Listen on
Overview
The Big Four accounting firms are now posting nearly 7% of their job ads for AI roles — more than they post for auditors. Graduate intake is down 30% at KPMG and 22% at Deloitte, PwC is reportedly planning to cut entry-level hiring by almost a third, and the global chairman of PwC has told the BBC he cannot find the engineers he needs. The natural read is that AI is eating audit from the inside. This episode pulls that narrative apart and finds something more interesting underneath: not a profession disappearing, but a value chain being restructured in ways most of the coverage is missing.
In this episode of Eventual Consistency, Ross Katz (Principal and Data Science Lead at CorrDyn) and Jason Bradwell (founder of B2B Better, host of Pipe Dream) work through what the hiring data actually signals versus what it means operationally. Ross introduces a four-phase model of the audit value chain — origination, analysis and production, judgment and synthesis, and relationship and sign-off — and argues that AI compresses the middle while making the ends more demanding, not less. As AI produces more, someone has to verify more.
The conversation gets into why codifiability, not seniority, is the variable that determines AI exposure; why AI automates at the task level rather than the job level; why the junior role is being transformed rather than eliminated, and what that means for how firms hire and train; and why the talent problem the Big Four describe is partly a signalling problem — and a window for smaller firms that read the shift correctly.
Key Takeaways
“AI is eating audit” is the wrong read on the data
The surface numbers look damning: more AI-specialist job ads than auditor ads, graduate intake down 30% at KPMG and 22% at Deloitte, equity partners quietly demoted, PwC’s chairman telling the press he cannot hire the engineers he needs. Read together, those facts get used to signal “AI is replacing auditors.” Ross and Jason argue the operational reality is a restructuring of the audit value chain, not its collapse. The work that audit sells — assurance someone is accountable for — does not go away when the production underneath it gets cheaper.
The four-phase audit value chain shows where AI actually bites
Ross splits professional-services work into four phases: origination (winning and scoping the work), analysis and production (the procedural execution), judgment and synthesis (deciding what the analysis means), and relationship and sign-off (standing behind the conclusion). AI compresses the analysis-and-production phase hardest, because that is the most codifiable. But compression of the middle does not shrink the judgment layer — it grows it. More output means more to verify, and verification is exactly the part AI cannot be accountable for.
Codifiability, not seniority, determines exposure to AI
The displacement conversation usually fixates on the wrong variable — senior versus junior, technical versus non-technical. The better question is how much of the work can be written as a finite sequence of rules. The more codifiable a task, the more exposed it is; the more it requires reading nuance, weighing competing considerations, or putting your name on a consequential decision, the more protected it stays. That reframing cuts across the org chart rather than down it.
AI automates at the task level, not the job level
KPMG has AI integrated across 95,000 auditors; EY’s Helix scans databases of up to 100 million entries. That is real productivity — workflows get faster, more structured, more consistent, and a human reviewer can cover more ground and catch more anomalies. But a job is a bundle of tasks, and it only disappears when every task in it can be codified. As long as a role contains judgment that cannot be expressed as an if-then rule, the job persists even as its tasks get automated around it.
The junior role is being rebuilt, not removed
The instinct is to read graduate-intake cuts as a permanent contraction. The episode’s more nuanced view: the junior role is shifting away from procedural execution toward synthesis and verification — supervising and checking what AI produces rather than producing it by hand. That transformation requires a different kind of hire, different training, and a different day-one expectation. Firms that understand the distinction will hire differently from firms that read the same numbers as a simple headcount cut.
Internal AI investment is a bet on operational readiness, not frontier capability
The firms struggling with AI transformation are not blocked because the models are not good enough. They are blocked because the data, the processes, and the accountability structures were not ready when the capability arrived. Using this window to get data in order, map and standardize processes, and build repeatable AI-assisted workflows builds something that compounds regardless of which models or vendors win. The accountability chain, the evaluation framework, and the audit logging are not compliance overhead — they are the foundation of a durable data capability.
Related: AI Strategy | Technology Strategy | Episode 22: If AI Can Do the Work, What Are Clients Actually Paying For?
