- SAP Managed Tags:
We are currently experiencing extraordinary times of multiple disruptions. Artificial intelligence will fundamentally transform not only technology but also work itself. How well we succeed in this transformation depends on our approaches to adoption, adaptation, and transformation. But what exactly are we talking about here? According to Wikipedia, adoption means:
"In computing, adoption means the transfer (conversion) between an old system and a target system in an organization (or more broadly, by anyone). If a company works with an old software system, it may want to use a new system which is more efficient, has more work capacity, etc. So then a new system needs to be adopted, after which it can be used by users."
From a more holistic perspective, it goes beyond the rather 'technical' implementation and usage of software to achieve comprehensive acceptance and productivity of users regarding digital tools. As always, it actually depends on the perspectives of stakeholders which are pretty diverse:
- management wants high ROI, cost savings, growth, productivity
- IT desires low integration costs, low TCO with high security
- Departments from a Line of Business want to achieve their individual goals
- The end-users want little frustration and they want their work to become easier, faster, or better
These are all wishes or projections that need to be considered beyond the purely technological perspective, especially for applications of generative artificial intelligence. Especially here, acceptance and productive use are highly important. Because in the introduction and use of generative AI, many fears resonate - ranging from dystopias like world annihilation to uncertainty and fear of individual job and relevance loss.
Therefore, adoption should be understood here as an extended people oriented process beyond technical use or process optimization and analyzed from a 'human centered perspective' as well.
Digital Adoption as a Process
In the context of our recent Digital Adoption Blogging Challenge, I have already distinguished between adoption and adaption from frustration, confusion, curiosity to mastery and advocacy, inspired by Maslow.
In the mentioned contribution, it was described that especially with the highest levels, we not only 'affectively' expand the understanding of adoption. We also can expand the degree of maturation towards adaptation and transformation.
After all, we now want to achieve more with artificial intelligence than just to use of tools for existing tasks and processes. Ideally the use of a disrupting technology like generative AI incorporates a rethinking of processes, business models, roles, and individual tasks. Which means the adaption of work according to the new opportunities - not just adoption in terms of usage in percentages.
AI Transformation Requires Extended Digital Adoption
Based on these general reflections on adoption, adaptation, and transformation, we now want to look at AI transformation and AI adoption. Key success components beyond purely technological questions are, as usual, communication, involvement, clear goals and vision, learning opportunities, understanding of resistance, motivations, interests - in other words, components that are generally known from organizational change management. They are especially essential to address fears or resistance which are currently arising in the context of AI.
Which frameworks, approaches and models can help in AI Adoption?
In the following, I will try to reflect several established models from the field of digital adoption.
The simple model of Should, Can, May and Want
'Should': This refers to a clear AI strategy, clear goals and KPIs of how success would be precisely measured. For a clear strategy it is important to identify the best use cases that are currently technically feasible, accepted, and bring the highest added value.
'Can' means: Everyone has the respective AI skills and competencies - or can develop them depending on the respective needs.
'May' means that the corresponding authorizations and empowerment is available to use AI tools like co-pilots, agents, or chatbots.
'Want' stands for the motivation, conviction and acceptance to use AI.
Everett Rogers' Diffusion Model
This describes how innovations are adopted and spread in an organization. It identifies five types of people (innovators, early adopters, early majority, late majority, laggards) and their different readiness and attitude for technology adoption.
For optimizing AI implementation it helps to target strategies to different target-groups.
- Innovators: Risk-taking tech enthusiasts. Strategy: Early involvement, resource provision, encouragement of experimentation. Leverage as champions, pilots and multipliers.
- Early Adopters: Influential opinion leaders. Strategy: Leverage their influence, tailored training, showcase success stories.
- Early Majority: Pragmatists. Strategy: demonstrate the value, support them, foster their collaboration.
- Late Majority: Skeptics. Strategy: Address concerns, simple integration, create incentives.
- Laggards: Change-resistant. Strategy: Personalized training, emphasize necessity, gradual introduction.
Crossing the Chasm by Geoffrey Moore deals with the challenge of bringing an innovative technology from a small group of early adopters to a broad majority.
For generative AI, this means that different target groups should be treated differently. For example, early adopters and innovators can be used as pilot participants or promoters, e.g., for change agent networks or coaches. It also helps to manage expectations: not everyone adopts in the same time and communication needs to reflect the needs of the different categories.
Clayton M. Christensen's Disruptive Innovation shows how new technologies or products replace old ones by being simpler and more cost-effective.
Generative AI is clearly a disruptive general purpose technology. Companies need to embrace it for the highest value use cases. However every individual should actively experiment to learn from own experience where AI can help them e.g. to improve or even replace existing tasks and processes. Psychological safety for this experimentation as well as incentives and benefits are key to reinforce it.
