This article is based on the Allwork.Space Future of Work Podcast episode Dave Wright on Why AI Strategy Keeps Failing. Watch or listen to the full episode.
The biggest AI mistake companies can make may be assuming the technology itself is the transformation.
Companies are spending heavily on AI, yet many are struggling to see meaningful returns. For Dave Wright, the problem often starts before the technology is even deployed.
Wright, Chief Innovation Officer at ServiceNow and co-author of Infinite: How Visionary Leaders Transform Today’s Businesses into AI-Forward Companies, joined the Allwork.Space Future of Work Podcast to discuss what companies are getting wrong about AI. His argument is that organizations too often use AI to automate work they already know how to do, without questioning whether the underlying process still makes sense.
That can leave companies paying for AI while essentially running the same business they ran before.
AI should start with the business, not the technology
Wright compared the current rush toward AI with the earlier era of digital transformation, when companies could add digital tools without making meaningful changes to how they operated.
He sees a similar pattern emerging with AI. Companies decide they need to “deploy AI,” then search for places to use it. That puts the technology at the center of the strategy when it should be serving a larger business objective.
“AI should be an enabler, not the strategy,” Wright said during our podcast conversation.
His preferred approach starts with the business itself; a company should first determine what it wants to accomplish, then work backward to identify the resources required and where AI could help create them.
For example, a company that wants to expand into a new market might use AI to reduce the amount of human capacity required for existing processes. The goal would not be to automate for the sake of automation. It would be to free up enough time and resources to pursue the larger business objective.
That also changes how companies choose their first AI projects. Wright recommends selecting an important workflow that crosses multiple organizational silos rather than starting with a mundane task that few people care about.
Employee onboarding or supply chain processes, for example, can expose problems across several functions while giving the company a meaningful opportunity to learn how AI changes the way work gets done.
AI changes the assumptions behind the workflow
The deeper issue is that many corporate workflows were designed around human scarcity. For years, workflow design has largely been about figuring out how to accomplish a task with a limited number of people. Organizations optimize processes around the amount of human time available and try to make those resources as productive as possible.
AI introduces a different possibility. If technology can dramatically increase the amount of work a company can execute, the assumptions behind an existing workflow may no longer hold. A process designed around having limited human capacity was never built for a world where certain tasks could potentially be completed at much greater scale.
That means simply inserting AI into the existing process can miss the larger opportunity. Wright gave the example of a 10,000-person company that automates 30 minutes of work for every employee. If the company does nothing else with that newly available capacity, the productivity gain may have little practical impact. The employees simply have more time available during the day.
But if the company treats that recovered capacity as a resource, it could potentially use it to expand into new markets, develop additional services or pursue other business opportunities. The question therefore becomes what the organization intends to do with the work it no longer needs people to perform.
Automation can expose the workarounds companies stopped noticing
Redesigning workflows can also reveal problems that employees have learned to work around. Wright described this as one of the potential benefits of applying AI to a process. Employees often know the informal rules that keep a workflow moving. They know who to ask for help, which workaround to use or when something needs to be submitted to get through the system.
Those workarounds may be invisible in the official process documentation. AI does not automatically possess that institutional knowledge. When it reaches a point where a process breaks down, the bottleneck becomes visible. Fixing that bottleneck can then reveal another one further along in the workflow.
For Wright, this makes AI implementation an opportunity to examine how work actually happens inside an organization, rather than simply automating the version of the process that exists on paper. That matters because some of the most important decisions inside a company may depend on experience and tribal knowledge that was never formally documented.
Ford shows the risk of removing judgment from a process
Wright pointed to Ford as an example of what can happen when companies automate a process without adequately accounting for the expertise built into it. He referenced Ford’s decision to eliminate 300+ quality-engineering roles and use AI for quality assurance, followed by the company’s decision to rehire those engineers after customer satisfaction suffered.
The example illustrates the difference between automating a task and replicating the judgment involved in performing it well.
Experienced employees may recognize situations that fall outside a standard process, draw on knowledge accumulated over years and understand when a seemingly straightforward decision requires additional scrutiny.