Full Transcript
Jason: Hi everyone, welcome back to another episode of Eventual Consistency. So there’s this data point making the rounds at the moment that sounds more dramatic than it probably is, but that gap between the headline reality is exactly where this episode lives. So for the first time, big four accounting firms are posting more job ads for AI specialists than for auditors. Nearly 7% of Deloitte, EY, KPMG, and PwC’s job listings now require AI skills. Audit roles sit in under 3%. Graduate intake is down 30% at KPMG, and equity partners are being quietly demoted. The global chairman of PwC is telling the BBC he can’t find the engineers that he needs. So the natural read here is that AI is eating audit from the inside. But in this episode, Ross Katz from CorrDyn and I spend some time pulling that narrative apart.
Ross Katz: What we find is that the more interesting story isn’t about how audit is going away, but it’s about how the value chain is being restructured, where AI compresses the analyst and production phase, but actually creates more work in the judgment and synthesis phase, not less. It’s about credence goods and why a name at the top of an audit sheet still matters in ways that AI can’t change. It’s about the difference between a one-time reset and a permanent contraction. And it’s about why your three-month AI rollout timeline probably needs revisiting, not because AI is hard, but because trust is, and trust takes time.
Jason Bradwell: So Ross, let’s get into it. Obviously, the FT has found that AI roles are now nearly 7% of big four job postings, which is over three times an increase since 2022. Does that fact surprise you?
Ross Katz: No, I can’t say that it surprises me. To me this headline and the study is perfectly crafted from a journalistic perspective because it’s the thing that when you read it as a headline you’re like, ‘Oh, this is really interesting. I can’t believe these big four audit firms are investing in so much AI capability.’ But then you look under the hood and it’s like, okay, the growth rate’s really high, which means the denominator matters, which means it was starting from a very low base. The headline is really: big four accounting firms are hiring people to do AI work — which honestly I think all companies and all professional services companies either are or should be doing. The reason is not because AI is going to eat audit, which is the thing that journalistically you want to gesture at with a number like this. It’s because internal automation is beneficial regardless of what the capabilities of these models end up doing, because the worst case scenario if you automate a lot of things internally is just that you become more efficient, and as a result you’re able to scale up even bigger than before or you’re able to do more with fewer people. Hiring people to help you do internal automation using AI is really important. But also if you see on the horizon the possibility that you might be forced into a business model change because the number of hours it takes you to conduct an audit or do an accounting screen are much lower — what’s the first thing you want to do as a company like this? You want to solidify your cost basis and you want to see how repeatable these processes can be. That’s why you hire people to do AI work: you need to get all of your data in order, all of your processes in order, you need to consider how you rework those processes to integrate AI and make that cost basis more predictable, and then only then would you consider on a selective basis changing your business model from something that’s more hourly to something that’s more fixed fee or outcome-based or subscription-based. In the meantime, you’re still going to charge by the hour, so you still have to charge for more hours in order to grow. That’s the thing about professional services firms charging by the hour: a person only has a certain number of hours in a day, so if you want to charge more hours you need more people in seats. My expectation is basically that yes, AI jobs are growing inside of these firms, but at least for the foreseeable future that doesn’t mean you’re going to hire fewer auditors until your organization is in a place where you have determined that you can continue to grow and make more money with the same number of auditors as before. That is not what I think we’re seeing, although that’s what the headline gestures at. What’s going on here is laying the groundwork — a little bit of risk management and preparation for the opportunity.
Jason Bradwell: Yeah. It also touches obviously junior hires. Part of the article was around KPMG having cut graduate intake by around 30%, Deloitte by about 22%, EY by 11%, and there were some leaked internal documents that show PwC has plans to cut entry-level hiring by almost a third over the next couple of years. What do you think? Is it a temporary recalibration or is it the beginning of something more permanent in relation to the junior workforce?
