Proof-of-concept (PoC) and testing are the key stages when proving the use case of artificial intelligence to be viable and value-creating in business. Even after a successful proof of concept, the transition of the solution to production is not self-evident, but requires understanding and preparation for this transitional phase.
In AI-TIE artificial intelligence accelerator Finland’s artificial intelligence accelerator FAIA, Haaga-Helia and Laurea cooperate in order to support the small and medium enterprises (SMEs) to recognize the possibilities the artificial intelligence has to offer. When the key use case of artificial intelligence from the business point of view is found, enterprises are encouraged into the planning and implementation of the proof-of-concept. At the end of the proof-of-concept the goal is to get go/no go decree to the use case of artificial intelligence. The stage of concept certificate doesn’t have to be wide in order to be thorough or to be able to prove suitability of the use case in particular business. Also, in proof-of-concept developing doesn’t have to be linear; the projects of artificial intelligence are often iterative, which means there is shifting between different stages, including the sprint planning, testing and exporting into manufacturing (FAIA – Finland’s Artificial Intelligence Accelerator 17 March 2020, 28). Often the development follows the CRISP-DM model, which is widely in use in IT projects and works also when utilizing artificial intelligence.
In the stage of proof-of-concept artificial intelligence’s suitability for use is being tested by mirroring it in business realities and operating environments. The idea is being worked on and one step is taken forward, for example technical statement is carried out, the strategic planning related to the use case gets to be implemented and the customer’s interest is validated. With proof-of-concept the risks are minimized, and the viability of investment is ensured, market interest is verified and a better understanding of the functionality of the artificial intelligence solution is obtained. This creates a base for artificial intelligence to be taken into manufacturing.
How to take artificial intelligence solution successfully from PoC stage to production?
In the stage of proof-of-concept, in order to make the succeed artificial intelligence solution to manufacturing, it takes patience, preparation and resources. It also involves research, a competent team and programming and specialist resources. When transferring to manufac-turing, the core challenges are unresolved data questions, weak ML-tools, insufficient competence in team and inadequate management (Harshil 2021).
Because of these challenges a good idea which has shown potential in the stage of proof-of-concept may not materialize. Next to the technical challenges, the human factors come first.
Rolan, which is taking part in AI-TIE artificial intelligence accelerator, is a forerunner in car-sharing in Finland and is currently on its way to the launching of Rolan Digital Platform. The solution makes optimizing transportation possible by collecting data among electric scooters and city bikes that are in use in cities. The potential of the Platform is huge, and it might change the car-sharing markets. The artificial intelligence solution could offer a partial solution to this wide and comprehensive reform of an information system, and the possibility of this matter is being investigated in critical transition phase from the experimentation to production. The need and interest are clear: business is hectic and there’s a need for ability to look further. If predictability now remains in every employee’s own head and in an expe-rience based on silent knowledge, artificial intelligence may help with predictability. From the clean industry’s point of view the important scope of application for artificial intelligence is for example, that in the future carbon emissions can be anticipated.
In the phase of PoC the need and role of artificial intelligence are specified. It may also be the case that in the final product launched on the market, it ends up using some other solu-tion, such as software robotics without an actual artificial intelligence component. This is usually valid in many companies, when it’s noticed that several problems and challenges in business are solved without artificial intelligence.
Rolan made a pilot of Digital Platform and based on that the company was active on planning and implementing to transfer it into the manufacturing. A technical partner for the programming was chosen and the evolution of the system was taken far. While transferring to manufacturing environment, the company has faced many challenges, such as developing user interface, scalability and making a terminal independent version. At the top of the pain point there are not technical challenges but how to get the authors of the company behind the idea and implementation and work together towards a common goal. At Rolan’s case the PoC phase has shown that the key to success is to get people to use the system, because in practice, if the master data is not maintained, the solution and the final product or service do not have possibilities to succeed. Master data does not update itself but requires people’s investment and work, which in turn affects straight to the quality of data and the functionality of the solution. Therefore, as a critical success factor, moving from PoC to production is making experience and competence possible inside the team and induction that has to be done extremely well.
Regular dialogue within the team and with people supports Rolan’s success in bringing the entire complex to the production stage. Regular communication is essential; otherwise, assignments and things agreed are left undone. The project also needs a product owner, who supports others and takes things further. However, at the same time a central factor of success is making the dialogue and the regularity of it possible. At the context of telecommuting and remote teams this need will increase, since people might read things and the expec-tations and needs in different ways – and it is important to be on the same page.
