven decision-making processes can bring huge benefits to companies, such as greater confidence in taking decisions, more transparency, and a higher chance of improving in the long term.

Gartner estimates that more than 33% of large organizations will have analysts practicing decision intelligence by 2023, and 75% of enterprises will shift from piloting to operationalizing AI, driving a 5X increase in streaming data and analytics infrastructures by the end of 2024.

But at the same time, businesses often have doubts on how to deliver data-driven decision-making projects successfully.

In a previous article, we talked about how important it is to choose the right project when implementing a data-driven decision-making process. We said that businesses should think about the goals they want to achieve, consider the available data, and the actionable solutions that might be implemented thanks to their data-driven knowledge.

Let’s now assume we have identified the project we want to put in production.

HOW TO STRUCTURE DATA-DRIVEN DECISION-MAKING PROJECTS

Once the project team has been created – it should be balanced, well defined, and with all the stakeholders involved – we advise you to include 3 phases:

• Project Scope (<1 week)
• Pilot (between 4 and 6 iterations, each iteration lasts for one week)
• Production deployment (between 2 and 4 weeks)

The phases can be tailored according to your needs but, in principle, we recommend you follow them as closely as possible.

Let’s imagine your data-driven decision-making project is highly complex and requires many months of work to get to the final results.

We suggest you reduce the initial scope to give you a first deliverable in 2 or 3 months.

This has two major advantages:

• You focus your efforts on something immediately usable
• You build trust in all stakeholders, who see results from the outset.

One of the advantages of Rulex is that it is extremely flexible and interactive thanks to its drag and drop functionality and eXplainable AI. As a result, it delivers value from the very first iterations.

The graph below sums up the life cycle of every Rulex project: at an early stage, little effort is involved and the return in value is already greater than the investment.

Over time, the management effort grows slowly, while the value of the solution grows exponentially.

HOW TO DEFINE YOUR PROJECT SCOPE

The scope phase is crucial to the success of any data-driven decision-making project.

If you make a mistake in this phase, for instance, setting an incorrect goal or misunderstanding your data, the project might be seriously compromised.

Typical activities during the assessment phase are:

• Goal definition: set out the goal in detail, identifying performance measures and expected KPIs
• Data definition and audit: identify the available data, evaluating data quality and suitability for analysis
• Deployment in production: plan how the new decision-flow will be deployed in production, e.g., how it will take place within the daily work routine and how it will interact with all company processes
• Assessment: evaluate whether objectives can be met or if corrective actions should be taken

If the assessment is positive, project sponsors should sign-off so you can move to the pilot phase.

DEVELOP YOUR PILOT, FOCUSING ON VALUE CREATION

In the pilot phase, the team should provide proof of concept, and clients should see the first results. With conventional methods, customers expect these steps to take several months, requiring a large number of billable hours from data scientists, and delivering little or no directly actionable insights.

With Rulex, customers will quickly receive tangible business value in the form of understandable, actionable insights.

In this phase, typical activities are data preparation, model creation and refining, and dashboard creation.

DEPLOY IN PRODUCTION FAST AND SUCCESSFULLY

In the final stage of the project, the proof of concept will be put into production, integrating the solution into the customer’s IT infrastructure.

The scenario for deploying the proof of concept strongly influences implementation activities. Among other factors, two main aspects are critical to determining the complexity of this phase:

• How different the input / output format is compared to the format used in the pilot phase
• The solution’s level of autonomy with respect to client IT systems

The good news is that with Rulex, production deployment is much easier than traditional methods, thanks to its no-code approach and numerous project monitoring features.

CONCLUSIONS

Data-driven decision-making processes are revolutionizing the world, and they can make a huge contribution to your business.

Once you have chosen the best data-driven decision-making project for your business (see our dedicated article on the subject), execution is essential.

A few reminders for you:

• Start small to achieve more ambitious goals over time
• Try to deliver value to all your stakeholders as soon as possible
• Stay focused on your final goal

 These best practices may seem obvious, but they are crucial concepts to keep in mind to avoid making the most common mistakes.