Ross Katz: There are two things I would think about here. First, these are unanswered questions as of today, so it’s important to hold them lightly. The first is: is AI better for experts? Does it make experts 10 times more powerful while everybody else stays the same, or is it better at lifting the floor for people who don’t have the level of experience and judgment that the more senior partners at an organization like this would have? If it’s better for experts, if your experts are getting so much more leverage from AI, then in the long term you’ll see these big accounting firms’ structure change. Historically — David Maister, Managing the Professional Services Firm — the thing he talked about was how to get leverage on your junior people. You hired a bunch of smart junior people and then you wanted to maximize the amount of work those juniors took on, because the more junior people a senior partner could oversee effectively to deliver high quality work to clients, ultimately the larger the profit margin for the firm. You want your lowest-priced people doing the highest amount of work. If those experts are becoming super-experts and they can now manage a bunch of AI agents to do the work rather than managing a bunch of junior people, then those junior roles are going to get squashed. But if AI is actually just lifting the floor for the same number of humans to do more work without having gotten that expertise, then what you’re going to see is just a one-time recalibration where right now they’re figuring out what that ratio looks like, reorienting their processes to consider how work looks now that this new capability exists. Once they figure out that they need juniors and that those juniors can get as much or more leverage than experts, then they’ll go back to hiring highly intelligent, AI-capable people directly out of college or whatever graduate program. My view is it’s a combination — we’re looking at more of a one-time reset. The other lens I would use is the value chain of a professional services firm like an audit firm. It starts with origination: can you find the clients? Then analysis and production: can you consume the documents they provide and produce the documents they need back? Then there’s judgment and synthesis, and then there’s a relationship/trust/sign-off phase. What AI does is compress the analysis and production portion of this value chain. It does not change the nature of origination, judgment and synthesis, or the relationship and trust/sign-off portion. If your model of the world is juniors do analysis and production, senior partners do everything else, then obviously you’d think, ‘We’re going to need fewer juniors.’ And the value is going to concentrate at the ends, so rather than hiring 10 new juniors we should hire one super partner who’s going to be able to do better origination, better judgment, and better trust and sign-off. I think in the short term that makes sense as a response to what they’re seeing on the ground. But what’s actually happening is that this oversimplified view of the value chain neglects the fact that as AI produces more in the analysis and production phase, there needs to be even more work in the judgment and synthesis phase in order to verify and account for what’s happened in analysis and production. The nature of junior roles is going to change to something that is more verification and synthesis oriented and less analysis and production oriented, and the job of seniors is going to be to impart the knowledge for judgment and synthesis to the juniors rather than expecting that they’re just going to be able to do the procedural stuff they might have been able to do right out of their graduate programs.
Jason Bradwell: This points to an interesting tension that was also brought up in the article and in the report, around the availability of talent in the market. I think it was the global chairman of PwC, Mohamed Kande, who said we’re looking to hire all of these experts to come and help drive our AI agenda but we just can’t find them. What does that tell you about where the professional services industry sits in the talent market relative to the tech market?
Ross Katz: Another thing that’s happening is they’re stagnating their junior salaries. Professional services firms have not raised their junior salaries over time, and the market for AI talent is so hot right now that you have to really want to go out and get people who have this degree of talent. It’s also been around for such a short period of time that if you’re focused on finding somebody with 10 years of generative AI integration experience — well, generative AI has not existed in its current form for four years, let alone 10 years. What you’re really looking for — and you’re maybe not in a good position to judge — is: what kind of capabilities does this person have, how have they demonstrated those capabilities previously, and how much of a bet are you willing to take on someone who’s learning really quickly and maybe doesn’t have a ton of experience navigating your internal landscape and implementing AI in ways that are going to get adopted across the organization and fundamentally standardize and make more efficient these processes? Those are just hard hires to find. In a fast-moving technology landscape, there are a lot of companies very used to sitting on the sidelines and letting the talent play out before they jump in, that are being asked to move faster and with more risk to make HR hires than they would otherwise make. Another point: this study is based on job posts put out there in the market, which is not just “I post job therefore I hire person.” There’s another role those job postings do, which is signaling to the market that you’re investing in AI. You put the job requisition out there, you see if anybody comes who really makes you excited, and then if they don’t come, you can complain loudly about how there’s not enough great AI talent out there — but really your sense of urgency about this transformation may be lower than the message you’re trying to send to your clients about how all-in you are on AI. I think it makes sense that these organizations should be trying to make work more standardized, make themselves less reliant on specialists, while also building their proprietary data moats — getting their internal data in order and getting their systems to shadow humans so they can capture the data to automate processes over time. I believe that they want to do that, but the path between wanting to do that and being willing to invest strategically in it is going to vary from organization to organization.
Jason Bradwell: Do you think this presents an opportunity? I’m thinking of compensation strategy for data teams — if you’re competing against the big four for the same limited talent pool, do you have a window to try and win people over who previously have defaulted to a big four offer, when this industry is showing stagnant growth in salaries and compensation? Is there a window for other technology businesses or smaller brands to win talent that otherwise would have just defaulted to going to a KPMG or to an EY?
