There is a discourse that is becoming increasingly common in business, which was shared with me by someone who works for one of the “giants” of AI just last month. It essentially states that if we are looking for value with AI, we should start with the back office. Internal, administrative, and support functions would therefore be the most rational, safest, and above all, most profitable area in which to achieve concrete results. Conversely, other analyses claim exactly the opposite, explaining that the value lies in the front office with its customer relations, revenue, and differentiation challenges.
This debate sometimes arises in management committees, pitting finance against operations and influencing a large part of investment decisions around AI. However, when we look closely at the facts, this debate says less about the location of value than about how businesses think about their business, their work, and their ability to transform gains into sustainable results.
It is a misleading debate, especially since it is fueled by reputable sources that appear to contradict each other, when in fact they are describing different levels of analysis that are often confused in operational discussions.
In short:
- The debate between back office and front office masks a deeper issue: the ability of businesses to integrate AI into their work processes in a sustainable and structured way.
- The majority of AI pilot projects fail not because of the technology, but because of a lack of real integration into workflows and clear governance of gains.
- The back office is often favored for initiating AI projects because of its operational stability, but the gains there are limited and rarely transformative on a business-wide scale.
- The front office has higher value potential (customer relations, revenue, differentiation), but requires a more profound transformation of work to avoid negative effects or failures.
- For a COO, the real question is not where to apply AI, but how and under what conditions to integrate it in order to produce real and sustainable value, regardless of the targeted function.
Lessons from failed AI projects
A recent MIT report that caused quite a stir threw a spanner in the works regarding the actual status of ongoing initiatives. According to this study, nearly 95% of generative AI pilot projects in business fail to have any measurable impact on the P&L (MIT report: 95% of generative AI pilots at businesses are failing). This figure is often used to fuel alarmist discourse about the immaturity of the technology or the incompetence of organizations. This is a fast and simplistic analysis that does not tell us that these pilots are not working, but simply that there is a gap between what the technology can do and what businesses expect (Technologies sell productivity, but businesses want revenue and AI from productivity to P&L: nothing happens by chance).
But what this report shows above all is that AI is being deployed without being truly integrated into work, existing workflows, and operational trade-offs. In other words, businesses are multiplying pilots, but they are failing to transform these trials into practices that are sustainable and that truly change the way the business operates.
This observation is not isolated. It is echoed, in other forms, by several recent analyses that describe the same difficulty in transforming local initiatives into sustainable value. Gartner, for example, estimates that more than 40% of agentic AI projects in business will be abandoned by 2027, not because of technological failure, but because organizations are unable to derive sufficient economic value from them on a business-wide scale (Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027). .
For a COO, this is an important observation because it immediately shifts the question from where to test to what to do once something works locally.
The back office as an obvious entry point
It is no coincidence that the idea that value lies in the back office is so widespread. Back office functions often involve repetitive activities, explicit rules, relatively stable volumes, and better-identified internal dependencies. Under these conditions, it is easier to introduce AI without immediately jeopardizing the promise made to the customer or the continuity of the business. At least in theory.
From an operational perspective, this makes sense. It allows for more visible unit gains, reduces certain frictions, improves reliability, and speeds up processing without disrupting the entire system. This is often where the first results appear, fueling the idea that this is where the value lies.
Much of the feedback from firms and analysts points in this direction, emphasizing that internal functions are often more conducive to initial industrialization, not for strategic reasons, but because the work is more easily standardized and controllable, as regularly noted in operational analyses of internal automation by Gartner and Deloitte.
But this reasoning has a structural limitation because the gains achieved in these areas quickly reach a ceiling. They improve local efficiency, and sometimes working conditions, but they do not fundamentally change the trajectory of the business.
For a COO, this translates into the impression that the system is working (perhaps) a little better, but that it is still working in the same way.
The front office concentrates promises and failures
Conversely, those who claim that value lies in the front office are not necessarily wrong. It is indeed in interactions with customers, in the ability to serve, convert, retain, or adjust an offer, that the most interesting levers in terms of revenue and differentiation are found.
The problem is that these areas are also where work is most dependent, most contextual, and most difficult to formalize. Workflows are less predictable, trade-offs and the need for judgment are more numerous, and errors have an immediate impact on the perception of value. Introducing AI into these areas without reworking the work itself often amounts to adding variability where the organization is already struggling to deliver on its promises.
This is exactly what several case studies show when analyzing successful uses of AI in customer-facing functions, while pointing out that these results require highly structured processes and strong execution discipline. (How Is Your Team Spending the Time Saved by Gen AI?)
This is precisely why many front office-oriented projects fail to scale. Not because the potential value does not exist, but because the increase in productivity does not translate into any measurable effect at the business level.
