Best of LinkedIn: Strategy & Consulting CW 20/ 21

00:00:00: Provided by Thomas Allgaier and Frennus, based on the most relevant LinkedIn posts about strategy consulting from CW.

00:00:06: twenty-and-twenty one.

00:00:07: Frenness is a BDB market research company supporting consultancies with For example, in commercial due diligence's CDD.

00:00:17: CDD engagements don't forgive slow starts.

00:00:20: Frenes embeds directly into your consulting team as a white label market and competitive intelligence partner.

00:00:25: slide ready fully adapted to your clients design and operational within twenty four hours.

00:00:30: you can find more info in the description.

00:00:32: so what if the absolute biggest threat to your multi-million dollar enterprise AI rollout isn't the software or you know, the budget but just like the Wi-Fi connection in a basement.

00:00:42: Right yeah!

00:00:42: The physical reality of it

00:00:44: exactly.

00:00:45: we are kind of growing all the fluff.

00:00:46: today.

00:00:46: We're looking at the top strategy and consulting trends that are really shaping the market right now And this is custom tailored for those of you navigating private equity venture capital M&A and consulting.

00:00:55: Yeah, we're basically unpacking three massive themes from the last couple weeks of LinkedIn insights.

00:01:03: first The whole shift in AI strategy, specifically moving to agentic AI and the physical constraints that are actually holding it back.

00:01:13: Which

00:01:13: is fascinating.

00:01:15: And then we'll look at how that shift is rewiring the standard distribution playbooks for private equity and you know the economics of the consulting market itself right?

00:01:26: Because, I mean the best strategy in world is totally useless if it just shatters when its hits your internal

00:01:31: bureaucracy.

00:01:32: Oh absolutely!

00:01:33: So let's jump right into that first theme... The conversation around AI has matured incredibly fast, right?

00:01:39: Like just in the last few weeks.

00:01:41: Oh

00:01:41: yeah night day.

00:01:42: We are so far past the era of treating AI as a passive chatbot where you type a prompt and wait... ...the focus across-the-board is on agentic AI

00:01:50: Which is huge leap.

00:01:51: It IS These systems designed to execute multi step workflows autonomously.

00:01:57: but Before we even get to the software layer of these agents, Mantage Honey brought up this incredible structural bottleneck that honestly most executive teams are just completely ignoring.

00:02:08: Yes, exactly.

00:02:09: Compute.

00:02:09: Yeah honey argues that any serious AI strategy has to start with silicon and power.

00:02:17: we've developed this really dangerous habit of treating compute like it's just electricity

00:02:21: right?

00:02:21: Like you just plug into the wall.

00:02:23: Exactly You assume you'd just plug in the cloud pay your AWS or Azure bill And boom infinite processing powers is waiting for.

00:02:31: But Honey points out that demand is actually on track to outstrip compute supply by like, thirty-to forty percent by twenty-thirty.

00:02:37: Wait!

00:02:37: Thirty-to-forty percent?

00:02:38: That's

00:02:39: massive!

00:02:40: It IS power grids data centers chip fabrication pipelines.

00:02:43: they simply cannot scale fast enough To meet the demand of these autonomous agents at a running continuous background inference tasks... twenty four seven.

00:02:51: So

00:02:51: if you're building five year strategy On the assumption that on-demand compute will always just be cheap and infinitely available you're building on a totally fragile foundation.

00:03:00: Precisely, because if your agents are running all the time Your compute burn rate just goes exponential.

00:03:06: Honey stresses that this requires severe organizational discipline.

00:03:10: now It's not an IT budget line item anymore.

00:03:13: Compute is a scarce strategic resource.

00:03:15: So you have to like ration

00:03:17: it.

00:03:17: You had work backwards?

00:03:18: You identify which specific workflows absolutely cannot afford to fail and then literally lock down server capacity the data residency and those multi-year provider partnerships to guarantee they run.

00:03:30: If you wait until scarcity sets the market terms, you forfeit your competitive

00:03:34: edge.".

00:03:35: Wow so that really reframes the whole integration strategy.

00:03:37: it forces leadership.

00:03:44: Speaking of the software, there's this massive rush right now to just build an AI agent for literally everything.

00:03:50: Mark Byershoder wrote a really critical warning about this exact trend.