Ross Katz: I think that’s probably the case. I also think it’s possible for small and middle market audit firms or other professional services firms that would previously have had difficulty taking on the scale of work that a big four audit firm can take on to win business by winning talent and by being faster to orient themselves around AI automation than other ones. But I also think that audit in particular — there are three kinds of goods in economics. You’ve got search goods where the quality is verifiable before you purchase the thing. Then you’ve got experience goods where it’s verifiable after you make the purchase. And then you’ve got credence goods where you can never fully verify it but you have to trust the company that gave it to you based on their reputation — and audit is your textbook credence good. Clients are paying for the name of the big four accounting firms to be on the top of that sheet, and that’s the competitive moat that AI isn’t really going to change. I understand why a larger audit firm would be more conservative, because AI doesn’t fundamentally change the nature of their competitive landscape the way it does in some other industries. But you could also see that as an opportunity for disruptive firms: the market for audit is changing and more companies are going to get audited that don’t need the name of PricewaterhouseCoopers or Ernst & Young at the top of that sheet. There’s the Clayton Christensen opportunity for something that from a credence perspective doesn’t carry the same degree of credence, but is more automated and more subscription-based. We’ve been expecting something like this to come about in the legal domain for a while. I can’t say whether or not that’s going to shake out, but fundamentally in this credence good ecosystem, people expect the firm to be accountable for the outcome of the audit — and that’s why the credence matters. If the client is willing to take the accountability, then the credence of the firm doesn’t really matter that much. If the regulatory regime changes where your risk of getting sued goes substantially down if you use one of these automated accounting firms, then that changes the nature of it too, but with the landscape as it stands, a company that’s already using Ernst & Young is probably not going to switch to a non-big four accounting firm because that name at the top is so important.
Jason Bradwell: No one ever got fired for hiring KPMG.
Ross Katz: Right, exactly.
Jason Bradwell: Let’s follow this thread and talk about the role of artificial intelligence in auditing and whether you’re seeing progress or whether you think it’s theater. We’ve got KPMG with their Clara platform, now with AI integrated across more than 95,000 auditors globally. You’ve got EY’s Helix GL tool, scanning databases up to 100 million entries. Are these from your point of view genuinely changing audit quality or are they layering technology onto a process that hasn’t really fundamentally changed? You’ve already touched on this a little bit, but let’s build it out.
Ross Katz: I am not in a position to speak to the quality of the AI applications that the big four accounting firms have put into place, so I just want to stipulate that before I talk about them. My expectation is that they are scaffolding existing processes and existing software applications with AI in order to speed up the human workflows, speed up the integrations with clients — for example, ingesting documents, putting them in formats that are easy to access, easy to query, easy to find what you’re looking for in those documents — and then also putting the human in a position where they can verify very quickly what the system is finding. We’re going from a fundamentally unstructured or semi-structured workflow to a workflow that has a lot of structure and much more standardization, little things that you can click that accelerate the tasks within that workstream in order to get to the final outcome. But what it’s not doing to my knowledge, and will likely not be doing for the foreseeable future, is executing every task end-to-end without human oversight at all. A lot of people are talking about this idea but I’ll just restate it: a job or an occupation is just a bundle of tasks and AI automates at the task level, but not the whole job. AI is not perfect at the task level — it may be above human quality at the task level, but if you let AI do the task and then you have a human verify it, you get typically a higher quality output across a lot of different tasks. But if there are still tasks within a job that require the human to execute the task, then the job fundamentally can’t go away; it’s just that their time allocation within that job changes. This is happening across the economy. The audit workflow is just one example of this. What we’ll see over time as companies attempt to rework their processes and understand where human capital is most valuable in delivering what the client expects is occupations restructuring into different bundles of tasks that are going to have a much higher verification component for verifying the outputs of AI, but also are going to be natural bundles of skills where a person can be expected to do the human aspects of these tasks. In doing that rebundling, the entire process, the entire workflow, the entire organization becomes more efficient. I think David Autor is the economist who’s also quoted in the article who researches this framework. It is great for all of these accounting firms to be trying to envision their workflows as a series of tasks, documenting those tasks in the form of software, and integrating AI on top of that software in a way that makes sense to accelerate and improve the quality and standardize the approach to audit. However, the judgment required in those verification tasks, the judgment required in evaluating the final output, is still going to be a critical component. The question is in audit or consulting or law, how much of a given role is codified — expressible as one or a very small sequence of if-then rules for a given task — versus how much of it is judgment-laden and you really need to rely on a human to detect the nuance in what they’re looking at. The more codified it is, the more exposed the tasks are to being automated, and the more judgment-laden they are, the more protected the tasks and the jobs containing those tasks will be with the emergence of AI.
