Ellmer & Dhar (BCG) — Is Your AI Rollout Built to Fail?
▶ Watch on YouTube · Boston Consulting Group · 24:36
Kristy Ellmer and Julia Dhar, co-authors of How Change Really Works and founders of BCG’s Behavioral Science Lab, explain why so many transformations fail and what the science says separates success from failure. They break down the “messy middle,” why people resist being changed rather than change itself, and how giving employees real agency turns resistance into momentum.
— Channel description, Boston Consulting Group (abridged: chapter list and links omitted)
TL;DR
A 25-minute episode of The So What from BCG, hosted by Georgie Frost. The guests are Kristy Ellmer (North American leader of BCG’s Transformation practice) and Julia Dhar (North American leader of People and Organization), who co-founded BCG’s Behavioral Science Lab and co-wrote How Change Really Works: Seven Science-Based Principles for Transforming Your Organization. Human-curated transcript.
It is a book launch, and the AI content is one thread in a general account of organisational change. That thread is what the wiki needs: most sources on enterprise-ai-adoption say adoption is a people problem; this one says which people problems, in which phase, with a named behavioural-science finding for each remedy.
- “People don’t dislike change. They dislike being changed.” The opening line, and the frame for the rest.
- Three failure points by phase. At the start, false alignment among executives. In the middle, assuming commitment will arrive by itself — the messy middle, where momentum “leaks”. At the end, failing to sustain effort over a long period.
- What makes AI transformations different: the intensity of change (“the single most disruptive change that any of the five generations currently in the workforce are likely to see”) and the threat to identity.
- Redesign the cockpit, don’t hand out the tool. The Paul Fitts story from WWII aviation, applied to AI rollouts.
- Agency over participation, via the IKEA effect: people value what they helped build.
- Constructed endings and fresh starts as tools against change fatigue.
No new data. The research cited (IKEA effect, Milkman’s fresh-start effect, a BCG figure that ~90% of employees expect to need significant reskilling) is secondary and not sourced in the episode.
The definition, and the opening claim
Ellmer defines transformation as “any deliberate effort to shift how people… get something done together” — an AI rollout, a merger, a change in customer service. The test is behavioural: “are people coming and acting differently six months from now?” Or: “It’s actually about changing how people show up on Tuesday morning.”
Dhar’s reason for the book: failure rates are well documented, “I frankly think we’re not scandalized enough”, but successes “are not random. They leave clues.” The book assembles those clues from psychology, economics and neuroscience, updating Jeanie Duck’s The Change Monster (2001) with research that did not exist then.
Where transformations fail, by phase
At the start: false alignment. Dhar calls the fix “the cheapest behavioral science hack of all time”: give each executive a piece of paper and ask them to write down the transformation’s goals and how they’ll know it worked. “You’ll be surprised by how often you get wildly different answers.”
In the middle: the messy middle. Ellmer’s description is the most useful passage of the episode, because it gives observable signals:
“The messy middle isn’t dramatic. It’s actually quite quiet… Meetings become updates instead of decision meetings. Initiatives maybe slip by two weeks. Maybe then it’s a couple of months. And you start to hear a lot of reporting increases but actually a lot less discussion and problem-solving… Momentum isn’t something that disappears overnight. It actually leaks.”
At the end: sustaining effort. A CEO on their third turnaround told Dhar, “I had forgotten how difficult it would be.” The advice is to plan for periods that are “difficult and boring and painful.”
These are contextual/internal-barriers in the W&W sense, and all three are inside the leadership team rather than in the workforce. The episode is explicit that blaming employees is the wrong diagnosis (below).
What makes AI different: intensity and identity
Dhar gives two differences. The first is scale — five generations in the workforce, and for all of them probably the biggest change of their careers. The second is identity:
“People are asking, ‘If I don’t write software code, who am I; what is valuable about being a person in this organization?’ at the same time as they are asking, ‘What should I advise my child to do; do I think my spouse or significant other will have a job?‘”
So the emotions are “likely to be more significant than in other situations.” The same identity effect appears, measured, in Kropp et al.’s BCG experiment: 13% higher identity uncertainty when AI is framed as an employee. The two come from different parts of the same firm. This bears on ai-employment-effects: the workforce question here is felt before any job is lost.
Asked what the end point of an AI transformation is, Ellmer says nobody knows. We are in a “change economy” where change is constant. But “humans need endings”, and organisations report “my people are tired.” Her answer is to construct endings — chapters with a visible start and finish — which improve confidence and reduce the cognitive load for the next change.
