The Hidden Costs of AI Implementation
A 2025 Roadmap for Success
Happy New Year to all of you! I meant to take one week off for a holiday and time with family, but ended up coming down with a bit of an illness, so it turned into two weeks.
Such is life, but I am excited to be back and ready to continue this journey with you on AI Strategy and how it can play into your business.
A Review of the End of 2024
First, a review of where I left off at the end of 2024. We discussed:
JPMC’s journey to incorporating AI and how they have taken a “future-proofing” approach to be able to continuously incorporate new technologies (such as the many flavors of AI) into their technology stack. They also are focused on bringing their employees along. You can catch up here.
BMW’s incorporation of augmenting quality control processes into their manufacturing processes, especially for the most tedious tasks. This way, the AI system can ensure that the “Gorilla” is spotted, even when the human eye misses it.
Amy Webb’s 2024 Trend Report that she published as part of her Future Today Institute. I appreciate her concepts of scaling AI across function or task as well as the uncertainties and challenges that continue to face companies in highly-regulated industries.
Then there was the class on Decision Intelligence that I took with Dr. Lorien Pratt. I love, love, loved this class. I appreciated the systems approach to AI and its implementations. Decisioning is growing in complexity and the levers and external implications can be hard to quantify, no matter how experienced you are. Check that out here.
As we enter 2025, I have learned that companies are probably missing the value of AI in their strategy.
Why is this? They are most likely focusing on the more visible, tangible aspects of utilizing Artificial Intelligence (and Machine Learning). They are focusing on things like the technology (“it’s a technology problem”) or talent (“where do we find the best AI, Data, ML, MLOps engineers?”).
All technology has always been a means to a solution for business problems. So, by starting with the business problem you are trying to solve (ie: growing revenue in a certain market, reducing risk in your supply chain), then you can better tie business outcomes to the technology that your teams are working to solve.
Cost comes in many forms: there’s the easy form: costs associated with things like capex and opex. Then there’s the nontangible forms that include the “squishy” stuff that make it more difficult to draw a direct line from what you are implementing and the outcomes you are working toward.
The Three Hidden Costs of AI Implementation
Data Infrastructure and Modernization Costs
If your company has been around for more than 5 or 10 years, there is a pretty good chance that you have some measurable technology and data infrastructure debt that needs to be dealt with.
As was demonstrated with JP Morgan, mid-sized and enterprise companies should play the “Long Game” with their technology strategy. That doesn’t mean that a company must invest millions of capex and opex into their architectures and stacks and wait 5-10 years before they see results.
By implementing a strategic approach to modernization, a technology leader can identify those “thin slices” of technology and architecture to deliver value, implement new processes, and incorporate critical security and regulatory requirements into the entire process.
Cultural and Organizational Change
As BMW demonstrated with their roll-out of their quality control system, the utilization of transfer learning allowed employees to train the model thereby reducing the cost of training neural networks while also keeping the workforce engaged and familiar with the suite of new tools.
By putting people first, BMW has been able to reap the rewards of the system that they’ve invested in with substantially fewer false positives, an improvement of production flow, improved employee moral and engagement while minimizing the amount of data needed to get the system off the ground.
Governance and Risk Management Framework Implementation
The details of Governance and Risk Management Frameworks deserve their own set of posts. However, one of the key pieces that I have seen across all industries regarding regulations can be boiled down into one word: trust.
Probably not that surprising, is it?
Trust should probably be the word of the decade the way things have been going.
Trust in AI really refers to the auditability of the data, models and agents throughout the system. Transparency in the models, the training data, and how it impacts individuals and systems should be considered.
I’ll be getting more into this next week, but by doing an assessment in what data, how it will be used, and the potential harms that could arise if the model(s) malfunction (ahem: hallucinate), it will save a of headaches if/when you try to scale.
Actions to Take to Avoid these Costs:
Assess where you are starting from and what you are starting with.
Anytime you go on a roadtrip, you know where you are starting and (hopefully) know where you’re going. At Bridgeway, we have developed an assessment to help companies understand where they are starting from. If you are interested in checking out the MVP that we have published, you can find it here.
The assessment we've developed at Bridgeway specifically addresses these hidden costs by evaluating three critical dimensions:
Data Infrastructure Readiness
Evaluates your current technology stack and identifies modernization needs
Assesses data quality, accessibility, and integration capabilities
Helps quantify the technical debt that needs addressing before AI implementation
Organizational & Cultural Preparedness
Measures stakeholder awareness and buy-in across departments
Evaluates current training and change management capabilities
Assesses team readiness for AI adoption and identifies skills gaps
Governance & Risk Management Maturity
Reviews existing compliance frameworks and their AI-readiness
Evaluates current audit and documentation practices
Assesses risk management capabilities specific to AI deployment
By understanding where you stand in each of these dimensions, you can better anticipate and budget for the hidden costs we discussed earlier. The assessment doesn't just identify gaps - it helps you prioritize investments and build a realistic roadmap that accounts for both visible and hidden costs of AI implementation.
Think of it as a pre-flight checklist for your AI journey. Just as pilots don't take off without confirming all systems are go, you shouldn't launch an AI initiative without understanding your organization's readiness across all these critical dimensions.
Understand and Engage Stakeholders (all of them)
If you build it, they probably won’t come. Kevin Costner got lucky with his baseball field, sorry.
Who are the people who will be interacting with the AI/ML system that you are looking to implement? These include the users, the security folks, regulatory audit, DevOps/MLOps, Infrastructure and (especially) the people most intimately familiar with the strategy and goals of the organization.
If you do not begin with a diverse group of cross-organization folks understanding the common vision, the project will go off the rails pretty quick. This will occur when one or more groups misinterprets their goals, risks are not identified and realized until just before or :gasp: after launch, or the shiny new (expensive) investment collects dust on a shelf because no one wants to use it.
Develop a plan to test, verify and validate your system
Counterintelligence Behavioral Analysis professional for the US Intelligence Services, Robin Dreeke, says in his book The Code of Trust that in order to inspire trust, you must do two simple things: “First: Be eminently worthy of trust. Second: Prove that you are.”
Proving that an AI system is transparent in its data and model structure is easier said than done. I’ll point you to that I found by Mark Dangelo.
Understanding what the representative data is and how the model is behaving provides the transparency that builds the needed trust in the stakeholders and other groups that may be impacted.
Meeting the demands of auditability and other regulatory frameworks such as NIST, and the EU AI Act may complicate the scale of AI in the short term, but I’ll be discussing this more throughout the next few weeks.
Looking Forward
As we move into 2025, successful AI implementation isn't just about having the right technology or talent – it's about understanding and preparing for the hidden costs that can make or break your AI initiatives. By taking a holistic approach that considers data infrastructure modernization, cultural transformation, and governance frameworks from the start, organizations can build a foundation for sustainable AI success.
Remember:
Start with a comprehensive assessment of where you truly are
Engage stakeholders early and often
Build trust through transparency and validation
Think of AI implementation as a journey, not a destination
The organizations that will thrive in the AI era aren't necessarily those with the biggest budgets or the most advanced technology. They're the ones that understand and prepare for these hidden costs, building thoughtful, sustainable approaches to AI implementation that consider the full spectrum of organizational needs.
Next week, I'll dive deeper into governance frameworks and regulatory considerations that every organization should consider when implementing AI. Until then, I'd love to hear about your experiences with hidden costs in AI implementation. What unexpected challenges have you encountered in your AI journey?
If you found this valuable, consider subscribing to AI Strategy Navigator for weekly insights on AI implementation and strategy.