Jason Bradwell: You’ve half-answered what was going to be my next question, which was around accountability — because the liability for errors manifested through AI tools, who do they fall on? Do they fall on the developers? The vendors? The deploying businesses, the professionals, end users? I don’t think there’s a clear-cut answer at least legally speaking for that. This is really relevant for audiences of this podcast and customers of CorrDyn because every data team is out there deploying these AI-assisted analysis tools and facing that same accountability gap, right, because if the model is flagging something incorrectly or they miss something that should have been caught, who’s signing that letter?
Ross Katz: You put the nail on the head and this is the relationship/trust/sign-off leg of the value chain. Ultimately, the work has to have somebody’s name on it. If the work has Ross.AI’s name on it, that’s very different than the work has my name on it. As people, we have built up architectures of trust and accountability to each other — enforceable accountability — over the course of thousands of years. A lot of the foundations upon which our society is based is the creation of distributed trust mechanisms and ways of developing trust with strangers that allow us to navigate our lives in safety without knowing each and every person of the 8 billion on planet Earth that we’re running into. Injecting AI in the mix and saying, ‘Well, now AI is accountable for the output,’ is not something I — or I don’t believe many business leaders — are going to be comfortable with for the foreseeable future, unless there is an entity standing behind it that is willing to be sued out of existence in the event that the catastrophic thing happens on the back of the output of that AI. This is another aspect of what we’re navigating across domains: how can we restructure trust-based businesses, judgment-based businesses, expertise-based businesses to allow for a distributed trust mechanism that allows me to have the level of confidence in this AI output that I used to have in Ross or CorrDyn. I think we’re pretty far away from a world that looks like that, and I don’t expect that to change drastically anytime soon unless there’s regulatory intervention. From where I sit, the regulatory intervention doesn’t seem excited about removing the barriers to entry for AI to do things that humans are being tasked with doing right now. I would argue the other direction is where it’s going — that the protection of the human trust and accountability mechanisms at the expense of AI is really where things stand. We’ve already hammered out how that works with the software-as-a-service ecosystem. If it’s going to change on the back of AI, I think we’re talking about a decades-long transformation, not something that’s going to happen in the next two to three years. That’s at least my viewpoint about where we stand from a trust perspective. It’s going to take a long time for those kinds of social structures to transform.
Jason Bradwell: The question is can we afford to wait that long for those types of social structures to transform? What is the impact of waiting a decade plus for these things to happen when everything is moving so fast and it’s totally transforming the industries that we all work in? From your perspective, when it comes to building out that accountability chain in data systems with your clients, are you thinking about that as best practice or is this now a requirement?
Ross Katz: There are different levels of trust. There’s trust for the entire project — the work that we’ve produced, you trust us to produce that project. But then there’s trust in the system that we developed for you. From an overall project perspective, we show you who we are every day when we’re working with you — that’s more of the human-based trust architecture I was talking about previously. From a systems perspective, the way that we develop trust is by demonstrating reliability in our systems. The way of demonstrating reliability in a software system and an AI system in a data system has been established over the last 30, 50 years: with lots of testing and evaluation, with consistent monitoring and audit logging, with reviews with all of the stakeholders to make sure that they understand how the system works and why it works the way it works, and can validate that it fits within the compliance regimes they’re subject to and the security controls they have in place. All of that combination of systems reinforcement — allowing for transparency, alerting, and teaching each other how the systems work and how they fit into the overall architecture of the business — gives everyone the trust and confidence they need to continue investing either with us as an external vendor or in the system and architecture they’ve created for themselves. None of that is fundamentally changed by AI in my opinion. AI just lays on top of it, and the way that you demonstrate trust is similarly through reliability, and the way that you demonstrate reliability is with a really strong evaluation set, a really strong benchmark data set that shows you what the system does in different circumstances and allows you to know when you make a change what things are getting worse and what things are getting better. And all of that work of developing the evaluation data set — and Microsoft and Satya Nadella have been hammering this home with their recent release of this idea that we’re all going to fine-tune our models on our internal data and on our data moat — that evaluation data set that you develop over time as an organization to say, ‘This is the expected behavior and we know when our systems are abiding by that expected behavior and not abiding by that expected behavior,’ is something that takes a lot of time and effort and energy to develop and it doesn’t happen overnight. It happens over the course of a long period of experience. Back to the big four audit firms: this is fundamentally what they should be targeting with their AI hires — laying that foundation and preparing for a world where more and more tasks are being done by AI, but not automating it and then pretending nobody has to look at it anymore. Automating it and then being able to evaluate the output, test it, draw human attention to where we think the AI is doing something wrong so that we can get more leverage on our humans — they’re attending to the things that have to be attended to and the rest AI can take over. But that’s a process that’s going to take years if not more than 10 years, and what you want is a foundation for growing into that future for your data moat, not to be slowly but surely left behind by other organizations that are building those AI capabilities internally.