Technology Acceptance Model (TAM)
The TAM explains how users adopt and use a new technology based on its perceived usefulness and ease of use.
TAM is probably the most empirically studied model and was extended several times - additional influencing variables have been added, such as social influence including subjective norm and voluntariness and cognitive processes like social influence.
More recent studies increasingly highlight users' trust in the technology as promoting acceptance - especially with AI, this is a central point. It underscores the need for being transparent about the ethical use of AI Tools and what happens with the data provided as input.
ADKAR
This Model is another well-known framework for change management. It shows the dimensions which need to be addressed to successfully go through changes.
It involves Awareness (creating awareness), Desire (creating the desire for change), Knowledge (imparting knowledge), Ability (developing skills), and Reinforcement.
It's important not to forget any of these elements when introducing AI and shows that awareness is not the same as knowledge or ability, and that consolidation through reinforcement and continuous development is important.
Agile Methods
Using agile approaches is particularly suitable in the IT sector, as agile models were invented in software development in order to cope better with increasing complexity and dynamics.
Scrum and Kanban, for example, are proven agile approaches that can be used in change projects. Scrum enables teams to work in short iterations (sprints), allowing for regular feedback and quick adjustments to the need for change. Kanban, on the other hand, visualizes the flow of work and helps to identify bottlenecks, which increases efficiency and promotes continuous improvement. Both methods support iterative approaches that involve continuous learning and adaptation in order to increase the acceptance of change and better involve employees. In addition, methods such as retrospectives and different approaches from design thinking can also be very helpful in change management.
Working iteratively with agile principles and practices or using single methods like using retrospectives and similar approaches can be very helpful when fostering AI Adoption.
Kotter's Model for Leading Change
This well known approach consists of eight steps for creating successful changes. It can promote the acceptance of generative AI by creating and communicating a strategic vision and narrative for the use of AI and its benefits, forging coalitions, removing obstacles, celebrating successes and achieving quick wins. The model is a bit old and although it has been further developed it does not reflect the latest state of experience regarding change management. Same with the next one.
The Change Curve
This describes the emotional phases people experience during changes. It can help in implementing generative AI by considering the different emotional reactions of employees to adapt communication and support accordingly. The phases vary and include anticipation, shock, denial, despair in the valley of tears, learning and adaptation, acceptance, and integration.
Those affected go through these phases at different speeds or completely differently depending on the context. It shows that emotions like fear, excitement, envy always play a role in changes and need to be checked and addressed depending on the stakeholder. It can be criticized that it is not evidence based and originally comes from the are of handling grief which is obviously a pretty different context. And no-one moves though those phases in organizations. However it is helpful as it reflects emotions which are often neglected.
SAP Activate
This is a methodology for implementing SAP solutions, where AI is increasingly embedded. It provides many templates and tools, so called accelerators to manage SAP projects. It includes, among others, the 'Solution Adoption' workstream, which addresses the needs of change management and enablement as a cross-topic and further education. See here the overview-page for Organizational Change Management.
There is no best model
With all frameworks and models, it is of course important to respond empathetically and intuitively to the interests, emotions, and culture inside a company - depending on what makes sense. The context and time in which above approaches originated also help for classification and reflection.
The frameworks are like toolboxes - it's good if you know and understand them. Next to the TAM model no models have been empirically validated. In the recent year agile approaches have been getting more use by practitioners.
Based on experience, research and a recent blogging challenge on digital adoption I would see 4 fields of actions to tackle AI Adoption.
Four Components of Successful Human-Centered AI Adoption
In the following figure, you will find four essential components of human-centered AI adoption beyond purely technical focus, with some important key points including links for the SAP context. Even if a certain chronology is likely here, these 4 components stand equally 'side by side'.
Human-Centered Design Mindset & Practices
In the recent years human-centered design has been proven to design and implement new solutions and are an essential basis for adoption, adaptation & transformation success. There are different ways to approach this. For crafting an AI Strategy, it is key to identify and prioritize amongst the many use cases which are available. There the proven design thinking criteria desirability (what do people desire?), feasibility (is it technically possible?) and viability (can the company profit from the solution?) can help.
For the selection of suitable AI scenarios, the SAP Apphaus Toolkit for generative AI - the Business AI Exploration Workshop and the Business AI Design Workshop are very helpful toolkits. More Business AI focused Methods are in development.
Whether design thinking, co-creation, iterative prototyping, agile coaching, or other practices from the human centered design area - it's always important to put the user at the center and think from their perspective. Solving real-world problems with real user data lays a foundation for user adoption helps also in the design and implementation of generative AI.
