The AI-augmented Operating System: Is your business prepared?
Rising labor productivity acts as a domino-effect for economic growth, higher wages, and enhanced living standards.
The sectors using AI the most (financial services, information technology, and professional services) are seeing labor productivity grow almost fivefold (4.8x) greater than sectors with lower AI exposure (such as transport, manufacturing and construction).[1] Yet, for a third of HR leaders, Gen AI and Large Language Models (LLMs) have brought no benefits to workforce productivity[2], while only 27% of employees expect that AI and/or automation will improve how their job is done over the next three years.[3]
Clearly, technology doesn’t always deliver. To cement future success, business leaders will need to lay the groundwork today to ensure AI is playing its part not only in boosting performance, but making work more accessible. Cue the AI-augmented operating system (AOS).
The prize with the AOS is two-fold:
-
Exponential gains in performance
Bending the demand curve of work will transform performance. For example, one major financial services organization redesigned its workflows around an AI-powered technology platform. The result was a 50%+ gain in productivity as data ingestion was automated, transaction processing accelerated, and customer relationship management augmented by greater insight into customer behavior and needs. In addition to the productivity gains, the significant reduction in errors and less variance in execution greatly enhanced overall performance.
The other source of productivity gains is the reduction of the “experience premium” as the gap between experts and novices narrows. One study estimated a 14% improvement in the productivity of customer service agents using Gen AI. The most pronounced gains were among novice workers who attained the capabilities of experienced agents in just three months rather than ten.[4]
Examples like these will be music to the ears of the fewer than half of executives (46%) who are confident their organization can meet customer demands with their current talent model.[5]
-
Increasing the accessibility and humanity of work
Bending the supply curve of work means increasing the speed and agility of workers and intentionally making space for learning and well-being in the flow of work. This will be how employees work smarter (not harder) and gain more fulfilment and satisfaction in their working lives. For example, on average, 40% of nurses’ jobs involve doing work “below the license”. What if we could substitute that work with AI, automation or junior-level talent so nurses have more satisfaction in the other 60% of work?
Given the trajectory of AI and technology more broadly, the AOS is an inevitability. But what does the migratory journey look like from where we are today to the desired state? Leaders typically start with assessing the feasibility of discrete use cases that focus on improving efficiency and productivity so they can learn in “safe zones” before they earn the right to move to broader business model transformation. Edge cases or high-risk scenarios will need more human oversight and control. Organizations will likely have bodies of work at each stage of the AOS continuum – with legacy human-driven systems at one end and an autonomous, machine-driven model at the other (and human-machine hybrid work in between).
The move to the AOS will center on use cases becoming pilots, and pilots evolving into systemic change that brings everyone, no matter their place in the business, along on the journey. Ultimately, making the jump from discrete AI use cases to full-scale business model transformation calls for a significant retooling of the operating system.
Before making the jump to an AOS, there are some fundamental challenges businesses will need to address to avoid becoming victims of the productivity paradox. This idea, outlined by Robert Solow, economist and Professor Emeritus at the Massachusetts Institute of Technology, suggests that with more investment in technology, productivity may go down instead of up.
There are two reasons for this:
First, early versions of many technologies are often flawed and unsuitable for widespread adoption. But Gen AI has a unique advantage over many legacy technologies. Gen AI:
- Is easy to use and requires relatively little expertise or training
- Can be delivered directly to users’ computers, accelerating adoption
- Builds on a recent legacy of advances in robotic process automation (RPA), machine learning and deep learning
- Is evolving at an unprecedented speed
- Requires limited human supervision
- Is the first technology demanded by both workers and leaders[6]
The second reason relates to the architecture of work: the processes, structure, decision rights, workers’ skills and culture. Digital skills have a shorter half-life than ever before — some technical skills are down to approximately two years.[7] With the latest wave of AI, including the impact of generative tools like ChatGPT, approximately 300 million jobs could be affected by AI and automation.[8] The World Economic Forum’s The Future of Jobs Report 2023 predicts that 23% of jobs will change within the next five years, with 44% of workers’ core skills being disrupted.
