PwC US to train 85,000+ staff to challenge AI by June 2027
PwC US is rolling out a critical thinking and judgment certification for more than 85,000 partners and staff and estimates more than $100 million in total investment, including university funding. Simulations have staff challenge AI-generated work like internal controls assessments. For any organization handing analysis to AI, the human reviewer is now the part that needs training.
This story was produced through MarketScale. See how Business Services teams put it to work with Executive Thought Leadership.
Key takeaways
In the KPMG and University of Texas at Austin study, 24.1% of early-career professionals did worse working with an AI agent than the agent did alone, which puts the human reviewer at the top of the training list.
A sharper question for any team where AI drafts and a junior person signs off: what has that reviewer actually practiced challenging, and does the workflow require the challenge or only a signature?
Free workspace
Turn your Business Services expertise into content.
Record interviews, organize footage, and write with AI on a free trial of the MarketScale platform for qualifying companies. No demo required, no credit card.
The exercise opens with an AI tool's assessment of a company's internal controls. The participant has to find the deficiencies in those controls, then push back on what the tool concluded. Yolanda Seals-Coffield, chief people and inclusion officer at PwC US, described that simulation to Bloomberg Tax as one example from a new critical thinking certification the firm will run across more than 85,000 partners and staff.
The rollout begins in December 2026 across PwC's US and Mexico offices and its Acceleration Centers, the offshore service offices that Bloomberg Tax noted sit in countries such as India. Completion is targeted by June 2027. The firm estimates more than $100 million in total investment, including program development costs and funding for universities.
PwC plans to start rolling out the certification in December 2026 and aims to finish by June 2027. The goal is to give its people a repeatable way to evaluate information and exercise judgment, particularly when AI is involved. PwC's reasoning is that professionals have always built judgment by doing the research and analysis themselves, and AI now does more of those foundational tasks. If the practice no longer happens on the job, someone has to put it on the calendar.
That makes the program useful well beyond accounting. For a controller, internal audit head or operations leader whose analysts now review AI drafts more often than they write first drafts, the program suggests what putting judgment practice on the calendar could look like.
Five stages from diagnostic to daily routine
PwC built the course on its own Critical Thinking and Judgment Framework, which it says helps people establish context, challenge information and assumptions, and communicate recommendations with clear accountability. The firm lays out the sequence this way:
- A diagnostic scenario that maps each participant's thinking preferences, strengths and growth areas
- Instruction in the framework, applied to practical situations
- Masterclass-style videos featuring PwC leaders and outside experts on the skills behind sound judgment
- Personalized simulations tied to the participant's own work, such as the internal controls review
- On-the-job toolkits, team collaboration scenarios and routines meant to carry the habits into client work
The diagnostic is the design choice worth studying. Seals-Coffield told Bloomberg Tax the goal is for people to discover what kind of critical thinker they are by default. Her example was the analyst type: someone who checks every input for reliability but dives into the details before deciding how much checking the work actually needs.
That is a calibration problem, and it suggests PwC is training people to size a review before starting it. The habit would matter more as AI produces drafts faster than anyone can check them line by line. A reviewer who examines everything at the same depth runs out of hours exactly where a careful look was needed.
Seals-Coffield also placed the certification inside PwC's wider learning on responsible AI use, describing the pieces as connected work rather than separate courses. For L&D leaders building their own AI training, that framing matters: the judgment module sits next to tool training, and doesn't replace it.
The apprenticeship AI is absorbing
PwC is not working on this alone. Bloomberg Tax reported that its Big Four rivals EY, Deloitte and KPMG are building AI, including autonomous tools that handle routine tasks, into their workflows while reworking talent strategies around both AI and traditional business skills. Claudine Cassar, founder and editor of Big4News, put the audit stakes plainly: routine procedures done under supervision were how junior staff learned to spot inconsistencies, question explanations and weigh evidence.
KPMG US reached a similar answer earlier this year, Big4News reported. It redesigned its audit internship around simulations and exercises that have interns investigate anomalies and make decisions with incomplete information. Two firms, one approach: when real engagements stop supplying the practice, simulated ones fill the gap.
Research by KPMG and the University of Texas at Austin, as summarized by Big4News, shows why the reps matter. The study followed 523 early-career professionals working with AI agents and compared their results with the AI working alone.
In the KPMG and University of Texas at Austin study, 24.1% of early-career professionals did worse working with an AI agent than the agent did alone, which puts the human reviewer at the top of the training list. Big4News described two ways people fell short. Some accepted the AI's conclusions with little scrutiny. Others did challenge the tool, but missed its most relevant weaknesses.
The second failure is the subtler one, and it lines up with the calibration problem in PwC's diagnostic. Skepticism aimed at the wrong details eats time and catches nothing. The study covers early-career professionals only, so it says less about how experienced reviewers perform, but early-career staff are exactly who AI-assisted workflows tend to put in the review seat.
Simulations versus years on engagements
The strongest objection is Cassar's: it is still an open question whether simulations and a formal certificate can build the depth of judgment earlier generations gained over years of supervised audit work. PwC itself says the framework is still being validated with academic input, including professors affiliated with the Wharton School and Harvard University. Until that work reports results, the certificate records that staff completed the training, and the evidence that it changes review quality is still to come.
Wayne Banks, a chartered accountant and former global CFO, has warned in Big4News about automation deference: professional skepticism wearing down as people grow used to polished AI work that looks credible. His remedy is procedural. Banks argues for documented verification steps, clear accountability and controls that require employees to test AI conclusions before they approve them.
Training and process controls attack the same problem from opposite ends. A certified reviewer inside a workflow that asks only for a sign-off is still working in a sign-off workflow. For organizations rolling AI out to junior analysts, the practical lesson from pairing PwC's program with Banks' argument is to do both: train the challenge, and make the process demand it.
From the firm to the freshman classroom
The university piece starts in early 2027, when PwC plans to offer its framework to any university that wants to teach it. Through its Student Human Skills Initiative, an extension of the firm's Human Skills Project, PwC also plans to fund selected universities so professors can build the framework into introductory courses, with access to PwC practitioner insights and resources.
That pushes the same vocabulary into the hiring pipeline. If adoption spreads, employers could start seeing graduates who arrive with a structured method for questioning AI output, which would change what a first-year training plan has to cover.
Before AI drafts anything a junior employee signs off on, ask what that person has practiced challenging.
The dates are set: rollout in December, the university release in early 2027, completion in June 2027. The signal worth watching sits between them, in what PwC's academic validation actually measures. If it ties the framework to better review outcomes, the kind of gap the KPMG and UT Austin data exposed, other employers will have a benchmark they can copy, and their AI governance teams will have something more concrete than a policy memo.
Sources
Your experts belong here
Every story in MarketScale Business Services starts with a company putting its consultants, practice leads, and account teams on the record. Buyers are already reading this topic. The only question is whose experts they find.
Clients hire the firm whose thinking they have already read, which means fewer cold conversations for your partners.
About the author
The MarketScale Newsroom reports on the companies, technologies, and trends shaping 16 B2B industries. It turns primary sources and expert commentary into clear, useful coverage for the people doing the work.