And beyond the operational dimension, we cannot fail to mention the relational and even reputational risk. It is at the front office level that customers perceive a brand’s identity and whether or not it delivers on its promise. Here, we are not only talking about operational excellence, but also about consistency between identity, promise, and execution.
How can we not mention the case of Klarna, the fintech company that wanted to entrust its customer service to AI? Contrary to what was said, the technology worked perfectly, but it was the change in the relationship that led to discontent and then customer churn. Klarna had gained in efficiency, perhaps in costs, but had lost what made it successful, the way its culture infused customer service. (When AI Turns Your Secret Sauce Into Ketchup). This is a very good example of what happens when technology is prioritized over business design (Efficiency vs. uniqueness: the false dilemma of operations).
But for every Klarna that has openly communicated on the subject, how many other failures have gone unreported?
The dividing line is not organizational
When we put these findings together, the back office versus front office debate appears for what it is: an oversimplification. The dividing line is not between functions, but between forms of work. Where work can be isolated, repeated, and integrated into a relatively stable flow, AI more easily produces observable effects. Where work is highly dependent, poorly articulated, and unstable, it mainly reveals existing weaknesses.
The MIT report shows this indirectly. The projects that fail are not those that target one function over another, but those that remain at the tool stage, without transforming work or governing gains (Without governance, the gains from AI are virtual). Conversely, when a use case is truly integrated into an existing work sequence, it can generate value in both internal and customer-oriented activities.
The wrong way to arbitrate
Having had a number of discussions with professionals in the field, arbitration is carried out in relation to “acceptable” risk. So yes, risk management is vital in such processes, of course, but there is a difference between using risk as a management tool and making it the sole criterion for decision-making.
What bothers me here is that the issue of transformation is secondary and that this approach leads, in my opinion, to decisions being made for the wrong reasons.
I was recently talking to the head of a major digital services company who told me that they prefer to steer clients towards front office issues for two reasons.
The first is that it is visible, gives an impression of modernity, and makes it easy to communicate (although I suspect that since Klarna, they have tempered their position somewhat or that their clients are challenging them a little more).
The second is that if you are aware of the risks associated with the front office, no one wants to take risks with their HR, finance, logistics, etc., because not only can this lead to operational risks that will eventually impact the front office, but there is also sometimes a legal risk involved.
All these arguments are valid and indisputable, but once again, for me, they should be tools for arbitration and management, but should not be objectives in themselves.
What this means for a COO
From an operational perspective, the question is not whether to choose between back office and front office, but rather to ask where the organization is capable of stabilizing practices without creating more disorder than it resolves. Starting with the back office may be a pragmatic choice, provided it is not confused with a value creation strategy. Moving towards the front office becomes inevitable if you are looking for a lasting impact, but only if the work and dependencies have been reworked beforehand.
What the massive failures that are much talked about today show is that most businesses look for value where they know how to deploy tools, not where they know how to transform their business.
Bottom Line
Value is not found naturally in the back office, nor magically in the front office. It is built where the business is able to integrate AI into stable workflows, manage the resulting gains, and make the necessary trade-offs. The MIT report says the same thing, even though it is often read as a statement of technological failure when it primarily describes an organizational failure.
For a COO, the real challenge is therefore not to choose sides, but to recognize that value appears where time has been taken to transform the work, and especially where decisions have been made in advance about what to do with the gains once they are there.
To answer your questions…
Neither, by nature. The value of AI depends above all on the business’s ability to integrate it into established workflows. The debate pits functions against each other, when the real difference lies in how work is structured, governed, and transformed.
Because they remain at the pilot or tool stage. Businesses improve productivity locally without defining how these gains will be translated into sustainable economic results or integrated into existing operational processes.
Activities are more repetitive, more standardized, and less risky for customer activity. This allows for visible gains to be achieved more quickly, even if these gains often remain limited and have little impact on the business as a whole.
The front office directly affects revenue and customer relations, but the work there is more complex and contextual. Without a redesign of processes and trade-offs, AI can degrade the customer experience despite its real technical efficiency.
By asking where the organization can stabilize AI usage and govern gains, rather than pitting back office against front office. The priority is transforming work, not locating use cases.
Image credit: Image generated by artificial intelligence via ChatGPT (OpenAI)
Related articles:
Who is handling your artificial intelligence projects? Probably not the right people.
AI in the workplace: going beyond augmentation to actually transform
HR and IT merger: Moderna redesigns its organization for and with AI
People Centric Operations 2.0: how AI is reinventing knowledge work at scale
AI adoption does not replace productive appropriation
What works today with AI without any particular effort
