Moving away from Shadow AI

We all are focusing on building AI organizations and tons of effort is going to learn it & adapt it. SAFe introduced AI in the framework and more details with SAFe6.0. Working with a few transformation leaders we came across the challenge that Forbes & Business Insider is also talking about - Shadow AI.

We all know Shadow IT and how most organizations are already learning how to deal with this issue. In Shadow IT, as the use of applications and infrastructure were not under the control of a central body that resulted in redundancy, less security, un-reliability and licensing problems. We have started encountering the same in world of AI as well . Every department and employee in the organization started consuming the AI tools & wind up with hundreds and thousands of siloed and unmanaged AI solutions. Some are less and few are more. Every Leader has a desire to do what makes sense for them & is useful to grow the business.

and I am sure you must have seen too, 25 tools for developer productivity, 30 tools for Product & Business, 40 tools for Project & Program management. But who is making employees aware that how to use AI and prompting in ethical way as well as awareness on sensitive data... We encountered spike in data breach cases soon after AI liberation was provided.

Without management by a central function in the company, the result can be mess of noncompliant, unsecured, and unreliable AI systems too as we can imagine...

Some one can argue that Shadow AI is a powerful force for enabling innovation at the department and the individual level. So what's the harm in it - Agile teams should have mastery & autonomy to drive it. We heard similar voices too- Therefore, in addition to providing the tools to empower every team member (employee with the AI), we need to think about environment to manage and control the resulting AI projects. In other words, you will need to implement an effective strategy to overcome Shadow AI @ scale.

At a Strategy level, discuss started that should it be treated like an independent unit/office for AI work with an independent Agile team enabling CIO/CTO or a team with each Solution Train as Shared team or an independent Agile Release Train.

** I may not use the exact approach we designed, but how in SAFe Framework we enabled it is a worth share & happy to understand your approach ****

Steps we Performed:

We are in the second year of the SAFe transformation and have 3-4 business portfolios operating in Agile Release Train, Solution Train and Product team construct. We are preparing for Lean Portfolio Management practices after the Annual Planning & Budget in to Quarterly buckets. This year like every other organization the key Strategic theme at the Portfolio level was usage of AI but there was a strategic budget ( to fund under Horizon 1 as an investment in the value stream, & horizon 2 to invest as emerging tech).

• We tried experimenting as Enabler Agile Release Train , though the funds may come from various portfolios. Begin with the ART was formed to build the capability as well as enhance/support the business of the other Solution Train and ARTs. Forming an ART was full of challenges. But ART launch tool kit helped to design and map the overall AI vision.
• Governance of an ART : Release Train Engineer, Product Management and System Architect as TRIO to drive the purpose of AI in an Organization being as Enabler than being driver. In the Backlog not only the Functional and Tool category was introduced but also the Governance setup, Trainings and Compliance/Legal aspects. It was unique opportunity to analyze & setup a LEAN process. Business Owners & Epic Owners were duo of business & technical SME that can work in hand.
• Driving unknowns in the PI planning , not commitments but continuous exploration & design thinking concepts
• Feature around Define & Design Access Control ( Access control - Manage who has access to the models, data, and outcome of the AI solutions, and ensure access can be revoked at any moment ( employee change project, company etc.)
• Feature around enable the Monitoring - Track the health and usage of every AI solution/Tool
• 2 Week Iteration , Quarterly Planning, Inspect & Adapt and more of every Iteration the flow was - Spike to analyze & then design or develop. Bringing Techno Functional PO for all the 5 teams we formed - Generic if we say one team of Data scientists to develop Models , One for Tools analysis ( like Thoughtspot Sage Platform, Alpaca or Vizcom, or it is empathetic AI with Hume or Azure cognitive - reviewing each use case from business ask perspective and will it be applicable for other business lines were always a debate. MVPs true meaning we explored with the Tools exploration team ) , One for Platform Use Case analysis (Azure, co-pilot, API design etc.) , One for setting up process, standards, communication, training ( handling Shadow AI instances and drive streamline process) . Kore. AI or Moveworks - comparing the support of helping employee was tough choice but these exploration enablers helped team to set up a process on how to evaluate a tool by capturing maximum business benefit it will give. Requirement gathering of the AI use case template went through almost 10 iterations so the learning was not around AI development but also AI process evolution
• We designed a 2-week demo of these teams together where we invited Stakeholders of the current as well as opportunity to invite new business where they can see what work is happening and they can take an opportunity to drive the Features in next PI using these use case/features /tools available in the eco-system. We found training the existing POs on these AI enabled opportunities as most challenging scenarios and a large effort is still needed to scale. Even if the tools became available there is huge opportunity in front of OCM group to now roll out those tools. Here traditional OCM can be a large blocker in any organization. We have an agile OCM team that is continuously driving this effort with the 4th team of ART but still there is a long way to go

I will share more as we will progress but happy to come on a call and see if you solved scaling AI in a different way or some strategy around change management and communication/training.

We are still in WIP :)

References

1. What Is Shadow AI And What Can IT Do About It? (forbes.com)
2. Why Shadow AI Is an Even Bigger Problem Than Shadow IT - Business Insider
3. Introduction to Artificial Intelligence (AI) - Scaled Agile Framework
4. AI - Scaled Agile Framework