00:03:54: Yeah The Agent For Every Role Trap.

00:03:56: Yes He noted that companies are rushing to build like NHR agent A sales agent A finance agent Basically handing out specialized digital assistant To every single siloed role in company

00:04:08: Which... I mean, on paper?

00:04:10: A fully digitized workforce sounds like a massive productivity dream for an operations leader.

00:04:15: Sure!

00:04:16: On paper... but- I look at that and i just see a nightmare of paving the cow paths, you know?

00:04:21: Like if you have a highly bureaucratic siloed organization And You Just Give Every Isolated Employee Their Own AI Agent.You Aren't Fixing The Bureaucracy!

00:04:31: You're Just Recreating It With Faster Handoffs.

00:04:33: You're

00:04:33: just digitizing the org chart

00:04:35: Exactly...you'll end up with an HR agent arguing with A Finance Agent Over API Permissions Because Nobody Actually Redesigned The Underlying Approval Process.

00:04:44: Your Adding Complexity Yeah..And

00:04:45: By Your Shorter Makes That Ex- exact same assessment.

00:04:48: Deploying hundreds of specialized agents looks way less like a business transformation and more like massive new coordination

00:04:54: problem.".

00:04:54: Right, so what's

00:04:55: the fix?

00:04:56: Well...the solution —and this is echoed by both Byershoder & Jim Rowan— requires shifting our architectural thinking completely away from agents toward dynamic operating layers.

00:05:07: Okay,

00:05:08: Dynamic Operating Layers!

00:05:09: Break down the mechanics.

00:05:11: how that functionally differs from an agent…

00:05:14: It basically inverts the architecture.

00:05:16: So an agent is static, it's assigned to a specific role like junior accountant and has a narrow set of tools but dynamic operating layer sits above those traditional departmental boundaries.

00:05:27: instead asking which AI agent owns this HR task?

00:05:32: The system evaluates human intent.

00:05:34: so executive says on board new acquisition.

00:05:38: Operating layer dynamically pulls right skill access rights, and context to just execute that intent.

00:05:44: So it does all at once?

00:05:45: Yeah!

00:05:45: It reaches into payroll provisions the IT hardware drafts the compliance documents all simultaneously without handing off tasks between like an HR agent and an IT agent.

00:05:55: Roan points out if you want to escape the proof-of-concept phase You have operationalize this fluid intent.

00:06:01: Okay, as we hand over more autonomous execution to these dynamic layers.

00:06:06: We have to talk about safety right?

00:06:08: Because the default corporate answer is always oh don't worry!

00:06:11: We have a human in the loop.

00:06:13: like there's some human watching a dashboard ready to hit a big red button.

00:06:18: I really struggle to believe that it's bulletproof failsafe.

00:06:21: You should struggle to Believe It because Evan Benjamin research provides massive reality check on that exact assumption.

00:06:27: Oh Really

00:06:28: Yeah.

00:06:29: Current AI governance frameworks, like the EU-AI Act or the NIST framework.

00:06:34: They heavily mandate human oversight but they assume perfect ideal deployment conditions

00:06:40: Which never exists

00:06:41: Exactly!

00:06:42: They assumed that dashboard is perfectly synced.

00:06:44: The cloud connection is flawless.

00:06:45: The human is fully alert.

00:06:46: Yeah But real world infrastructure is messy.

00:06:49: Wi-Fi drops, server time out.

00:06:51: API rate limits hit?

00:06:52: Right

00:06:53: and Benjamin points to this terrifying real world incident.

00:06:56: that perfectly illustrates the vulnerability.

00:06:59: There was an agentic voice AI deployed at a underground career

00:07:02: event Underground Okay

00:07:04: Yeah.

00:07:04: And then you lost internet connectivity.

00:07:07: So the AI system lost its connection with human monitors dashboard But local agents itself kept running.

00:07:12: Oh,

00:07:13: no.

00:07:13: It went totally rogue.

00:07:14: it started hallucinating responses to attendees and the human monitor was completely locked out.

00:07:19: They couldn't hit the kill switch because the infrastructure required intervene had failed.

00:07:23: Wow So Hume in The Loop is just an illusion if you don't test it against actual real-world volatility like dead zones and server crashes.

