PwC Says Firms Are Watching and Waiting with AI. That’s How You Lose

Written by Mark Evans MBA, CMgr FCMI (Aka Rogue Entrepreneur)

PwC Says Firms Are Watching and Waiting with AI. That’s How You Lose

I sit in a lot of boardrooms. The line I hear most often is calm and tidy. “We will watch and wait, once things are clearer, we will act”. It sounds responsible. It lowers the pulse in the room. Unfortunately, It also hands your advantage to someone else.

A chief executive said it to me again last month. They wanted the dust to settle. That is not how this kind of change behaves. The dust never settles. It moves to the businesses that get their hands in the work and learn how the pieces fit together. They do not wait for clarity. They create clarity by activating early change process, running small, disciplined experiments inside their own context.

If you are not testing how AI fits your operations, what exactly are you waiting to see. A safe majority. A tidy map. A peer who tells you when it is time. None of that will help you understand how AI interacts with your data, your controls, your customers, or your people. You learn that only by doing. The organisations that move first build muscle, feedback loops, and internal legitimacy that compounds. You notice their tail-lights and wonder when the gap opened.

Figure 1: (Both series shown as index, baseline 100. AI skills wage premium 156 vs 100 baseline. AI exposed productivity up to 400 vs 100 baseline. Source: PwC 2025.)

Figure description for readers:

This figure shows two PwC signals on a common scale so they are easy to compare. The wage premium for roles that require AI skills is shown as 156 on an index where the baseline role is 100. Productivity growth in AI exposed sectors is shown as up to 400 on the same baseline. The message is simple. Employers are paying for capability. Sectors that put AI to work are pulling away on output per worker. Use this as a signpost, not as an anchor. Measure in your own context and act on what you learn.

PwC’s 2025 Global AI Jobs Barometer is a useful signpost. It shows that industries with higher AI exposure are pulling ahead on productivity and revenue per employee, and that roles requiring AI skills carry a clear wage premium. The headline numbers will not run your business for you, but they do tell you where the wind is blowing. Treat them as guide marks on the road, not as the road itself (PwC 2025a; PwC 2025b).

What follows is a working map for leaders who want to move with care and speed at the same time. First, why watch and wait quietly puts you behind. Then, what the best evidence shows about adoption and results. After that, four simple lenses that help you act without the theatre. Finally, a first-year plan that builds capability without throwing the organisation into chaos.

Why watch and wait quietly puts you behind

Waiting feels like prudence. It gives everyone the chance to breathe and carry on as ‘normal’. It also creates a structural disadvantage that is hard to see in the moment and plain to see a year later. By the time the path looks safe, other firms have already done the hard miles inside their own systems. They have learned where value lives and where it does not. They have rewired dull pieces of process that never make the press release but decide the outcome.

There are four losses that stack on top of each other with AI watch and wait.

You lose learning time. Firms do not wake up one morning with a sudden ability to absorb new knowledge. They build it by doing related work early, by cleaning the data, by testing the workarounds, and by writing down what failed and why. Academics call it absorptive capacity. In plain language, it is the muscle that lets you recognise a useful idea when you see it and fold it into your operations before the window closes. If you do not build that muscle now, it is not there when you need it later (Cohen and Levinthal 1990).

You lose balance. Every business needs both exploration and exploitation. Exploration looks for what is next. Exploitation makes what you do today faster and cheaper. When exploration is paused in the name of safety, the culture drifts into neat routines that feel good and age badly. You then spend months trying to reintroduce experimentation to a team that has been rewarded for avoiding it. That is an expensive way to save face (March 1991).

You lose dynamic capability. The firms that navigate change well develop three habits. They sense where the next value might be. They seize opportunities with real investment rather than slogans. They transform operations when the evidence is strong enough to justify the pain. Those habits are not memos. They come from repeated cycles of trying, learning, and adjusting with the core business still running. You cannot wait your way into them. You practice your way into them (Teece, Pisano and Shuen 1997; Teece 2007).

You lose ambidexterity. The best leadership teams protect small bets while keeping the day job steady. They do not confuse noise with risk, and they do not let a single project hijack the main business. When you postpone the small bets, the capacity to run them fades. You end up with a tidy core and no pipeline of what comes next. That is not prudence. That is a slow loss of options (O’Reilly and Tushman 2013).

The signal from the market

You do not need to take this on faith. There is a consistent pattern across independent sources.

