Schedule AI Assistant
Product Context
The Unified Supply Chain portfolio supports users within the crude oil supply chain, from purchasing crudes, through to planning, scheduling and executing the refinery operations. There are multiple user types that span this supply chain, each with their own objectives and needs, each communicating with the users up and down the supply chain.
​
While working at AVEVA my main focus has been within the Scheduling part of the portfolio and further still, within a new tool called Schedule AI Assistant.
The Scheduling Problem
Schedulers are responsible for ensuring the continual, safe and economic running of the refinery. The refinery must have a constant feed of material to all of the process units at the same time as outputting the right quality of products, whilst factoring in the varying qualities of crude oil coming in and the different processing constraints on each and every process unit. It is the responsibility of the Scheduler to create day by day, hour by hour schedules of the weeks ahead, which are then communicated and executed by the operations team.
​
It is an exceptionally complex and time dependent job.
Problems with Solutions To Date
Because of the huge complexity of refineries, all the processing units within them and the materials which they process, computing power has been limited in its ability to assist Schedulers. Solutions to date rely heavily on users to create schedules for the refinery manually with computers being able to check these schedules for their feasibility. Only some, if any, automation exists.
Schedule AI Assistant - The New Approach
With cloud compute offering more power to express the complexity of any given refinery, the opportunity to utilise automation and AI techniques to assist Schedulers is apparent. Instead of Schedulers manually creating schedules, Schedule AI Assistant will rapidly generate feasible schedules that the Scheduler can select and refine if required.
My Process Overview
My role and work
• Sole designer, spanning Discovery through to Delivery
• Cross collaboration with mathematicians, front-end developers and back-end engineers
• Varied work from usability improvements through to supporting vision and roadmaps
• Working in a dual track, Discovery & Delivery Kanban process
• Interviews and testing with expert users, customer support
• Generation of workflows and designs to support them
• Utilisation and contribution to AVEVA Design System
• Cross-collaboration with engineers for input and review of design approaches
Domain understanding
Scheduling AI sits within a complex product family, within a complex industry, with domain specific vocabulary. Before tackling any design work, I made it my priority to familiarise myself with the tool set and the industry they sat within. Multiple discussions with subject matter experts over the course of two months immersed me in the domain. As my understanding grew I created diagrams that could be reviewed and iterated to ensure this understanding was correct. Key activities included:
• Reading subject matter books and literature
• Research into competitor tools, functionality and user feedback
• Stepping through user manuals in the tool
• Discussions with expert users, product managers and customer support
• Diagrams summarising my understanding, reviewed with users and PMs
• Mapping refinery constraints
.jpg)
.jpg)
.jpg)
Research & Initial Concepts
Project focus: Supporting Schedulers during reconciliation
As scheduling is an ever evolving, day-by-day process, an important aspect of a Schedulers role is to bring in current data from their site to ensure their schedule is in line with reality. This process, called reconciliation, often means that tweaks, rather than major changes, are required.
​
This piece of work aimed to examine the reconciliation process and how Scheduling Assistant would best support schedulers undertaking it.
The key activities were:
• Interviews with users to understand the reconciliation process
• Observation of a reconciliation process in the desktop tool
• Creation of workflow diagrams to communicate to the development team
• Analysis of research and generation of approaches
.jpg)
.jpg)
.jpg)
.jpg)
Initial Concepts & Prototypes
Project focus: Supporting Schedulers during reconciliation
Once an understanding of the needs and workflows of the scheduler during reconciliation had been gained, work could focus on how the AI could support these needs and how the scheduler would communicate their intent to the system via the UI.
​
I created a framework of Fix, Flex, Forget to denote the degree of freedom the AI would have in respecting a marked activity.
​
The key activities were:
• Working with the AI team to understand the technical implications of the work and collaboration of possible solutions
• Prototyping in Adobe XD to communicate and test design concepts
• Testing of prototypes with users
• Review with the full-stack team
• Iteration of designs
• Utilisation of AVEVA Design System
Final Solution
Schedule AI Assistant is a web-based application that can be opened from the existing desktop tool, allowing schedulers to adopt the tool in a way that does not disrupt their existing workflows. When opened, the landing page allows the scheduler to specify preferences for their schedule as well as being able to fix certain activities that the Assistant must not overwrite. When the scheduler is satisfied with their settings they can press optimise, initiating the AI engine.
On completion the AI returns a proposed schedule that the users can review prior to applying back to the desktop tool, where they can export the schedule to operations.
Further design strategy has been conducted, outlining additions to the product that include; visualising multiple results alongside their KPIs, exposing global settings that currently reside in the database to the UI to make initial setup simpler and further visualisations to help with assessment of the AI proposals
Summary
Since joining AVEVA, Scheduling Assistant has been my main focus. Embedded within the team, my work has been many and varied; from supporting detailed UI additions and modifications to working alongside the Product and Development managers in defining short, medium and long term visions for the product. All this work has been undertaken in an environment with domain experts on hand and validation of designs and ideas conducted.
​
I have been proud to represent designs and make space for design within a team that has previously had no design involvement, modifying processes to factor in collation of user needs, idea validation and development collaboration and review. I have particularly enjoyed developing a close working relationship with the mathematical optimisation engineers, which has allowed me to act as an advocate for the users in the underlying behaviour of the product.