Whitepaper – AI Maturity
Learn how an Analytics Translator helps organizations overcome the most common difficulties when building AI solutions.
Whitepaper – The Analytics Translator
The Analytics Translator is the liaison between senior management, the business, and data experts. On some days, you are the gatekeeper of the project funnel, brainstorm ideas with executives, and work with the data experts to groom the backlog of viable ideas.
Let the BEAT drive your AI Strategy!
What values drive your AI strategy development and execution? Many companies proclaim their values prominently on their websites and in marketing. Some companies even print their values on the office walls. These values reflect the company's cultural and ethical principles towards their customers, employees and society in general. In day to day decision making employees on any level in the company may use these values as the ultimate measure to evaluate different scenarios and drive their decisions.
At GoDataDriven our 4 values are People First, Sharing Knowledge, Customer Intimacy and Quality without Compromise. These values lead us every day in our work for our customers and help us whenever we are in doubt about the best course of action.
This blog is written around the statement that AI Strategy development and execution also requires general values and principles. Values, the BEAT as I will explain later on, that drive your decisions both in designing your AI organization and processes and in putting this organization and processes into practice.
Frequently check whether the BEAT is right and you know you are aiming for success with AI.
These values probably won’t make it to your office walls, they are a bit too practical for marketing purposes, but they will definitely help you in getting value out of your AI initiatives!
AI Strategy Development and Execution
Your AI strategy covers what benefits you're aiming for, how you allocate resources to turn your ambition into reality and how you set up the governance structure and processes to ensure this is done according to your plan and principles. In developing your AI strategy many trade-offs need to be made, as discussed in my previous blog What is your AI Strategy?. This blog mentioned some example trade-offs to be made:
- Business pull & Data/Technology push
- Day-to-day business & Business strategy
- Business opportunities & Ethical behaviour
- Data driven & Business driven
- Make & Buy
- Enabling technology & Hiring expertise
Making a clear choice on direction from these trade-offs during strategy development guides you in strategy execution; translating the longer term strategic choices into the shorter term tactical and operational roadmaps and processes.
BEAT values for a successful AI Strategy
From our experience in developing AI products and helping customers setting up and organizing their AI capabilities we have identified 4 key success factors for AI products. Keeping these factors in mind as the values driving your decisions on AI strategy development and execution guides you towards success with applying AI for the benefit of your customers and organization. The values are:
B usiness relevancy
E thical acceptability
A nalytical quality
T echnical quality
The acronym BEAT is the memory aid for this list of values. The AI strategy should lay the foundation for application of the BEAT values, paving the way and setting the right directions and requirements to ensure compliancy to the values of all efforts on AI. Below I will elaborate on each of these values.
Well, that is a no brainer isn’t it? Why even bother analyzing data, creating dashboards, developing machine learning models, trying to predict something when it is irrelevant? As trivial as it may sound, unfortunately we often see a lot of effort going into creating data driven insights for the sake of the insight alone.
We are not aiming at nice figures and charts on slides alone, you want your limited AI resources to produce tangible results and actual impact. That's why it is important to align your AI initiatives with your business strategy, since on this level, you already decided on the right focus for all company resources for the next couple of years. Next to operational continuity related objectives it is your business strategy that determines the right focus for where to look for relevant AI opportunities.
New opportunities should be evaluated on a solid business case in terms of added value, reduced cost or mitigated risk being addressed. The relation to and potential impact on operational and/or strategic KPIs should be made clear in advance. Structural benefits should be favored over one-off cases. Read our blog on AI use case generation for an example on how this process could work for you. And please note, you should consider not even to spend just a minute of brainpower to a potential use case without the commitment from the business owner, responsible for the affected KPI's, on adoption of the AI product.
This check should be done in advance and often during the full AI product lifecycle, from ideation, experimentation and development up to production. The B in the BEAT is our reminder for this.
