AI and the Credibility Crisis of Digital Transformations

Digital Transformations and Their Impact on Internal Credibility

In the business world, every failed digital transformation initiative leaves an indelible mark on the internal credibility of the organization. The arrival of artificial intelligence (AI) in companies already exhausted by successive changes could exacerbate this situation. Without a prior assessment of the actual conditions, the promises of AI risk eroding faster than models can produce tangible results.

A few years ago, a large European group undertook an ambitious transformation by adopting the SAFe (Scaled Agile Framework) framework. This process included mapping value streams, identifying Agile Release Trains (ARTs), training teams, and planning program increments (PI Plannings). Although operations began to function after six months, eighteen months later, 40% of the features were blocked due to unresolved dependencies with three external cross-functional teams. While SAFe offers solutions for these issues, such as shared services or expanding the ART's scope, none of these solutions were implemented because no one had the authority to integrate these teams into the train.

Accumulated Credibility Debt

This scenario is not isolated and has been repeating for two decades. ERP deployments in the 2000s, cloud migrations in the 2010s, and the adoption of large-scale agility since 2015 have all produced results below expectations. The usual diagnostics often point to a lack of support, resistance to change, or poor implementation. However, these explanations mask a deeper and rarely measured cost: the loss of internal credibility.

Employees who have experienced multiple transformation cycles have learned to wait for the storms to pass. They adopt the current jargon, participate in rituals, fill out required documents, but continue to work as before. This superficial adaptation is a rational response to an environment where promised transformations do not materialize.

AI and the Acceleration of the Disappointment Cycle

Generative AI enters this already tested context with an additional challenge. Formal AI programs, those that go through the executive committee with a budget and promises of results, are not comparable to the informal adoption of ChatGPT or proof of concepts in shadow IT. Where an ERP required several years to deploy, and a cloud migration took between eighteen and thirty-six months, AI significantly reduces these timelines. An executive committee that has attended a demonstration expects results within weeks. Disappointment therefore arises more quickly and in front of a larger audience, as AI touches on sensitive areas such as customer relations, document production, and decision support.

To deploy a large language model (LLM), it is essential to have accessible, well-governed, and sufficiently high-quality data. However, this prerequisite is rarely met. The main issue lies elsewhere: when an AI program requires teams to structure data, validate results, or review their processes, it addresses collaborators who are already exhausted from previous agile transformations, cloud migrations, and product reorganizations. The power of the model does not compensate for the accumulated fatigue.

Reasons for a Lack of Prior Diagnosis

Why do organizations not conduct a diagnosis of their actual state before launching a transformation program? The answer lies in the very structure of the decision-making process.

Decisions to adopt a transformation program are generally made by four actors whose interests converge towards the same outcome:

• The vendor or integrator, whose business model relies on the adoption of the program.
• The internal champion, who needs a visible program to secure a budget and legitimacy.
• The executive committee, which buys a promise of results without committing to an audit of technical debt.
• The IT department (DSI), which knows the actual state of information systems but avoids confronting it directly to not be perceived as a hindrance.

Each of these actors finds an individual benefit: a signed contract, a budgeted program, a presentable story in the boardroom, an IT department that is not seen as an obstacle. The collective result, a launch without diagnosis, is no one's decision. No one has the mandate to say no.

Crossing Technical and Governance Dimensions

In a previous article, I explored how organizations optimize each area locally without diagnosing the constraint that limits the system as a whole. The issue of unverified prerequisites follows the same logic: each component is evaluated separately without crossing the two dimensions that determine what a program can actually produce.

• The first dimension is technical: the state of information systems, the level of application coupling, the quality and governance of data, and the realistic decoupling capacity over three to five years.

• The second dimension is governance: the effective level of delegation on priority decisions, the ability to arbitrate between areas when value streams contradict each other, the actual tolerance for prioritization conflicts, and the level of residual credibility of transformation programs with teams, which is rarely measured.

SAFe addresses this dimension through several mechanisms: Lean-Agile leadership, Lean portfolio management, participatory budgeting. The Team Topologies model assumes a consumable self-service platform that cannot exist without clear investment arbitrations. AI requires data governance that touches on the prerogatives of business units. Each approach documents the importance of this dimension, but none make it a go/no-go criterion.

A program can progress with a partially coupled information system if governance allows for real-time management of dependencies. A well-architected system produces nothing if priority decisions remain centralized by silo. The two dimensions interact. Evaluating them separately amounts to not evaluating them at all.

Concrete Actions for Organizations

The following recommendations are aimed at organizations that have already gone through several transformation cycles, where credibility debt exists, and where at least one decision-maker has measured its cost.

• Change the deliverable of the diagnosis: Value Stream Mapping already identifies dependencies, transfers, and bottlenecks. The problem lies in how it is used: in the SAFe process, these findings serve as inputs for designing ARTs, not as a map of constraints to address before slicing. Using VSM as a diagnostic deliverable rather than as a step towards slicing would change the nature of the program. The same logic applies to AI: a cross-sectional assessment of data quality and governance before the catalog of use cases. Minimum condition: a sponsor willing to accept a deliverable that is less appealing than what they had planned to present to the committee.

• Make foundational elements budget-visible: in almost all observed programs, prerequisites are supposed to be built in the process. They are never built, systematically arbitrated in favor of delivery. Without a distinct line, without clear ownership, without an independent roadmap, foundations do not survive the first prioritization review. What does not appear in the budget disappears from arbitrations. Minimum condition: a sponsor who agrees to sanctify a line without immediate business deliverables.

• Introduce an actor whose mandate does not depend on the launch: if the four actors structurally converge towards "let's launch," the only way to produce a reliable diagnosis is to mandate someone whose interest is not aligned with adoption. An evaluator before the go decision, with a simple deliverable: here is where you stand on the two dimensions, here is what this positioning allows you to expect, here is what it does not allow you to promise.

The Current Challenge for Organizations

The next wave of AI will produce results in organizations that still bear the marks of previous waves. The model will be powerful. The use cases will be real. And the teams that will be asked to engage will have already learned, program after program, that the promises of transformation have a limited lifespan.

The question that will determine the future is not technical. It can be summed up in one sentence: does this organization still have the necessary credit for people to believe in it once more?