The value gap
A faster step is not a faster journey.
Imagine a claims specialist who can summarize a file in seconds. If the next decision still waits on a missing document, an ambiguous policy rule, and an approval in a different system, the claim has not moved much faster. The task improved. The workflow did not.
That distinction appears across otherwise different studies. In its 2025 global AI survey, McKinsey reports that only 39% of respondents see an enterprise-level EBIT impact from AI, even as adoption grows. BCG found that just a quarter of surveyed executives reported meaningful value from AI initiatives. These are different surveys with different definitions; they are not interchangeable measures of the same phenomenon. Together, they describe a gap between access to AI and enterprise-wide results.
The question for leaders is not simply, “Which tasks can AI do?” It is, “What needs to change so that an improved task actually changes an outcome?”
What the research says
The work between the tools.
Fund the function, not the feature.
BCG reports that companies it identifies as AI leaders direct more than 80% of AI investment toward reshaping key functions and inventing offerings. The comparison is not a proof that an 80% budget mix causes superior returns. It does suggest that the object of investment matters: a whole function can change its economics in ways a stand-alone assistant cannot.
Measure the entire path.
Bain notes that coding assistants can boost developers’ productivity by 10–15%, yet saved time often does not become higher-value output. Coding is only one portion of the path from idea to launch. A gain inside one stage may disappear into testing, coordination, review, or release. The same logic applies to intake, adjudication, and payment.
Change the sequence.
Research discussed by MIT Sloan models work as interdependent tasks, not a bag of isolated activities. How tasks are sequenced and handed off between people and machines affects the value of automating any one step. That makes workflow design an economic question, not merely an interface decision.
Redesign decisions, too.
PwC frames the AI-powered enterprise around changing how work, decisions, and accountability operate, rather than just making individual tasks quicker. McKinsey likewise observes that its AI high performers are more likely to redesign workflows. Neither source establishes that any single redesign method guarantees a return.
The Loresse view
Build around the exception, not the happy path.
In regulated work, the value of a workflow is often decided at the moment the standard path breaks. A record is incomplete. Evidence conflicts. A policy needs interpretation. Someone must know which expert can resolve the exception, what authority that person has, and what must be recorded for the next review.
That is where institutional knowledge lives: in the judgment that connects evidence, rules, context, and action. Our interpretation of the research is to start with that knowledge, map the full journey from trigger to resolution, and then decide what AI should assemble, recommend, or route. People should retain explicit authority over consequential decisions. The record of why a decision was made should travel with the work.
This is a design thesis for Loresse, not a finding or endorsement by any of the firms cited here. It also is not a claim of measured customer results. The proof for any particular workflow is its observed end-to-end outcome, including quality, cycle time, cost, and the experience of the people doing the work.
A practical test
Before adding another AI tool, ask five questions.
- 01. What outcome changes? Define a completed case, not just a faster subtask.
- 02. Where does work wait? Trace handoffs, missing context, rework, and exception queues.
- 03. Who owns judgment? Specify authority, escalation, and review before automating a decision.
- 04. What can be inspected? Preserve evidence, decisions, and approvals in a usable record.
- 05. How will we know? Compare the full process before and after, including quality and unintended effects.
Sources & method
Read the original work.
This is an independent Loresse synthesis of published research and perspectives, not a joint study or an endorsement. Survey findings describe associations, often based on self-reported answers; they do not establish that workflow redesign alone causes financial returns. The insurance example and practical questions above are Loresse’s interpretation.
- PwC — The blueprint for an AI-powered enterprise
An operating-model perspective on redesigning decisions and work around outcomes.
- BCG — From Potential to Profit: Closing the AI Impact Gap
2025 AI Radar survey of 1,803 executives across 19 markets.
- Bain — From Pilots to Payoff: Generative AI in Software Development
An example of why faster individual tasks do not automatically shorten an entire delivery cycle.
- McKinsey — The State of AI: How organizations are rewiring to capture value
Global survey on adoption, enterprise impact, and how high performers work differently.
- MIT Sloan — How AI is reshaping workflows and redefining jobs
A research-based account of how the sequence of interdependent tasks shapes AI value.