Modelling: Creating the right structure and flow
Modelling is an integral part of analysis and provides actual results.
Stakeholders in this phase:
Data scientists, development team.
Obstacles:
Building complicated models that don’t satisfy business needs (accuracy, legality, speed) or can’t be transferred into an integrated software solution (communication with the IT team about feasibility).
Dangers:
Situations where you have a nice model but it is not applicable or has no real value.
Engagement with stakeholders:
Data scientists must be aware that they need to share their thoughts with business and IT teams.
Evaluation: Present and visualise results, and assess benefits
What works best in this stage is great data visualisation and simple, clear results. You need to know how to present the results and communicate honestly no matter whether your goal was achieved. ERNI has developed evaluation frameworks that can be used in this stage.
Stakeholders in this phase:
Data scientists, visualisers/UI experts, business analysts, business end users.
Obstacles:
(a) The communicated outcome is too complicated to understand; (b) the result might not be satisfactory and data scientists may try to hide it.
What can go wrong:
(a) Results can’t be communicated appropriately, and the business does not see the benefit or success of the project and stops it; (b) the reverse might also happen: the failure to properly communicate that the experiment has produced negative results with respect to the business goals and that another iteration is needed can lead to the implementation of an unsuccessful application.
Difficult but correct decisions:
Communicate bad results and openly say when it is not worth it to investigate further.
Engagement with stakeholders:
Engage early, show results regularly and visualise in an appealing and simple way.
During the evaluation stage:
1. Present results in a clear, targeted and simple way.
2. Communicate failure.
3. Promote the benefits.
In this stage, your team makes a decision to operationalise a data product (or moves from an offline analysis to the pilot, or from the pilot to a large-scale solution), or even makes the hard decision to put an end to the project.
Make it work in real life
What makes a successful implementation? Good old project management, system integration, process integration and a wisely chosen interdisciplinary team with broad technology know-how.
Stakeholders in this phase:
Data scientists, visualisers, UI experts, development team, data engineers.
Obstacles:
(a) Poor integration of the data component into the existing software system; (b) a good model but a resulting integration that is not user friendly; (c) poor visualisation can make even a good model and its results hard to understand; (d) you get stuck during the introduction and training of the new tool.
What can go wrong:
A project that, despite a successful start and promising results, must be stopped because of operational failures, or when the outcome is not accepted by the end users.
Engagement with stakeholders:
Show early pilots of the final solution to business stakeholders. Build one team out of data scientists and software developers.