Addressing this issue requires systemic work redesign by deconstructing jobs, redeploying tasks and creating new ways of working. Over the next three years, AI and automation will completely reshape work:
From pilots and discrete use cases to intentional design
This evolution is best highlighted by Walmart and IKEA.[9]
Walmart uses Generative AI to handle the negotiation of supplier contracts. This is an example of traditional AI deployment at the strategic level. Walmart launched such an AI pilot to negotiate all its supplier contracts in 2021. The pilot was strictly focused on negotiating not-for-resale items, like shopping carts, fleet services and other equipment used by Walmart to serve end customers. The negotiations were complicated as they included price discount offers based on sales volumes or assortment of items, a variety of payment schedules combined with discounts for early payments, but also different extended payment terms, and the negotiation of different options of contract termination. These pilots resulted in three times the expected rate of successfully reaching an agreement, an average of 3% in cost savings, and an extension of payment terms to 35 days.
A different example of a retailer using AI is IKEA. IKEA's vision is to empower their customers and to be seen also as an Interior Design Consultancy. They have been deploying AI towards Do-It-Yourself (DIY) services and for personalized human interior design advice. Since 2021, they have been reskilling their call center staff to become remote interior design advisers as they delegate routine customer inquiries to their AI chatbot, Billie. In mid-2023, IKEA reported that their chatbot, Billie, effectively managed 47% of customer inquiries since launch and they have successfully reskilled 8,500 call center workers to serve customers as interior design advisers. In 2022, they also launched IKEA Kreativ which offers customers a way to design and visualize home and office spaces from their own computers or smartphones. IKEA Kreativ combines decades of IKEA expertise with the latest developments in spatial computing, machine learning and 3D mixed-reality technologies.
An AOS business has integrated foundational models not only in specific areas of its business but into its entire value chain. This would provide contextual feedback loops along the entire value chain of the business, which includes the enterprise, its ecosystem and its customers.
Contextual services for customers, and enterprise and ecosystem capabilities that can be ‘homogenized’ are the north star of AOS businesses. Building machine learning systems for a wide range of applications – this is homogenization – is standard when designing foundation models and is a core differentiator between an AI-native and digital-native business.
The art of the possible
How will the AOS be experienced by different people?
- 1 Business leaders
- 2 Customers
- 3 Employees
| From | To |
|---|---|
From
|
To
|
| From | To |
|---|---|
From
|
To
|
| From | To |
|---|---|
From
|
To
|
Making the AOS a reality
1. Start with a data strategy
Over two-thirds (67%) of organizations adopt new technology without transforming the way they work.[10] But without the fundamentals of a digital transformation and a data strategy, simply layering AI over the top of your current operating system isn't the solution (especially in the context of free-flowing tasks). It’s critical to understand the relationship between the data underpinning Gen AI before going any further.
Ask yourself: Has the Gen AI tool been trained by trusted data from diverse sources? Where has this data come from? What are its limitations? Is the data all gathered from the same region? What are the implications of this? How will cultural context change how helpful that data is for different employee needs?
2. Assess the opportunity (and risk) by focusing on the objective of the work
This nuance will require organizations to make work design a core capability, enabling them to analyze specific tasks and their objective function or return on improved performance (ROIP).[11]
Here are four prototypical ROIP relationships to illustrate this idea using tax preparation as an example:
-
Eliminate mistakesThis type of ROIP ranges from performance at a very low level, with many mistakes or missing deadlines, to minimally acceptable performance that generates a small positive value. Fewer mistakes would result in tax forms being completed correctly and on time. For work of this type, a human co-pilot must ensure minimum acceptable standards are being adhered to regardless of the situation. This is particularly true when the risk associated with the work is high. For example, if a restaurant relies on AI for food delivery orders and a customer receives a pizza on their doorstep instead of a curry, there is no real harm done. But in a hospital, if AI misinterprets a patient’s data and so gives the wrong medication dosage, the consequences could be severe.
-
Reduce varianceThis applies when performance differences don’t impact value, such as when there are many ways to reach the same goal. Reducing variance produces value not in improving the outcome but by reaching that outcome more uniformly, often reducing costs or confusion. In our example, this would include completing the tax form before the due date, since getting it completed earlier adds no more value than completing it on time. This might be an opportunity for AI to operate on autopilot.
