It is more a craft
than a science

A consistent focus on waste reduction, quality improvements, time-to-market, customer satisfaction, and yield increase. Persistence is key. Technology too.

Itility focuses on using data science to make your business more effective. It is tricky to expect a lot from just combining the data scientist with the domain expert and wait for results. No magic will happen using that approach.

Although data science techniques are perceived as promising, applying these often shows to be difficult. If you zoom in, it is less a magic tool, and more an engineering practice which has been known for a long time. We do not believe in ‘just start and see what happens’. We believe in a systematic approach, where you shape your thoughts into models, rethink them, test them out in a lab, refine them, and then work your way through industrializing to actually get the results.

A consistent and aligned set of best practices are to be applied when turning data into value. For this, Itility designed a comprehensive framework to assist you in your digital journey: Itility Applied Analytics.

Your benefits

Itility Applied Analytics takes the magic out of data science and treats it as a systematic engineering approach.

Close the gap between business owner and the data

Our framework offers the structure to first draw out the problem to solve.
An early glimpse of the algorithmic models that could be useful and the data requirements such as volume, variance, and quality. This enables the business owner to assess feasibility in a matter of days.

Enable the domain expert

Facilitate the domain expert, looking for tangible results. Express low hanging fruit in data and data models. We ease collaboration between domain expert and data scientist by reducing line-based labeling. Then we can steer toward classification in groups, for which a domain expert can validate in a more general way if the group is good or bad.

Predict realistic timelines

Plan for the whole and see it through. Magic can happen overnight, a craft needs time. Plan the project from the start on, to have a realistic expectation of what is required — accurately estimate the required effort for data gathering, algorithmic modeling, and implementation.

Organized for speed

The framework addresses the required skill set for manning the team. Experience shows this to be critical for fast success and ongoing results.

Itility Applied Analytics is implemented at:

How does it work?

3 steps to craft toward true data science value

The shaping starts with 3

In our Itility Applied Analytics framework we start with three roles: a digital consultant, data scientist, and domain expert.

The digital consultant is the translator between the domain expert and the data scientist. He acts as product owner and thinks about how things could be approached differently. The digital consultant understands the process and the added value that should be delivered in the end.
In parallel he will define the minimal viable product (MVP) and knows when to involve others: the software engineer, the process expert, or the data engineer.

In the lab — data deep dive and modeling

Data science starts with diving into the data. But not just any data. A skillful data craftsman thinks about requirements for the data to fit the prediction to be made.

To be predictive, the data set should contain the features and correlations required to predict the event one would consider as an event worth predicting.

To be able to recognize patterns, the data volume and variety should be sufficient to allow for learning and thus proper recognition output.

In parallel several data models are suggested to work with the data. The model parameters are tweaked and tuned. Preprocessing steps are becoming clearer. Train and test sets are defined.

Eventually a model is selected based on the available data. Data quality, volume and variety are perceived as sufficient. The model is evaluated again. No magic; it shows to be hard work.

From lab to production — rethinking the system

The data scientist leaves his lab, armed with a model performing on the data set he got earlier. Knowing his proof of value is adequate to allow for some tests in the real world.

The domain expert shows up, to evaluate the model and its results. But can this be done at sufficient precision in real life? A decision is needed to put this in practice, to run this in pilot, and validate the results. Tough for the domain expert as evidence is needed in practice, a catch- 22.

How can the domain expert prove that the system is good enough? What should be done if the model loses context and safety is at stake? Thoughts about proper fail-safes pop up. Fail-safe scenarios are defined, evaluated, and implemented. Change of processes and systems, the impact is substantial.

In parallel, the data scientist is hungry for new data to further train the created model. Where lies the trade-off? Data versus certainty versus effectiveness: no easy decision making.

Only a DevOps model can support this. Multiple people will be involved; software developer, process expert, data scientist and other roles are to assist in driving the system or process to a new data-driven based setup. True redesign.

Our framework for
crafting data science

Itility Applied Analytics offers a framework to foster a data science mindset on your business processes or products.

The framework describes how to launch first initiatives via a methodical craft that will work for all your use cases. It not only covers the actual modeling, but drives toward embedding the craft deep into your running business, resulting in tangible data-driven competitive advantages.

Itility Applied Analytics is developed as a result of numerous use cases that we have dealt with in practice. It covers methods to turn your organization into a data-centric one, and identifies all pre-conditions in the area of processes, technology platforms, and people skills.

Use case

During the use case, the digital consultant, domain expert(s) and data scientist use our Proof of Value (PoV) approach to identify possible use cases and their (data) feasibility. They do this in order to deliver the most viable cases, enriched with the first set of data and implementation-requirements.

Use case

Gate 0

Gate 1

Gate 2

Usable value

SHAPE

DELIVER

RUN
Gate 0: experimental setup in place

The data scientist then models this use case by defining the requirements to the data (variance, quality, volume), feature sets, and required data science models. He works through the steps to be taken to train the model, setting the hyper parameters, and to visualize the results. The first data science modeling has been successful. Lots of requirements on scaling the solution to production.

Use case

Gate 0

Gate 1

Gate 2

Usable value

SHAPE

DELIVER

RUN
Gate 1: final model ready

The model works, data is flowing, performance of the model is based on the requirements, and the model is ready to be run in production. The domain expert has fitted the model into his way of working.

Use case

Gate 0

Gate 1

Gate 2

Usable value

SHAPE

DELIVER

RUN
Gate 2: the model runs on a regular basis

The model runs in production, the business has adopted its usage and trusts the outcomes. Still, the business teams know it is an aid, continuous monitoring and learning is in place. Weaknesses can be anticipated, fail-safes are implemented, and falling back to regular business rules is an option when required in operations.

Use case

Gate 0

Gate 1

Gate 2

Usable value

SHAPE

DELIVER

RUN
Usable value

Diagnostic, descriptive, prescriptive, or fully automated control based on data. Any area, any process, any system made better by AI; from a waste reduction perspective, effectiveness perspective, or quality perspective.

Use case

Gate 0

Gate 1

Gate 2

Usable value

SHAPE

DELIVER

RUN