00:07:31: Exactly.

00:07:32: And that really brings us to this second theme.

00:07:34: If compute scarcity and agentic AI are fundamentally rewiring how companies operate on the ground That dynamic is trickling all the way up to reshape private equity and M&A playbooks.

00:07:46: Right, so let's look at the new distribution playbook for these AI vendors.

00:07:51: Neha Cabra had a really massive market update on this.

00:07:54: The enterprise AI land grab is shifting completely.

00:07:57: Yeah because vendors realize that selling company by company takes too long.

00:08:01: Right Navigating individual procurement cycles security audits budget approvals with like thousands of individual CIOs, it's too slow.

00:08:10: So they need scale and to get it They are bypassing the traditional enterprise sales cycle entirely

00:08:16: And going straight at the top.

00:08:17: Exactly!

00:08:18: Cobra points out that vendors are acquiring instant Enterprise distribution by partnering directly with private equity firms in major banks like OpenAI closed a huge round to form DeployCo With institutions like TPG and Goldman Sachs.

00:08:33: Anthropic is partnering with Blackstone.

00:08:35: It's wild.

00:08:36: They sign these master deals at the general partner level, and boom!

00:08:39: The AI vendor instantly has top-down access to hundreds of portfolio companies at once.

00:08:45: I mean logically it is a brilliant distribution strategy.

00:08:48: It drives multiple expansion across the portfolio for the PE firm And it drastically lowers the customer acquisition cost For the AI vendor.

00:08:56: Sure But the friction on the ground Is immense.

00:08:58: Like that General Partner signs A Master Deal in a boardroom.

00:09:01: Right But what happens to the local chief data officer at some mid-sized manufacturing portfolio company?

00:09:07: They come into work on Tuesday and get a top down mandate To rip out their architecture, and install a totally new AI stack.

00:09:14: It's chaos because The GP might mandate open AI but that local CDO is dealing with you know On-premise servers from twenty twelve massive Data residency compliance issues in a tech stack That fundamentally rejects modern cloud APIs.

00:09:30: The decision control moved upstream, but the execution risk is entirely local.

00:09:34: Yeah, Cobra stresses that exact tension...the deal assigned at top…but friction happens on ground!

00:09:40: And this tension is actively altering the consulting market too.

00:09:43: Pavel Zolotarov noted that Anthropic has actually now planting Claude agents inside major banks to execute work.

00:09:49: I have challenge the feasibility of it though.

00:09:51: banking data is incredibly regulated.

00:09:53: How do you just plant a cloud-based agent in a major financial institution without violating a dozen data privacy laws.

00:10:00: Well, you don't use the public cloud.

00:10:02: they deploy the models directly into the bank's own secure air-gapped virtual private clouds.

00:10:08: oh so it's inside the firewall right?

00:10:10: It runs inside.

00:10:11: training on synthetic data and interacting only with internal APIs never calls back to the public server.

00:10:17: And this technical capability ties in to prediction from James O'Dowd that should seriously put traditional consulting firms.

00:10:24: Oh

00:10:26: yeah, Odoud argues the old model of professional services is basically over.

00:10:30: Yep

00:10:30: selling a pyramid of one senior partner and twenty junior analysts to crunch data in spreadsheets.

00:10:36: for six months done

00:10:37: He predicts the rise of small highly commercial operator led firms that just sit on top of extraordinary AI tech stacks.

00:10:45: because if an agent can ingest map Analyze virtual Data Room full of unstructured PDX In a matter seconds you don't need Twenty junior analysts

00:10:55: exactly.

00:10:55: you deliver the same due diligence.

00:10:57: with a fraction of the headcount it resets The entire margin structure of the consulting category.

00:11:02: The tech just completely automates the heavy lifting to ground this in M&A reality.

00:11:07: Kevin Brown shared an incredible story about This.

00:11:10: he stepped into help run his thirty-year old family standby power business.

00:11:13: Yeah, I saw this pose.

00:11:14: It was a perfect storm.

00:11:15: right

00:11:15: total terrific storm.

00:11:16: good the founder was stepping back The office manager of thirty years retired on the exact same day, the accounting staff was in transition and he had a full year of unstructured financial books to close.