-
Incrementally improve valueThis is used when performance improvement produces a constant incremental increase in value. In tax form preparation, this might involve the clarity and writing quality of the summary letter accompanying a client’s tax form. While a minimally clear letter satisfies the minimum requirement, a more clearly written letter or one that highlights more important issues is incrementally more valuable. This might be another opportunity for an AI autopilot.
-
Exponentially improve valueThis type of ROIP often represents rare or creative performance that surprises and delights a customer or disruptively improves a process. In our example, this might be discovering an obscure deduction or a sophisticated way of restating income to significantly reduce taxes, both of which are opportunities for AI to act as a co-pilot.
An AOS will require business leaders to rethink business plan development, including collapsing the capital planning cycle to become more integrated. Connected to this is the need to level the playing field in terms of how a business accounts for its investments.
While Generally Accepted Accounting Principles (GAAP) might treat different sources of work in different ways with employee costs being expensed versus technology costs being capitalized, management accounting should be using measures like the Total Cost of Work.[12] This normalizes all of an organization’s different “sources of work” to enable a true picture of cost and return. When a business thinks about growth in an AOS, they’re more likely to get a multiplier effect by investing into technology that can learn on its own and benefit employees across departments, compared to investing in a new hire where the immediate gains are limited to that one person.
3. Co-creation with employees
Employees will use AI tools regardless of restrictions, so it is better to involve them in the dialogue and ensure informed usage. Redesigning roles and involving colleagues in the redesign process can help them see how they can spend their time differently. Employees estimate that one-third of their time is spent on mundane and repetitive work currently.[13]
One-third (32%) of executives believe AI will add the most value to their organization by amplifying intelligence to enable higher-quality work.[14] This will require getting close to employees’ day-to-day work life to understand what’s getting in their way. Combine AI pilots with brainstorming sessions to identify new ways to disrupt the business while bringing employees along on the journey.
As aforementioned, co-creation will require a carrot and a stick. Ensure employees are motivated to learn, develop and help evolve the business model, designing space and time for learning and experimentation within the flow of work while penalizing people who choose to stand still.
4. Retain strategic flexibility for perpetual reinvention
5. Encourage adoption and best practice
While the initial step is ensuring business leaders and IT are on the same page regarding AI adoption, the AI model in question cannot learn without the workforce embracing and inputting into it too. To encourage positive human-machine teaming, the “why” should be clear to employees. To do this, highlight how the integration of AI is a partnership (not a threat) that can help to differentiate the employee by enhancing their work, improving efficiency, adding value to their day-to-day, and contributing to an improved employee experience.
Consider AI champions, roll out “test and learn” labs, and/or collect examples of AI best practice from different teams featuring periodic feedback to avoid group think. There is often an imbalance in the adoption of AI tools among senior leaders, middle managers, and junior staff. Ensure these champions and success stories reflect the different personas within your business so they feel relatable.
6. Evolved organizational design and governance
Governance should be multilayered, with different levels of oversight based on the risks involved. This may involve cross-functional teams. For example, IT and HR departments are increasingly being brought together under one lead for better resource management, as Nestle has done under the leadership of Beatrice Guillaume-Grabisch. Governance may also include a centralized method for stakeholders to track, assess and mitigate risks (which can then be used to further educate employees).
Some leaders see potential in cultivating new partnerships with vendors or working with competitors to improve their resilience to cyber threats or address some of the thornier issues of ethical AI usage.
Change is inevitable, and the AOS will be an integral part of it. Business leaders must get to grips with the potential (and the risks) so they can grasp this opportunity with both hands. Success will hinge on a deep understanding of the existing operating system and the data implications of the AOS, before intentionally co-creating the system with employees and staying aware of risks.
The AOS will shake up business models everywhere, changing how we work and who we work with. Every organization will need to see the potential for collaborating with an ever-increasing pool of stakeholders (including competitors). Likewise, leaders must design their organizations and work with greater AI-centricity so the technology can make good on its promises of productivity.