Across the G7 and Brazil, the OECD reports that adoption of AI is uneven and closely tied to organisational capability and digital maturity. Larger and better prepared firms move first. Smaller and slower firms tend to wait. The gap does not stay still. It widens with time because the early movers learn faster inside their own context, and then scale the pieces that work (OECD 2025).

Inside the workplace, a series of studies point to practical gains when tools are matched with process and governance. The OECD finds that most workers using AI report higher performance and a better day to day experience. The caveat is simple and important. Benefits are strongest when adoption is accompanied by transparency, consultation, and clear safeguards. In other words, when leaders treat this as operations rather than spectacle (OECD 2024a; OECD 2024c).

At the level of measured productivity, a peer reviewed study tracked a generative assistant rolled out to more than five thousand support agents. The average output per hour rose by about fifteen per cent. The largest gains were seen among less experienced staff, which should make every talent leader sit up. That is what learning by doing looks like when it touches the front line. It is not a silver bullet. It is a practical shift in how people work and how fast they get to a good answer (Brynjolfsson, Li and Raymond 2025).

Set those signals next to the PwC barometer and a picture emerges. Capability is concentrating where firms experiment. Wage premia attach to skills that let people work with the technology rather than around it. Sectors that adopt quickly are pulling away. Independent coverage has also noted a period of turbulence for early adopters as systems and habits adjust, followed by a rebound for firms that push through. Plan for that wobble and you will not mistake it for failure. Ignore it and you will declare defeat at exactly the wrong moment (Reuters 2024).

Four lenses that help you act

You do not need a textbook. You need a few handles that keep the conversation honest.

Diffusion and the chasm. New ideas spread across groups that behave differently. There is a gap between early adopters and the cautious majority. You do not cross that gap by watching. You cross it by creating credible examples inside your own context and telling that story to your staff, your board, and your customers. Waiting hands the narrative to someone else.

The Technology Organisation Environment lens. Adoption depends on three things. Fit with your technology and data. Readiness inside your organisation in skills, structure, and culture. Pressures and rules in your external environment. Leaders cannot control the external. They can decide to improve fit and readiness. Waiting for neat conditions outside your walls is a way to avoid the work inside them.

Acceptance at the coalface. People use tools that feel useful and easy and supported by the team around them. You can design for that from day one. Put usefulness, ease, social proof and basic enablement in your requirements, not in your postmortem. This is the difference between a pilot that people crowd around and a pilot that sits on a shelf because nobody trusts it or understands what it is for (Davis 1989; Venkatesh et al. 2003).

Dynamic capability and ambidexterity. These two sit together. Sense early, seize with small real investments, and transform in steps while the core stays steady. Protect exploration and exploitation at the same time. Treat this like fitness. You do not get fit by observing a training plan. You improve by doing the work, keeping score, and showing up next week to do it again (Teece 2007; O’Reilly and Tushman 2013).

Where watch and wait hits your P and L

This is where theory turns into money.

Data loops. Early movers build feedback systems that improve with use. They clean data at the source, capture edge cases, tag events, and instrument decisions. These loops are not bought later. They are earned by running real work through the pipes. The longer you wait, the more you rely on other people’s lessons and the less those lessons fit your business.

People. Builders want to build. If your firm signals that experimentation is for next year, the very people you need will go to the firms that are learning now. The strongest gains in the support study came from less experienced workers who had tools, examples, and support. If that learning curve plays out in another firm, you pay a premium to hire it back later and you still have to rebuild the missing muscle inside your own operation (Brynjolfsson, Li and Raymond 2025).

Process and architecture. Workflows harden around manual reviews and offline decisions. When you finally try to move, you discover that your controls and training and audit are all designed for a world without real time assistance. You then try to change process and policy in a hurry and learn why that is a poor plan. The cheaper move is to start small now and let your governance mature with your use.

Narrative and trust. Clients and regulators do not want slogans. They want evidence that you can use AI safely and to purpose. If you have run a series of small, visible experiments, you can show the results and the limits and the fixes. That is how trust is built. The same holds inside your firm. People adopt when they can see how a tool actually helps their work and how the guardrails behave when things go wrong (OECD 2024a; OECD 2024c).

Your first 90 days

1. Choose two pilots

   One internal operations task with obvious latency or rework. One decision support or knowledge task. Scope small and agree metrics up front.