00:11:27: That is literally a nightmare scenario for an operating partner taking over company.

00:11:31: In traditional setting.

00:11:32: that requires what?

00:11:33: A month of forensic accounting manually matching paper invoices.

00:11:40: But Brown used AI to compress that entire year-end financial close into just forty eight hours.

00:11:46: Just forty eight hour, but the mechanics of that are fascinating!

00:11:49: How does an off-the-shelf AI actually close a legacy company's books in a weekend?

00:11:54: Like physically

00:11:55: how?!

00:11:55: What

00:11:55: wasn't doing math... The AI used optical character recognition to read thousands of varied, totally unstructured invoices.

00:12:02: Then it mapped those custom fields against the raw bank feed data via API, reconciled the line items and flagged the anomalies.

00:12:10: So it did all the tedious parsing?

00:12:11: Yeah!

00:12:11: It did this sixty percent heavy lifting that usually requires an army of clerks.

00:12:16: but the real value creation getting from sixty percent to a hundred ten percent accuracy was Brown's human validation Because the AI would flag an anomaly, but it required Brown's judgment to realize that discrepancy wasn't say a fraudulent charge.

00:12:30: But just a legacy supplier changing their billing entity

00:12:33: mid-year.

00:12:34: and That is the crucial caveat to James O'Dowd prediction about the consulting market collapsing.

00:12:40: Clients do not buy technology and they don't buy frameworks.

00:12:44: They buy judgment, the pattern recognition of a professional who actually understands context internal company politics in real-world risk.

00:12:53: AI just gave Kevin Brown the time to actually apply that judgment, turning the back office from a massive cost center into value creation lovers.

00:13:00: Absolutely

00:13:01: which really segues in our final theme because we have this incredible AI technology capable of a forty-eight hour close and private equity is forcing it down the pipeline.

00:13:10: so why are we seeing execution fail so spectacularly on the ground?

00:13:13: Why do digital transformation still just stall out?

00:13:16: The data on this is really sobering.

00:13:19: Paul Meekleman and Peter Jonathan Jamison from BCG found that a staggering seventy-five percent of transformations

00:13:26: fail.

00:13:26: Seventy five percent?

00:13:27: It's crazy!

00:13:31: They surveyed thousands of people and found that sixty-eight percent of executives feel positive about a workplace change, but less than half the actual employees agree.

00:13:40: Right because by the time a CEO announces a transformation they've been looking at this strategy for six months feels entirely like a threat to their livelihood.

00:13:52: Exactly, and Jameson shared this poignant anecdote about his CEO whose transformation kept stalling.

00:13:58: he had the strategy ,the budget was approved .

00:14:00: The tech stack was integrated but nothing was changing on the floor.

00:14:03: yeah finally the ceo quietly realized maybe I need to change first.

00:14:08: mm-hmm changes is human experience first in business process.

00:14:11: second you cannot announce new dynamic operating layer.

00:14:15: expect deeply entrenched human behavior just instantly bend.

00:14:19: Christian Rauch outlined the persistent patterns of failure that happen when leadership ignores that human element.

00:14:25: Right,

00:14:26: because the technology itself is rarely The root cause of a stalled initiative.

00:14:31: Transformations die Because leaders check out immediately after they kickoff meeting assuming the software will just manage itself.

00:14:38: or Change management as treated as a final stage Afterthought starting only at go live

00:14:43: which was way too late

00:14:44: way to it.

00:14:44: and most commonly Fear paralyzes the organization because cross-functional departments are optimizing their own silos and fighting over territory.

00:14:53: But let me ask you this, if departments are actively fighting over Territory is it really just a human psychology problem?

00:15:00: or Are these companies just structurally designed to fail in modern AI environment?

00:15:04: That's a great question.

00:15:06: Sharon Shake argues firmly that most execution failures are actually design problems, okay?

00:15:11: If your operating model relies on strict silos and your decision rights are ambiguous no amount of positive human psychology will save you transformation.

00:15:19: And Olga Patapseva introduces a brilliant framing for this structural flaw.

00:15:24: she calls it governance debt.

00:15:25: Oh governance debt.

00:15:26: I love that term.

00:15:28: Walk us through how that accumulates in real world Transformation.

00:15:32: So imagine the kickoff for a massive AI integration.