At the stage of Rolan Digital Platform’s proof-of-concept the manual work was done, and the old systems were used. At that point there was no need to learn to make things in a new way, until when the solution went into production, it required a new system and its effective implementation. This demands a need to rationalize and clarify the need for reformation for the employees involved in the process and their role in it. It has got to be proven that concurrent use of the new and the old systems is temporarily inevitable, even though this kind of mode of operation is more laborious in the transition phase. The end result is a unified task management application for employees to use. A broader meaning and investing in its opening is essential for employees to understand the critical value of their work: when you are involved in the development process during the transition, you are therefore involved in the development of the company.
In a small company everyone has their role in development
A successful artificial intelligence project is primarily the people implementing it – their competence and co-operation skills. Rolan is a demonstrating example of this. An important, separating factor between SMEs and large companies and larger organizations is that in smaller companies the target of resources need to be optimized. Even in an SME, the successful implementation of artificial intelligence solutions requires a good product owner who is motivated and takes responsibility of development. The product owner is responsible for ensuring that the development work progresses and that everyone knows their role. The development work requires sufficient time, at least several months, for the people to have time to commit to the new thing. Learning new things, mutual dialogue, brainstorming and joint development do not happen in an instant.
When taking the experimentation to manufacturing, every employee in SME is inevitably involved and everyone has their personal key role in developing and succeeding together. In addition to the product owner, the support of the management in particular is necessary so that the implementers of the artificial intelligence project have sufficient resources and the results can be used. If management support is lacking, it can easily happen that even a successful PoC is not put into production and the project is quietly discarded.
The preparedness of the team and people can be fostered and supported. In order to take artificial intelligence in use in enterprises, it is necessary to have strong and diverse compe-tence, technical skills, expertise of artificial intelligence and upmost a vision about how to create value with artificial intelligence solution for the business objectives. In case the nec-essary competence cannot be found inside of the enterprise – which is common especially in SMEs – it has to be purchased from the outside. Therefore, consults and enterprises specialized in artificial intelligence are demanded. The introduction of AI solutions and the move towards the production stage can change or even disrupt familiar processes and the work of many stakeholders, and the right buy-in and acceptance is needed to involve users in the production of carefully built ML models and technology solutions (Leung 28 December 2021).
Proof-of-concept may have given green light for the new solution, but if the company and its people do not understand the potential of it, the solution probably will not make it to production or will face many obstacles. Interaction between people is in the very core of an effective and successful artificial intelligence solution, in which case understanding and developing cooperation models happen by the side of developing the technical solution. As an example of it there is Tekoäly tulee – Tuki, osaaminen ja yhteistyö kuntoon!, a project that investigates how artificial intelligence changes work and develops the new role of knowledge work in the transformation of work. The introduction of artificial intelligence-based solutions in companies, require proceeding of cooperation by the side of multiplexing and strong digital information work: when planning the cooperation of a machine and a human it requires new ways of approaching and operating (Kärnä etc. 2021). This is particularly important in small teams and SMEs.
When planning the company’s artificial intelligence project, PoC and taking the results into production, it is good to keep the following matters in mind:
- Artificial intelligence solutions support taking the leap to clean industry in a meaningful way. Taking the use case of artificial intelligence to the stages of designing, concept certificate and manufacturing is iterative process.
- Enough time and resources are required and need to be allocated to the transition from proof-of-concept to production phase, while making sure that the project has the support of the company’s management and team. A successful AI project is primarily the people implementing it; their skills must be supported, incl. technical skills, artificial intelligence expertise and insight into how an artificial intelligence solution creates value for business goals.
- Especially in the case of artificial intelligence, not all know-how can be found within the organization, in which case it is necessary to consult experts in the field, for example in the form of purchasing services.
- The introduction of new technologies and artificial intelligence solutions must be phased in. Changes are implemented more smoothly when the different phases and the related needs have been identified and clearly communicated to key stakeholders.
- Regular interaction enables implementing the tasks, strengthening the shared vision and intervening to the problematic points on early stage. The fluency of team’s inner cooperation needs to be invested in and team members’ cooperation skills developed.
Translated from the Finnish by Sari Benford and Petteri Saloranta.
FAIA – Finland’s Artificial Intelligence Accelerator 17 March 2020. AI Playbook. Liikkeelle tekoälyn hyödyntämisessä.
Harshil, P. 2021. Proof of Concept to Production. Neptune.ai. Updated 13 December 2021.
Kärnä, E., Nikina-Ruohonen, A. & Humala, I. 2021. Entrepreneurial spirit of knowledge workers as a key asset in strategic change. HHBIC 2020, 17–18.11.2020, Online.
Leung, K. 28 December 2021. Bridging AI’s Proof-of-Concept to Production Gap. Medium.
Shearer, C. 2000. The CRISP-DM Model: The New Blueprint for Data Mining. Journal of Data Warehousing, 5, 4, pp. 13–22.