Our KTP Journey

The Problem

Back in 2012 we identified that the UK homecare market was not currently benefiting from any form of automated and optimised rostering software. This could be something to help care providers cope with the increasing pressures they are faced with.

Home care providers are constantly required to juggle client care with business. This means that often, someone misses out and the result is that either clients receive poor or inconsistent care or providers are running inefficiently, spending more than they need to on care staff and mileage etc.KTP diagram

We decided that incorporating some advanced and innovative computer algorithms into our rostering software would enable us to provide our customers with this balance. The idea was to let the software do the work and make the decisions with minimal human intervention, whilst letting providers also retain as much control as possible.

So we began to look for a third party supplier who we could join forces with to provide this ‘super rostering’ system – but failed! The home care workforce optimisation problem is extremely complex and isn’t getting easier; with life expectancy increasing, the management of this is becoming vastly comprehensive. Even optimisation specialists were not capable of providing an engine for the complex workforce scheduling and routing scenarios that we wanted to automate and optimise.

We needed to give care providers a way to quickly and easily find the balance; to automatically produce a roster which truly optimises whatever area of the business they want to. After evaluating the options available, no suitable solution could be found.

Embarking on a Knowledge Transfer Partnership

It soon became clear that if we were serious about solving this problem then we would have to start from scratch and do it ourselves! Forming a KTP (Knowledge Transfer Partnership) was the perfect mechanism to do this.

KTP’s are a nationwide programme to help innovative businesses benefit from the UK Knowledge Base by giving them access to knowledge, technology or skills at universities and other research organisations.

We had discussions with a number of universities who were interested in working with us on this project before we were approached by Dr. Dario Landa-Silva, Associate Professor in Computer Science at The University of Nottingham. Dario and the rest of the ‘Automated Scheduling, Optimisation and Planning’ research group had heard about our vision and thought there could be some common ground; they were looking for commercial, ‘real-world’ scenarios to help with their research and we were looking for highly specialised expertise to realise our idea. And so the KTP was formed to solve a problem for which we could identify no existing solution, and neither could The University of Nottingham.

The next stage was to apply for Government funding towards this project. This involved both Webroster and The University of Nottingham producing comprehensive business plans to detail the commercial and academic value of such a project. These were submitted to Innovate UK; the UK’s innovation agency which focuses on finding and driving the science and technology innovations that will grow the UK economy. Innovate saw the value of our proposition and understood the impact that this project is likely to have on both the care industry and academia and therefore decided to back us financially!

The final piece of the jigsaw was to find a KTP Associate; the person who would be working with both the University and Webroster to actually solve the problem, write the algorithms and work with the Webroster development team to produce a working tool to automatically and efficiently optimise homecare rostering.
We interviewed several applicants but didn’t feel we had found the right person to take on such a big challenge. This was worrying as without a KTP Associate the project couldn’t event begin, but we didn’t want to settle for second best. As the search took so long, our funding offer expired, so we had to then re-apply, but thankfully we were also successful the second time around.

Finally, we recruited Rodrigo Pinheiro who had just started a PhD at The University of Nottingham. The broader issue of workforce optimisation is a key interest of his research at the University and so we felt there was a perfect match.

And so it began…

Rodrigo began to work on this pioneering project whilst continuing with his PhD. His focus was solving ‘the problem’ with the Webroster.net rostering system in mind whilst exploring the scientific aspects of the problem. He split his time between our head office in Peterborough and the academic resources in Nottingham. In addition to Rodrigo and Dario, two other students in the research group were involved in the project; Wasakorn’s focus is on mathematical solvers and Haneen’s focus is on evolutionary algorithms. Together, the group of academics used Webroster’s commercial evidence and real world situations to gather knowledge and learn about the problem.

Rodrigo quickly started work on analysing the characteristics of the problem, such as constraints and how they relate. The first stage was to look at a multi-objective analysis of the problem; how improving one aspect of rostering (ie; costs or continuity of care) impacts other aspects. When this was determined, he was able to look at a solving mechanism; specialised algorithms to solve the problem efficiently. After that, he considered software development methodologies for research and development projects tackling optimisation.

March – September 2014: Solver Prototype

The first stages of work involved developing the solving algorithms and to implement a prototype for the tool. This was heavily research based and so it was conducted with thorough supervision from the academic team. We started by identifying the underlying optimisation problem and its features. We thoroughly investigated literature until we decided which direction to take and later we implemented the solver prototype. The initial results were encouraging, surpassing our expectations and showing that the choice of algorithms was correct and consistent with what we found in the literature. We managed to deliver the solver much faster than we had originally planned which was an encouraging start!

November 2014: Customer Feedback

We presented the optimisation project to current and prospective customers at events and meetings, this gave us invaluable feedback; Rodrigo was able to hear first-hand what home care providers want and need from this project. He learned about their preferences and gathered knowledge to be used in the research. During this time we also set up partnership agreements with selected Webroster.net customers who were interested in working with us to pilot the finished tool.

September 2014 – May 2015: Optimisation Server

The finished tool would need its own hosting environment and so we needed to develop an optimisation server. We defined an architecture, communication protocols and security. With the full definition in hands, we worked out the formal UML documentation of the server and later we implemented and tested it both on a contained environment and in Webroster.net’s own hosting environment, AWS.

June 2015 – June 2016: User Interface

To finalise the integration of the optimisation servers into the Webroster.net product, we identified the need for a tailored user interface, so Rodrigo went on two specialised courses to help with this task. First off, we analysed our typical system user and created some example user profiles to help us design a wireframe of the interface. Rodrigo worked closely with the Webroster.net development team and other departments, taking their feedback on board and polishing the design until we were satisfied.

Research Papers

During the project eight research papers were produced as a result of the work, which have been published in leading academic journals and international conferences. The papers cover a range of research outcomes facilitated by this KTP project: Problem benchmark data sets from real-world scenarios (based on customer feedback) provided by Webroster; formal optimisation models for these benchmark problems; various optimisation algorithms including mathematical programming and tailored heuristics to produce high-quality solutions and application programming interface (API) to facilitate the implementation of the optimisation engine.

Success!

In April 2016, the KTP project came to a close and was awarded ‘Outstanding’ – the highest possible grade – by the KTP Grading Panel for meeting and exceeding its original objectives which were to improve homecare workforce utilisation by developing an adaptable software optimisation engine that solves any workforce management scenario that includes both rostering and routing.

The results of the KTP are beyond expectation. Webroster has gained significant knowledge, understanding and appreciation of optimisation problems and has been able to produce a unique solution to the problem. Added to this, the research profiles of the academic team have been strengthened in several aspects.

Having the three-way relationship was crucial to the success of the project; as a group, we achieved results that otherwise wouldn’t be achievable. We extensively used the University access to academic journals and material on our research. Without this, we would not be able to obtain the algorithms we developed.

For more information or to get involved in this project with us, click here.