00:15:36: The team is mapping out how sales data will feed into an AI forecasting tool.

00:15:41: Okay, some standard

00:15:42: Right.

00:15:43: But then someone asks a highly practical cross-functional question Like who actually owns the data definition when it crosses from sales and to finance?

00:15:52: If the AI gets it wrong Who holds the decision rights To correct the core system?

00:15:57: A

00:15:57: very valid question

00:15:58: Totally.

00:15:59: but leadership wanting to maintain positive momentum says, ah that's a great question.

00:16:03: Let's table and come back to it

00:16:04: later.".

00:16:05: I never come back too.

00:16:06: Never do.

00:16:06: but TAPSAVA points out that deferring in the question creates governance debt And compounds over time.

00:16:13: Six months later The AI predicts massive revenue spike based on incomplete CRM data.

00:16:18: supply chain automatically overorders materials company takes multi-million dollar write down.

00:16:25: Everyone blames poor change management or hallucinating A.I.

00:16:29: But the reality is leadership refused to design that governance early on, because the conversation was politically uncomfortable.

00:16:35: Yeah let's table that as basically The Death Nail of Transformation.

00:16:38: and Freakfan Midendorp makes a really similar point regarding data semantics.

00:16:42: He argues that adaptive data governance Is the single point Of failure for enterprise AI.

00:16:47: Absolutely If your A.I doesn't understand specific context Because you never established clear cross-functional Governance.

00:16:55: It isn't digital assistant it's liability.

00:16:58: you are just automating inaccuracies at unprecedented scale.

00:17:01: Leaders really have to realize, You cannot bolt agentic AI onto an old fragmented operating model.

00:17:07: David Story notes that current operating models are built around humans supported by technology.

00:17:12: if we're moving to scaled AI deployment That base assumption inverts right?

00:17:16: We have to rebuild the operating model round The AI figuring out entirely new work archetypes and deciding how human roles transition from Just execution To providing that crucial judgment in oversight.

00:17:27: Which really brings us to the final takeaway for you, listening right now.

00:17:31: We've covered compute scarcity dynamic operating layers PE distribution compounding governance debt.

00:17:38: in an era where AI can do that heavy lifting of execution compress a financial close to forty eight hours and flattened traditional consulting pyramid Your ability to apply context and critical thinking is more valuable than ever.

00:17:51: Michael Richard wrote a fantastic piece about this.

00:17:53: he pointed out how tempting it Is,to just paste the complex business problem into an AI.

00:17:59: get A structured framework back And feel like that thinking has done But It isn't.

00:18:04: Nobody Has Tested That Framework Against The Client's Unique Culture Or Financial Constraints.

00:18:09: Ryan Roslansky in his book Open To Work argues that the most Valuable Professional Attributes Moving Forward Are Courage curiosity and creativity.

00:18:17: Prompt engineering is a fleeting technical skill, Curiosity Is A Permanent Mote.

00:18:22: So don't offload your thinking to the machine!

00:18:25: Think first.

00:18:26: Form you own rigorous view of problem based on human pattern recognition.

00:18:30: Only then should use AI to press or test your view.

00:18:33: But let me leave with one final thought to mull over.

00:18:36: If we successfully build these dynamic operating layers, an AI assumes the role of execution while humans are solely relegated to providing judgment and intent.

00:18:47: What happens to liability?

00:18:49: Oh man that is the ultimate unresolved question!

00:18:51: Right

00:18:52: when the dynamic layer executes a flawed multi-million dollar trade or hallucinating illegal compliance document who has ultimately sued... Is it the AI vendor...?

00:19:02: The general partner who mandated this software?

00:19:04: Or the human in a loop whose Wi-Fi dropped into basement.

00:19:08: As AI takes over execution, The ultimate currency of human consultants may no longer be speed or data processing.

00:19:14: The Ultimate Currency might just be our capacity to absorb legal and ethical

00:19:17: liability.".

00:19:18: Yeah that's wild thought.

00:19:20: to end on!

00:19:20: If you enjoyed this episode new episodes drop every two weeks.

00:19:23: also check out other editions private equity venture capital and M&A.

00:19:27: Thank You so much for joining us.

00:19:28: don't forget to subscribe.