Jason Bradwell: For me it talks to this idea of advising on disruption whilst you yourself are being disrupted — the big four specifically, and you see this in marketing circles as well. They’re grappling with AI on two fronts: trying to implement it internally and figure out how to reorient their business around this new technology, while simultaneously helping their clients do the same. They’re effectively treating themselves as client zero for this AI transition. How much of the advice they’re giving is ahead of where they themselves actually are? For people listening to this, that makes me think if firms with these billion-dollar AI transformation budgets are still figuring it out, then your three-month AI rollout timeline probably needs a bit of revisiting. What do you think? Are they running before they can walk?
Ross Katz: It depends on what task they’re trying to automate — that’s really fundamentally it. If you can figure out where the highest-leverage tasks are in your organization and you already have the data capture mechanisms in place to understand the context for those tasks, then a three-month rollout for that task seems achievable. But if what you’re pitching to yourself or your investors or the public is we’re going to reorient our entire business processes around AI within the next three months — including all of the tasks and all of our human capital and our business model simultaneously — you’re not going to do that. What we say to our clients consistently is: let’s focus on the highest value use cases up front, let’s demonstrate value on those use cases, and then let’s build confidence and capabilities over time. We’re not waiting on a transformation miracle that’s never going to come — we are taking a stairstep approach and not skipping any steps on the road to the transformation that you want. The transformation is stepping up those 20 steps in a row and doing it in a way that is thinking intentionally about the future so that in two years you’re where you want to be, but along the way you’re getting the value and building the confidence and learning the capabilities that you need internally to really ingest the change you’re trying to undergo. That is our general viewpoint on digital transformation generally and AI transformation specifically. I don’t think the nature of AI projects is fundamentally different than the nature of machine learning projects or data warehousing or data pipeline projects or decision augmentation projects. All of them have this kind of structure where you can develop the foundational capabilities you need as you are demonstrating the highest value use cases you have.
Jason Bradwell: As we look at wrapping up this segment of Eventual Consistency, I want your hot take — what does a big four firm actually look like three years from now? I appreciate you don’t have a crystal ball and I appreciate you reserve the right for this to change not just in the next year but in the next three weeks. We’re seeing some of these firms moving to more of an obelisk structure, others talking about going to a box model where they’re trying to match senior and junior headcount more closely. What are your thoughts? Headcount, ratio of human to AI systems, what a partner is within these firms. Three years from now, what does big four look like?
Ross Katz: This is really funny because I’m going to give you the coldest ice cold take. I’m going to give you the arctic take.
Jason Bradwell: I love it.
Ross Katz: My take is that these firms are going to look roughly similar to the way they do now with maybe a smaller base and a wider point at the top. They’re not going to look that different because the professional services model has survived for hundreds of years. This is not something that emerged yesterday, although it’s been honed over the last call it 100 years. There have been waves of technological transformation previously that have caused people to believe that job roles like audit are going to go away or be fundamentally transformed. If you think about the auditor of 100 years ago, we already moved from a world where mental math was the most important thing to something much closer to law — procedural, legal, compliance-oriented. You understand the way that financial structures work and the language of finance. Mental math still helps with intuition, and AI’s going to help with procedural, legal, and compliance knowledge in the same way that computers and calculators helped with mental math. But fundamentally, the integration across all of these different capabilities is the reason why you want a human there in the first place, in addition to the accountability structure that needs to exist for audit to be an industry in the first place. My viewpoint is this is changing the nature of jobs across industries in professional services and audit in particular, but the rebundling of tasks is not going to make the structure of the organization look that different as long as audit remains a credence good. If audit is no longer a credence good, then you’re going to see companies that are 90% software, 10% auditors — all different sorts of organizational structures. But my bet is three years from now the big four are still doing what they’re doing because they’ve been doing it for a long time.