Organizational Change Management
Beyond the individual perspective, AI adoption always mean the use of change management. In the digital Adoption blogging challenge several several points were highlighted.
- Clarity, transparency and value-add: Advantages of the new technology must be clearly communicated for organizations, teams as well as for single roles or personas. Employees should understand how the technology makes work more efficient and better.
- Trust and ease of use support acceptance: Human-centered design from the user's perspective, easy access, and user-friendliness of the GenAI applications are crucial. Trust in the tools is becoming increasingly important. Especially with AI, trust-promotion such as transparency of models or checking for possible bias are important.
SAP offers free tools and learning opportunities, communities for experience sharing, as well as Organizational Change Management Services such as the data-driven Deep Transformation Insights Service. This involves a combination of operational insights and feedback data from regular surveys. These are based on 5 change dimensions from a users point of view: Awareness, Acceptance, Enablement, Empowerment, Commitment.
The technical implementation is carried out using tools such as Signavio for process analysis and modeling and Qualtrics for the surveys.
Learning and Enablement - also called AI Upskilling
The most important AI investments currently are retraining or upskilling, next to automate tasks and processes to achieve higher productivity and cost efficiency of course.
In addition to general understanding and knowledge of various tools, the competence of prompt engineering is important. Some name this AI literacy. Moreover, it can be helpful to learn advanced prompt engineering techniques such as chaining or multi-prompting. Human language replaces programming language and with generative AI low- or no code development will become reality.
Besides learning basic knowledge, it's about the use of generative AI in the respective daily work and job. E-learning offerings are suitable for the basics. Some content is even available for free, for example via the Hasso Plattner Institute or the AI Campus. Special in-house academies are often set up for important target groups such as software development, R&D, customer service, or IT.
Since the goal of upskilling is always competence like implementing or using AI Tools, AI training should go beyond basics. They need to include practice, trying out and experimentation, reflecting, and the exchange with others.
Possible methods include workshops or hackathons where job-relevant challenges are solved. Every job has tasks where generative AI can help, like in faster or better content generation incl. summarization or translation, data processing and analysis, software development and more.
Another possibility is to learn in peer groups, like in the own team. The LernOS framework e.g. includes an AI learning guide. This guide structures a self-directed learning journey of small groups with input and exercises around AI. This allows learning at the workplace, including analyzing and working on individual use cases and can be realized easily and relatively cost-neutrally.
With the Corporate Learning Community we e.g. designed and run a hybrid Promtathon. Points to consider for such hands-on session are
- Use of central communication tools, e.g. a digital whiteboard
- Analyze and use Challenges from the daily job-life
- Allow self-organization & diversity of the learning groups
- Give minimal input and focus on maximum practice and output
- Various formats are possible from digital, hybrid to on-site.
- Duration can last from 2 hours, half a day to one day
- Documentation and sharing of results e.g. prompts is key
Example: Use of AI in SAP Product Training
Another field of application is training in SAP products and technologies. Every year, millions of SAP customers, employees, or partners learn new SAP skills. Where do we use AI?
AI Services can be used for different targetgroups
- AI services for authors: These can be text-based, such as for creating quizzes & assessment questions, scripting, or translation. Other possibilities are media services around the creation of audio, videos, and image material. For example, summaries were created via ChatGPT from SAPs internal AI Playground and the video with text-to-video via Synthesia. (see here an example course with text summaries and video avatars).
- AI services for learners: This involves Knowledge Discovery, which means AI-optimized searching and finding, personalized recommendations, up to digital learning advisors.
Further topics such as the further transformation of learning services are in discussion. Initial pilots show that there is not a single tool, but a hybrid, constantly expanding landscape. Especially when it comes to new knowledge like in product training, which no large language model can know yet, the strength of AI is supporting and augmenting experts in learning material production. Structured metadata and grounding techniques can help to reduce hallucinations.
Another example is the use of AI for SAP end users: Learning content and documentation for SAP products can be automatically translated per click in the SAP Enable Now tool and either accessed in the software itself via a so-called companion or via central knowledge databases.
Via WalkMe, the digital adoption platform recently has acquired, users get AI based recommendations on next best possible actions in the software in a context-aware copilot. Users also can access on-demand AI chat capabilities for conversational search and task automation. So we see that the boundaries between learning and software use are increasingly blurring - AI becomes performance support increasing user productivity.
SAP AI products and technology as well as related topics can be learned via SAP Learning Hub or SAP training. Currently, there are already 135 free learning contents on "Artificial Intelligence" on SAPs digital free learning platform.