2. Co design the work

   Put the process owner, a risk lead, a data engineer and a sceptic in the same room. Let objections become design constraints. Bake in usefulness, ease, social support and enablement from day one (Davis 1989; Venkatesh et al. 2003).

3. Run with light governance

   Logging, bias checks, human override. Adjust as issues show. Do not wait for perfect policy before you start to learn.

4. Measure learning

   Adoption, time to value, overrides, error types, user comments. You are building judgement and speed, not chasing a single model score.

5. Decide

   Scale, pivot or stop on day 90. Treat pilots like options with expiry dates. Avoid sunk cost guilt.

6. Invest in people

   Pair tools with coaching and examples. Give time to tinker and to break safe things.

   
   

Address the AI fear directly

Under watch and wait lives a set of familiar worries. Waste. Reputational risk. A fear of moving ahead of the regulator or the market. Each is real. None requires paralysis.

Work in chapters with audit trails and human oversight. Use the Technology Organisation Environment lens to surface constraints early and line up the dependencies you can control. Protect experiments with firm limits while the core stays disciplined. Treat absorptive capacity as a leading indicator. Ask each quarter if you are getting faster at spotting useful ideas and applying them in your context. If not, ask why, and adjust your plan (Cohen and Levinthal 1990; Teece 2007; O’Reilly and Tushman 2013).

Be honest about turbulence. Many firms experience a dip before the climb. That is the cost of changing workflows and habits. Plan for it. Resource it. Do not rebrand the first bump as failure. Teams that stay the course with a clear cadence tend to rebound stronger, because the pain bought them real learning that competitors still have ahead of them (Reuters 2024).

Entry level work and the graduate question

Several employers have paused graduate intake while they reassess entry level work under AI. It is understandable. It is also risky in the long run. The wider labour signal is more nuanced than a simple substitution story. The PwC barometer links higher AI exposure to faster productivity growth and to rising wage premia where AI skills exist. That is a market clue, not a press line. Roles are being reshaped, not simply removed (PwC 2025a; PwC 2025b).

Use that reality. Make early AI projects part of development pathways for junior staff. Give them real work with guidance and visible accountability. Let them learn how your data, your processes, and your guardrails behave. If you want future ready teams, build learning roles, not waiting rooms.

What I tell boards

Avoid grand transformation stories. Announce two small pilots and a review rhythm. Hold leaders to both financial and learning milestones. Share results internally, especially where things broke and what you changed. Invite the sceptics and give them instruments to test assumptions. Put a date on the scaling decision and keep it.

Most of all, stop treating AI like a show. It is how the work gets done. The firms that win will be the ones that treat it that way. Quiet movement. Consistent learning. Rising standards.

If you are still tempted to wait, ask what you hope to see. A stable map. A safe majority. A signal from a peer. None of that will tell you how AI fits your data, your pipelines, or your people. You learn that by doing. Start small. Move often. Keep your hands close.

The advantage goes to the builders.

Bibliography:

Brynjolfsson, E., Li, D. and Raymond, L. 2025. Generative AI at Work. The Quarterly Journal of Economics, 140(2), pp. 889 to 936. Available at: https://academic.oup.com/qje/article/140/2/889/7990658 (Accessed: 10 October 2025).

Cohen, W.M. and Levinthal, D.A. 1990. Absorptive Capacity: A New Perspective on Learning and Innovation. Administrative Science Quarterly, 35(1), pp. 128 to 152. Available at: https://josephmahoney.web.illinois.edu/BA545_Fall%202022/Cohen%20and%20Levinthal%20%281990%29.pdf (Accessed: 10 October 2025).

Davis, F.D. 1989. Perceived Usefulness, Perceived Ease of Use, and User Acceptance of Information Technology. MIS Quarterly, 13(3), pp. 319 to 340. Available at: https://parsmodir.com/wp-content/uploads/2018/11/TAM-Davis-1989.pdf (Accessed: 10 October 2025).

March, J.G. 1991. Exploration and Exploitation in Organisational Learning. Organization Science, 2(1), pp. 71 to 87. Available at: https://pubsonline.informs.org/doi/10.1287/orsc.2.1.71 (Accessed: 10 October 2025).

OECD. 2024a. Using AI in the Workplace. Paris: OECD Publishing. Available at: https://www.oecd.org/content/dam/oecd/en/publications/reports/2024/03/using-ai-in-the-workplace_02d6890a/73d417f9-en.pdf (Accessed: 10 October 2025).