Jason Bradwell: It’s time for my favorite part of Eventual Consistency, which is learning about what you have been watching since our last episode, Ross, in the world of AI, data, and technology. So, what’s caught your eye?
Ross Katz: I’m going to switch gears because what I’ve been watching is the NBA playoffs, and I love the NBA. I’ve always loved basketball. I’ve never been as good at basketball as I’ve wanted to be, but being a data science person who also loves basketball, it’s fun to watch a sport that is so analytically rich. There are so many metaphors from the NBA that are relevant to everything that we talk about in the AI and data landscape. You see five players on the floor who have these different combinations of capabilities and skills, and you can tell when a group of people on the court has a sense of what each other are going to do and can interact seamlessly. You can also tell when the competitive team recognizes a weak link on the floor from a defensive perspective — the playoffs always reveal this. There’s always somebody who’s put in the crucible where they’re being asked to defend the best player because they’re considered to be the worst defender. There’s always a situation where we’re discovering whether an emerging star like Victor Wembanyama can step up in the biggest moments and carry his team from an offensive perspective, and how the spatial nature of the floor changes when you have somebody who’s nearly eight feet tall who can protect the rim the way Wembanyama does. I don’t have a perfect nugget to share in terms of how this influences the way that you should be doing business today, but as humans in organizations we are constantly looking at each other and trying to understand: what are your capabilities today and how do your capabilities complement mine? Are we playing the same position and there’s too much overlap in our capabilities for the whole to be greater than the sum of its parts, or are we complementing each other so that we’re able to cover more territory as partners in order to do more as a team than we could do as individuals? Do I need you as my colleague to be able to hit that shot for us to win as an organization? And how can I put you in a position to succeed based on the strengths you have, and how can I give you the confidence to hit the shots that you’re good at by the way that we structure our offense or our defense? I love thinking about these things and I have to be honest about what I’m watching, which is the NBA playoffs — I hope you have a chance to watch as well. It’s something I really enjoy.
Jason Bradwell: Incredibly well put. If you had to call the championship, who’s taking it home this year?
Ross Katz: Ah, it’s impossible. The Knicks are peaking at the right time, the Spurs have just been through a grueling seven-game series with the Thunder, and so if I were giving the edge, I would give it to the Knicks. Knicks in six — they’ll be partying in Madison Square Garden — but honestly if Victor Wembanyama finds his footing and is able to stay healthy and be the star that the Spurs need him to be, then the Spurs could very well win the next four games in a row. It’s really just a question of whether the peaking team with more experience wins out or this emerging talent boom in San Antonio is able to win out. I’ll pick the Knicks, but I know about as much as anybody else if not less.
Jason: That’s a wrap on this one. If there’s a frame that I take from this conversation, it’s Ross’s value chain breakdown: origination, analysis and production, judgment and synthesis, relationship and sign-off. AI is compressing the middle but is making the ends more demanding, not less. The question for any data leader right now isn’t whether to automate the analysis layer, it’s whether you’ve got the judgment layer ready to absorb what comes out. The other thing that stuck with me is the credence good point. Audit isn’t just a service, it’s a trust mechanism that society has built up over a very long time. AI doesn’t change the nature of that oversight and probably shouldn’t, but the stairstep model that Ross described for AI transformation—demonstrate value on the highest leverage tasks first and build confidence then compound—that’s the approach whether you’re a big four firm or a five-person data team. Next episode is coming out in two weeks and in the meantime, if Eventual Consistency is useful to you, share it with someone navigating the same kind of questions that you’re looking to answer. And if you want to get in touch with the CorrDyn team, you can find them at corrdyn.com. That’s C-O-R-R-D-Y-N.com. We’ll see you on the next one.