This should not be a vague political continuum of opinions, but rather a binary metric with veto power. The AI strategy defines what is appropriate and what not, based on the code of ethics and risk appetite of the company. In practice this turns out to be still quite difficult but at least we should aim for this. Deciding on minimal requirements with respect to data privacy and security is a good start. On top of that the application of AI needs guidelines on transparency, fairness and interpretability of models.
A lot of research is being done in this area and new ideas on measuring fairness and making machine learning models interpretable are published continuously. These ideas and methods become more practical and available as open source tools at great speed. Your AI strategy for the next couple of years definitely should cover the importance of fairness in AI to your customers and company and requires continuous attention during AI product development.
The analytics behind an AI product is the combination of source data used and methods applied to this data. The quality of your AI products depend on the quality of your data and the quality of your analytical methods.
At least the meaning and quality of the data with respect to the intended purpose of the data should be known. That means, defining what good quality data looks like is the first step, actually knowing the quality is second and improving the quality of the data comes third if even considered possible. It is a well known empirical fact that gathering, exploring, interpreting and preparing data for AI projects may take up to 80% of the time of your data experts. Your Data & AI Strategy should make data availability, quality and documentation a company wide shared responsibility e.g. by prioritizing good data governance.
The other aspect of analytical quality is the methodological approach on turning data into value. Do we really understand the data? Does the algorithm fit the feature and target space? Added complexity in the analytical approach should be balanced against the added benefits. Furthermore it is important to care for reproducibility of analytics. If some time later on we (or the customer, or some authority) want to understand why the model predicts a certain value we have to be able to reproduce the model outcome from source data all the way up to the actual prediction. Simplicity in the entire pipeline makes documentation and reproducability of analytics much easier. This may very well mean that we choose the rather basic logistic regression model instead of the much fancier but not even that much more accurate deep learning model. As a last note, it is better to reuse battle-tested analytical components than reinventing every line of code every project again (which unfortunately is common practice).
Developing a prototype and proving the potential value of an AI solution in a data laboratory setting is nice, but most often the prototype is not mature enough to move forward to production. You don’t want to restrict and slow down the experimental phase too much by requiring production grade testable code, but a complete rewrite of the code base between experiments and production is error prone and very time consuming. The balance between rapid prototyping and time to market after the experimentation phase of AI products should be determined by your AI strategy, facilitating both phases with an acceptable level of drawbacks on the others.
Like for the analytical components, reusage of existing technical components should be favored over developing new ones. Maintainability and robustness of the technical solution is key; we are not developing AI products just to prove we are able to, we are developing AI products to have them add value for many years. This means that the required effort and cost of running and maintaining the product should be considered in setting the definitions of done for the development phase of the AI product. Again basic simple solutions may be favored over fancy but complex solutions for this reason. The AI strategy should guide us on when to choose what solution and the T in the BEAT makes sure we regularly check whether what we are doing still makes sense from a technological point of view.
Assign responsibilities for checking the BEAT frequently
From AI strategy development all the way to AI product development, the BEAT values are important success factors. Everyone involved should keep them in mind to continuously check whether AI efforts are heading in the right direction.
A data scientist working on the A should take a step back and zoom out every now and then to check whether the B, E and T are still on an acceptable level and mutually in balance.
This already indicates that, while everyone should feel responsible for the entire BEAT, each one has a specific responsibility in this field. We advice you to assign responsibilities on each of the four values to senior data experts, e.g. as follows:
- Business relevancy -> Analytics Translator
- Ethical acceptability -> Privacy Officer
- Analytical quality -> Lead Data Scientist
- Technical quality -> Lead Data Engineer
These seniors lead the development of best practices on their topics with the BEAT values in mind, and facilitate the team and the rest of the organization in applying them.
The BEAT values act as the guiding value framework to turn your AI ambition into reality, during both AI strategy development and AI strategy execution.
At GoDataDriven we support companies to design and execute their AI Strategy and to organize for success with AI. We build AI solutions, we train people and we manage data platforms and AI services.
Do you want to know where you are on your AI journey? Please contact us for information on the AI maturity scan.
Would you like to join us and help companies to define and execute their AI Strategy? Please have a look at our Careers page on the AI Strategy Consultant position!