Your AI team has developed a stellar AI model based on deep learning. The initial results are extremely promising. You have integrated the model into a tool. AI adoption is happening. Your employees will use the tool to improve their work. Ready to roll.
Not so fast. You are going to face lack of trust to AI capabilities, changing workflows, and many other obstacles. These obstancles might render your fabulous AI tool obsolete. Therefore, consider these three best practices for AI adoption to change your business for good. With the right approach, your fancy new AI-tool will be actively used.
In this blog post you learn how to succeed with intelligent tools. You will learn how to think about AI implementation in your organization. It’s not enough that you have the tool. You also need to think how it is adapted and coordinated in the organization to be truly taken into action
Cornerstone in AI adoption: One size does not fit all
AI solution needs to be adopted into the workflows of employees. Therefore, local testing with different user groups is essential. People need to see how they use the tool and how does it impact the way of working.
Often, leaders are too confident that their solution is suitable for everyone. They might conceptually know that one size does not fit all. Still, they fail to give people room for validation and customization. Their Gantt-charts move directly to deployment after the introduction of the tool. The validation and customization periods are missing.
1. Encourage local validation and customization of you AI solution
Extensive research by Jillian Chown at Northwestern University provides guidance for local validation and customization. She found that teams had to go through two stages to benefit from new solutions.
First, they had to validate if the tool fits their context. They would reflect and test if using the solution was feasible and if it truly added value for their processes.
Second, if yes, they would customize the tool to suit their context. They would make small tweaks here and there to make it easier and faster to use. They would also change some of their working practices to make the tool use smoother.
As an alternative second step, if the first step indicated the tool would not fit the local context, the local teams would find an alternative solution. Typically, the tools are not implemented for the sake of having the specific tool. Rather, tools are introduced to improve team or organizational productivity. In cases where the initial tool did not fit the purpose, the teams created alternative solutions for the need.
Hence, when you are plannign AI adoption in your organization, remember to give the organization time for validation and customization. Teams won’t get more productive if you force them to use the same tool in the same way regardless of their local conditions.
2. Define boundaries for local customization of the AI solution
Local customization is good. But things will get out of control if you have no boundaries. This applies also for AI adoption. Hence, you need to define basic parameters each team must comply with.
For AI tools, you need to standardize data inputs. Regardless of how each team chooses to use the tool, they need to record specific data. This is important because AI can learn only with sufficient and valid data. If teams use different standards for recording their actions or outcomes, AI’s learning is compromised.
Murray and colleagues’ illustration of New York Post’s editorial decisions shows how teams could use AI in different ways. The editorial team could either accept AI’s recommendation of the lead story as such. Alternatively, the team could debate on the recommendation and ultimately make its own choice. Either use would be fine and part of the local customization of the AI tool.
However, what should be standardized among all users of the AI tool is the data. The team should record the ultimate decision into the tool. The tool should also have full access to the performance metrics of the chosen lead article. In this way, the tool would collect standard information (action and its outcomes) even if the team had the right to use the tool in a customized way.
3. Get buy-in for AI adoption through concrete decisions, not only through vision and instructions
Discussing vision and values is good. But only for a while. If you want people to commit, you need to get concrete.
This applies also to AI adoption. You might spend 30 minutes preaching about digital transformation and the virtues of data-driven practices. But then you have to get to the point.
And the point for most people is what do they need to do differently. An excellent study by Stanford professor Melissa Valentine brought this point home strong: When change agents organized value workshops and spoke in generic terms, they failed to make any real progress. In contrast, when they negotiated with organization members in concrete terms and forced them to make decisions, change progressed effectively.
So, when you are bringing in the new AI tool, be concrete. Make different managers and teams negotiate what they should do differently and why. This makes them commit to the required new behaviors. In addition, the negotiation process helps them build understanding of the organization-wide goal of the AI implementation.
Conclusion: To succeed in AI adoption customize concretely within clear boundaries
Your AI tool is going to boost your organization’s productivity. But only if people use it. They won’t use it if they feel it does not help them. You need to let them customize it for their context and make them commit to specific behaviors. Hence, to succeed with AI adoption take these three steps:
- Encourage local validation and customization of you AI solution
- Define boundaries for local customization of the AI solution
- Get buy-in through concrete decisions, not only through vision and instructions