OECD. 2024c. The Impact of Artificial Intelligence on Productivity, Distribution and Growth. Paris: OECD Publishing. Available at: https://www.oecd.org/content/dam/oecd/en/publications/reports/2024/04/the-impact-of-artificial-intelligence-on-productivity-distribution-and-growth_d54e2842/8d900037-en.pdf (Accessed: 10 October 2025).

OECD. 2025. The Adoption of Artificial Intelligence in Firms. Paris: OECD Publishing, with BCG and INSEAD. Available at: https://www.oecd.org/en/publications/the-adoption-of-artificial-intelligence-in-firms_f9ef33c3-en.html (Accessed: 10 October 2025).

O’Reilly, C.A. and Tushman, M.L. 2013. Organisational Ambidexterity: Past, Present, and Future. Academy of Management Perspectives, 27(4), pp. 324 to 338. Available at: https://www.hbs.edu/ris/Publication%20Files/O'Reilly%20and%20Tushman%20AMP%20Ms%20051413_c66b0c53-5fcd-46d5-aa16-943eab6aa4a1.pdf (Accessed: 10 October 2025).

PwC. 2025a. The Fearless Future: 2025 Global AI Jobs Barometer. London: PwC. Available at: https://www.pwc.com/gx/en/issues/artificial-intelligence/job-barometer/2025/report.pdf (Accessed: 10 October 2025).

PwC. 2025b. AI linked to a fourfold increase in productivity growth and 56 percent wage premium. PwC Global Press Release, 3 June. Available at: https://www.pwc.com/gx/en/news-room/press-releases/2025/ai-linked-to-a-fourfold-increase-in-productivity-growth.html (Accessed: 10 October 2025).

Reuters. 2024. AI intensive sectors are showing a productivity surge, PwC says. 20 May. Available at: https://www.reuters.com/technology/ai-intensive-sectors-are-showing-productivity-surge-pwc-says-2024-05-20/ (Accessed: 10 October 2025).

Teece, D.J. 2007. Explicating Dynamic Capabilities. Strategic Management Journal, 28(13), pp. 1319 to 1350. Available at: https://sms.onlinelibrary.wiley.com/doi/10.1002/smj.640 (Accessed: 10 October 2025).

Teece, D.J., Pisano, G. and Shuen, A. 1997. Dynamic Capabilities and Strategic Management. Strategic Management Journal, 18(7), pp. 509 to 533. Available at: https://josephmahoney.web.illinois.edu/BA545_Fall%202022/Teece%2C%20Pisano%20and%20Shuen%20%281997%29.pdf (Accessed: 10 October 2025).

Venkatesh, V., Morris, M.G., Davis, G.B. and Davis, F.D. 2003. User Acceptance of Information Technology: Toward a Unified View. MIS Quarterly, 27(3), pp. 425 to 478. Available at: https://www.jstor.org/stable/30036540 (Accessed: 10 October 2025).

FAQs

1. What does PwC’s 2025 Global AI Jobs Barometer show? That industries most exposed to AI are seeing faster productivity growth and higher wages for roles requiring AI skills.

2. Why is “watch and wait” a risky AI strategy? Because firms lose learning time, data feedback, and internal know-how while competitors move ahead.

3. What is absorptive capacity? It is a firm’s ability to recognise, absorb, and apply new knowledge. Early experimentation builds it.

4. What is the Technology-Organisation-Environment (TOE) model? A framework showing that adoption depends on technology fit, organisational readiness, and external pressures.

5. How can small firms start with AI safely? Run small pilots with clear metrics, light governance, and visible human oversight. Learn before you scale.

6. Does AI really increase productivity? Peer-reviewed studies and OECD data show measurable gains, especially for less experienced workers when tools are well-designed.

7. What is ambidexterity in business? The ability to run efficient operations while experimenting with innovation at the same time.

8. How should boards measure AI readiness? Through pilot outcomes, data maturity, learning velocity, and governance confidence rather than vague transformation targets.

9. Are entry-level jobs at risk from AI? Many are being redesigned rather than removed. AI skills now carry wage premia, and junior roles are evolving into learning positions.

10. What is the best first step for leaders? Pick two controlled pilots. Measure learning. Share outcomes. Scale what works. Stop treating AI like theatre.