An AI system can follow the workflow it has been given without necessarily possessing that same contextual understanding.
For companies redesigning work around AI, that makes it important to identify where human judgment actually sits within a process before deciding which parts can be automated.
The AI workforce may require fewer people doing more
That question is also connected to how companies think about growth. Wright pointed to AI-native companies that have reported very high revenue-per-employee figures compared with traditional enterprises. His broader point was that companies may increasingly be able to grow revenue without increasing headcount at the same rate.
That creates two possible approaches: companies can use AI primarily as a cost-cutting tool and reduce their workforce, or they can use the capacity created by AI to grow the business while keeping headcount growth lower.
Wright sees the second approach as the more creative opportunity.
It also changes the traditional relationship between company size and employee count. Businesses historically used workforce size as one indication of scale. If AI allows companies to generate substantially more output with fewer people, that relationship could become much weaker.
The implication is less about predicting a specific future headcount and more about reconsidering how organizations are designed when human capacity is no longer the only constraint on growth.
Companies need guardrails before they deploy AI
The more responsibility companies give AI systems, the more important those design decisions become. Wright argued that organizations should establish their AI governance before implementation rather than deploying systems first and attempting to add rules afterward.
That means deciding in advance what AI can do, where human intervention is required and what AI should never be allowed to do. Companies should also determine how they will measure the system and what return they expect before deciding how and where to deploy it.
This becomes especially important as companies move toward more agentic systems that can take actions rather than simply generate information.
Wright described the distinction using a self-driving car analogy. Giving a system an objective without defining the boundaries around how it should achieve that objective can produce an outcome that technically satisfies the instruction while violating what the human actually intended.
For now, he sees “human-on-the-loop” as a more realistic model than completely handing over control. Humans can retain authority over what AI is permitted to do while the technology takes on more of the execution.
Employees need to be able to say when AI isn’t working
Governance also has a cultural component — Wright said employees need an environment where they can question AI without being treated as resistant to technology. That is particularly important because AI can carry an unusual pressure around it. Employees may hesitate to say that an AI system is producing poor results because questioning the technology can make them appear like they are against the company’s AI strategy. That can become a serious problem when organizations start encouraging widespread AI use without establishing a clear purpose for it.
If employees are told that everyone needs to use AI, they may begin using it in situations where it adds little value. The company then accumulates AI usage and associated costs without necessarily improving the underlying work.
Subject-matter experts need to be able to point out when AI is making a process worse, when the technology is being applied to the wrong task or when the process itself needs to be redesigned. That requires transparency from leadership about why AI is being introduced and what the organization is trying to accomplish.
Different generations may have different instincts about AI
The workforce is also developing different attitudes toward when AI should be used. Wright shared an observation from his 20-year-old daughter after an internship — she had noticed how extensively some millennials used AI, including in situations where it might not be necessary.
Wright said his impression was that Gen Z and Gen Alpha can have a more measured relationship with the technology, including a greater awareness of what AI cannot do. He gave examples such as recognizing that AI does not have genuine taste when giving style advice or genuine emotional understanding when providing emotional support.
The point is less about assigning a universal AI behavior to an entire generation and more about recognizing that employees are developing different instincts around the technology. For companies, that means AI fluency cannot simply mean getting employees to use AI more often. It also needs to involve understanding when the technology is appropriate and when human judgment remains more valuable.
The real transformation starts after the AI is installed
For Wright, the companies that get the most from AI will be the ones willing to reconsider the work itself.
That means defining the business objective first, identifying the workflows that stand between the company and that objective, and then determining where AI can create additional capacity. It means documenting the judgment embedded in those workflows, establishing guardrails before deployment and giving employees permission to challenge systems that are not producing better outcomes.
The technology is only one part of that process; the bigger question is what a company does once AI changes how much work it can perform.
If the answer is simply to make the existing workflow faster, the organization may end up with a more expensive version of the same process. If it uses the additional capacity to reconsider jobs, workflows, decision-making and growth, AI can become something much more consequential: a reason to redesign how the company operates in the first place.