Technical Support & Consulting
In addition to technical 'low' and 'high touch' support and consulting during the implementation, ongoing consulting plays a key role, even beyond the classic project lifecycle.
Since AI is embedded in SAP products from HCM to SCM or BTP, support in the cloud is offered as a support subscription with SAP Preferred Success or in the private cloud with SAP Cloud Application Services.
Via this Preferred Success Service customers can e.g. use the Discover SAP Business AI service – which guides customers in their exploration of use cases & prerequisites, prioritization of those use cases and initial project planning. It is all remote and 10h effort in total. Discovery workshops make sense for everyone in the beginning and of course drive adoption as they help to focus how and where AI is implemented.
Another practical example of the technical support is the SAP Joule Activation for SAP SuccessFactors" Adoption Lab. Here different customers meet online after some prework and prerequisite checks. Then they are guided by an expert how to activate SAP Joule – which is SAPs digital assistant leveraging GenAI.
For technical support there are of course also tools which support Adoption (next to the above-mentioned digital Adoption Platforms).
These are e.g. tools for process management, -analytics and -mining, for collaboration and communication like enterprise social networks with wikis, forums etc. as well as tools for surveys to measure the qualitative adoption.
Reflection and summary
As a fan of Ethan Mollick and his works I would first like to reflect some of his thoughts which guide the effective use of generative AI. His 4 rules for co-intelligence are a good advice for everyone:
- “Always invite AI to the table” – use AI wherever possible. Think of it as your digital working student or assistant - always try to delegate tasks to the AI.
- “Be the ‘human in the loop’” – we all know that AI is not 100% correct and does hallucinate. So always use your empathy, domain and expert knowledge or even common sense to validate AI outputs.
- “Treat AI like a person, but tell it what kind of person it is” – give the AI like a LLM as much context as possible, like persona, goal, steps, example data, tone or output-format. It is like with all briefings: the more exact it is, the better the output.
- “Assume this is the worst AI you will ever use” – The applications get better and better - be open and try out new tools, models and approaches.
This contribution can only touch on the topic of AI adoption. It will hopefully motivate discussion on why and how we can ensure the success of this important change in our time beyond purely technical implementation.
For the Implementation of generative AI I think it is very important that everyone experiments actively and tries to use AI wherever possible in their tasks. Like with driving a car you can read something about – but for really mastering it you need to experience and practice it.
So organizations need to offer the toolsets, but also space for learning. Next to the toolset it is also helpful to reward people for revealing AI use. If productivity gains happen, workers need to benefit as well. Psychological safety is often the key to a willingness to share innovation. So it is helpful to promote such a culture of psychological safety, including learning from mistakes and including living by example. Executives should be obviously using AI and sharing their use cases and experiences within the company like everyone else. As a sidenote: I see little discussions around AIs impact on managers knowledge work so far. It is more about strategy and imperatives - however I miss the own experimentation and exploration. This would be a separate blog I guess.
Talking of sharing use cases: It is certainly helpful to build AI benchmarks for your organization, build and share prompts and tools that work and perhaps don’t work… yet. Like Agents.
The important role of Ethics
To promote trust and avoid misuse, the use of AI in the company should always run along ethical principles and guidelines - which must be defined and concretely operationalized. For example, one should then determine,
- which use cases need to be examined more closely, e.g., when using personal data is used or with regard to possible bias,
- which use cases are a no-go, e.g., surveillance or influencing behavior.
Sooner or later, most knowledge workers will use AI tools for more productivity or creativity. Whether special tools or the so-called co-pilots or agents that are integrated into existing software like Microsoft Office or Apple Intelligence. Not to forget SAPs digital Assistant Joule. It will thus change tasks, roles, processes, and jobs. This will also influence the design or redesign of organizations. In addition to the usual criteria such as desirability, feasibility, and profitability, sensible criteria are also how digitizable, automatable, and cost-intensive tasks are.
Regulatory, technical, and ethical questions that elevate the topic of adoption to a societal level are to be further discussed.
Due to the dynamic development in the field of AI, adoption, acceptance, upskilling, and more as described will be relevant for knowledge workers beyond 2024.
Without focusing on people, AI implementations will not only be less successful - they can even be harmful. Let's prevent this and work together on a human-centered use of AI.
Sources and further information
Newsletter on AI & Education from MIT Professor E. Mollick
Co-Intelligence, Living & Working with AI - Book from Prof. E. Mollick
LernOS AI Guide for peer-learning
Harvard Business Review (2023). Reskilling in the Age of AI
Where Should Your Company Start with GenAI? (HBR, 2023)
In our podcast EducationNewscast we often talk about AI, however most episode are in german language.
1 Comment