That is a direct disagreement with Ross and Schneider (Egon Zehnder), who argue that transformation and change management presuppose a fixed destination and should give way to adaptability. Both sides agree the destination is unknown and that adaptability is the capability that matters — Dhar opens the episode by saying “adaptability is something that we can learn, and you can build it into the system.” They disagree on whether bounded programmes still have a role. For Ross and Schneider they don’t. For Ellmer and Dhar, bounded chapters are how people get through continuous change.
Redesign the cockpit
Ellmer’s best AI-specific point is a story. In WWII, US pilots crashed often, and the Air Force hired psychologist Paul Fitts, who found the cause in cockpit design: controls placed against how people naturally use them. Redesigning the cockpit improved performance immediately.
“If we think about that story and we put AI into it, right now we’re sort of giving people a tool and we’re saying, use AI, right? Use it into your workflow. And what people are doing is sort of small changes. They’re sort of playing with it. We actually need to think about how do we redesign the cockpit with AI.”
This is the micro-productivity-trap seen from the operator’s seat, and it matches Dutt et al.: a tool dropped into an unchanged workflow produces task-level tinkering, not firm-level change.
Dhar makes the same point another way: “If your assumption is that a car is just a faster horse, you miss a lot of the chances to think about the driver.” She also criticises “get on the bus” messaging. People already know they must reskill — BCG research says close to 90% of employees expect to need substantial retraining. What a bus has, and most AI rollouts lack, is “a large sign on the front that tells you where the bus is going” and planned stops.
When momentum fades
If an effort seems to have fizzled, Dhar separates feelings from facts. Check progress against the original objectives, and measure how the people expected to change feel about it. If it has fizzled, there are two options. One is to find someone to blame, usually in the form of “people are not very engaged or not very motivated.” The other, following Katy Milkman at Wharton, is to declare a fresh start: change is more likely at emotionally significant milestones, and “just declaring a fresh start can be psychologically incredibly powerful” — “not an admission of defeat.”
Ellmer treats momentum as a management system: break initiatives into smaller pieces to “show some more wins on the board.” Dhar: “all of us are addicted to progress.” Making progress visible is “a free gift to people doing the hard work… and… to the shareholders.”
Agency, not participation
The chapter on ownership is where the book’s argument is most distinctive. Ellmer cites the IKEA effect: people who built IKEA furniture valued it 63% more than people who bought it assembled. Dhar: “people don’t burn down houses that they build.” So the transformation should let people make micro-decisions and join working groups that have real say — digital-sensing/digital-mindset-crafting and strategic-renewal/organizational-culture as a design brief.
Dhar takes aim at three phrases:
- “We need to bring people along” — “my own mental image of like dragging people into something.”
- “We just need to get people excited” — telling people how to feel “is a bit like telling people not to get defensive.”
- “Change champions” — without the ability to raise and resolve issues, “they just become propagandists for the state.”
This is where the episode lines up with Carucci. Both reject the premise that employees resist change as such; both put the cause in how the change is being done to them.
Her COO anecdote makes the practical case. The COO planned to enforce a new procedure by standing in the control room. “Solid plan until the point at which you have to go to the restroom or get a coffee, and then the whole change agenda falls apart.”
She also notes that “all of the hyperscalers, all of the model companies are talking now about human agency”, which nobody was two years ago when they planned the chapter.
What leaders should do
Ellmer’s three things for a CEO:
- Treat the change as a product employees have to buy into, and design the system to make buying in easy.
- Design for how employees will engage, not the “30,000-foot… investor view” of pillars, KPIs and EBITDA impact.
- Don’t scale back after launch. Leaders put in the effort to prepare and then pull back just as the change reaches the organisation; they should “dial in”.
Dhar’s three next steps: make clear and specific requests; for a struggling change, ask honestly “why would anyone do this?” — “because I told them to” is not enough; and make progress visible and positive.
Ellmer’s close is the line that sums up the book: “Leaders sometimes think successful transformation requires extraordinary people. And I don’t think it does. It actually requires ordinary people working in systems that help them succeed.” The same allocation shows up in Hines-Pierce’s 70% people, 20% process, 10% technology.
Linked entities and concepts
- Entities: Boston Consulting Group (channel; both guests’ firm).
- Concepts: enterprise-ai-adoption, micro-productivity-trap, ai-employment-effects, dynamic-capabilities.
- Dangling (single-source mention, deferred): Kristy Ellmer, Julia Dhar, Georgie Frost, Katy Milkman, Paul Fitts, Jeanie Duck; the book How Change Really Works.
Source quality
Human-curated captions. A book-promotion podcast by the authors’ own firm; BCG sells transformation work. The behavioural findings are named but not sourced in the episode, and the 63% IKEA-effect figure and the ~90% reskilling figure are unverified here. The book itself is not in the corpus. The value is the vocabulary (false alignment, messy middle, constructed endings) and the observable signals of a stalling change, not new evidence.