Uphoff on Media Podcast

Every new dollar of revenue, another card on the wall.This fall I've heard the same story three times, from three very different businesses: a B2B agency, a PE-backed marketing services firm, and an AI consultancy. Each has rebuilt real work around agentic workflows, and the gains they describe don't round to a percentage. They round to a multiple. Ten times. Sometimes more.The productivity is real. The payoff isn't automatic. A multiple of output is not a multiple of anything you get paid for.If you bill by the hour and your team now does in one hour what used to take ten, you just gave your clients a 90% price cut. You worked hard for it. They didn't ask for it. And the invoice looks exactly the same as it did last year, only smaller.The number everyone quotes isn't the number that paysTwo numbers are getting blurred together in almost every conversation about agentic AI in services businesses. Pull them apart and the picture gets much clearer.The Output Multiple: how much more a person produces on a given workflow now that agents do most of the execution. This is the multiple everyone quotes.The Scale Multiple: how much more revenue each person brings in: revenue per employee now, divided by revenue per employee the year before agents. $150,000 a head to $300,000 is a Scale Multiple of 2. This is the number that shows up in the P&L, and eventually in the valuation.The gap between the two is what you're handing your clients for free.Last week, in Before You Flatten the Org Chart, I argued that revenue per employee is the number to measure before you restructure around AI. This is the next question: how did the number go up? Two businesses can post the same improvement and be building two different companies.Revenue per employee is the right yardstick because it's the one software companies use to grade themselves, and services businesses are about to be graded the same way. The median private SaaS company now does about $175,000 of ARR per employee, according to Benchmarkit's 2026 data. The largest agency holding company runs well below that. WPP booked £11.4 billion of revenue less pass-through costs in 2024 with an average of 111,281 people, per its 2024 results: about £102,000, or roughly $130,000, per person by my math.Margins tell the same story. The average digital agency earned a 13% net margin last year, according to Promethean Research. Its size data is the whole problem in one line: agencies under ten people run 19% margins, and agencies with more than 50 people run 8%. In the agency business, getting bigger makes the margin worse. That's been the history of the services model, where every new dollar of revenue shows up with a new salary attached.The holding companies are raising revenue per person the old-fashioned wayThe largest agencies already know the labor model is breaking, and they're responding by taking people out.WPP went from 108,044 people at the end of 2024 to 97,388 at the end of June. In March its CFO, Joanne Wilson, told analysts the company intends to "move away from time and materials and decouple revenue from headcount." Omnicom, which bought rival holding company Interpublic (IPG), expects to end the year with roughly 105,000 people, down from about 120,000. The cuts are the core of a $1.5 billion annual savings target from the merger, $1 billion of it from labor. Its CFO added that the cuts "really have not impacted client-facing people very much at all." One holding company CEO, quoted anonymously in Forrester's Predictions 2026, put the plan in a sentence: "By 2028, we'll double profits and halve the people."That raises revenue per employee. It does it by shrinking the denominator, not by growing what each person produces. I've led companies through structural disruption more than once, and the pattern holds every time: cost cuts buy time, and the only thing that buys a future is a new answer to "what exactly are customers paying us for?"Publicis is the useful exception. It grew 5.0% organically in the third quarter, after 4.8% in the second, while its peers cut. That's a data business winning share, built on Epsilon and now LiveRamp, not a smaller team doing more. Scale still works for the company that owns the data layer. It stops working for everyone billing hours on top of it.Outside services, the gap is starker. On October 21st, Meta reported third-quarter revenue up 26% from a year earlier, with headcount down 15%, to 66,775. That’s revenue per employee up roughly half in a single year, from more revenue and fewer people at the same time. Meta isn’t an agency. It’s the comparison your board will make anyway.The market doesn't pay for margin by subtractionCutting your way to a better ratio has a second problem: buyers and investors can tell the difference.Kroll's Summer 2026 software sector update makes the point with public software companies. Median EBITDA margins barely move across growth rates: 28.9% for companies that are shrinking, 25.1% for companies growing faster than 20%. The multiples move enormously, from 2.4 times next year's revenue for the shrinkers to 7.2 times for the fastest growers. The market is paying for growth, not margin.Services businesses won't be graded more kindly. A firm that gets to 40% EBITDA by cutting people is more profitable and no more valuable. A firm that gets there by growing revenue per person has built something a buyer will pay a multiple for.Where your business sits on the mapHere is an Uphoff Advisory Framework you can download and use in your business. Plot your Output Multiple against your Scale Multiple and you land in one of four places.Download The Scale Multiple Map →View the full Uphoff Advisory Framework library →What each square tells youBusiness as usual. Neither number has moved. The agents are a pilot, a tool, or a line in the credentials deck.Giving it away. Output is up sharply and revenue per person hasn't followed. This is where most of the productivity stories live today: real gains, priced by the hour, passed straight through to the client.Margin by subtraction. Revenue per person is up because headcount went down, not because anyone produces more. This is the holding company playbook. It works until there's nobody left to cut.Team Scale. Both numbers are up, and the gain shows up in the P&L. Small senior teams, agents doing the execution, revenue per employee climbing past the software median.Team Scale is the next rung up from the model I described in Solo Scale: The New Business Model AI Just Made Possible. Solo Scale is one expert with agents, running from $250,000 to $5 million in revenue. Team Scale is a small senior team doing the same thing at roughly $400,000 of revenue per employee: $2 million to $3 million with five to eight people, up to $25 million to $30 million with 60 to 75. At the holding company rate, booking that same $30 million takes more than 200 people.Three ways firms are getting there: build it, buy it, or sell itThe three conversations I opened with map to three different routes into that top-right box. I've changed identifying details; the math is theirs.Build it. The PE-backed marketing services firm built a series of agentic workflows. Before, it hired three people, on average, for every incremental $750,000 of revenue. Since then it has added about $5 million of revenue, from new clients and from expansion inside existing ones, with headcount flat. Under the old ratio that $5 million would have meant 20 new hires. It made none. Its Scale Multiple will take a few quarters to read. The twenty salaries it never added show up in the P&L today.Buy it. The agency is using an acquisition to add AI capability, with a target of taking EBITDA from 20% to 40%. That plan can go two ways, and the math shows why the route matters more than the number. Cut about a quarter of the cost base with revenue flat and 20% becomes 40%. That's margin by subtraction. Or hold the team roughly where it is, grow revenue by about 60%, and keep cost growth under a fifth, and 20% also becomes 40%, with revenue per person landing right around the Team Scale line. Same margin. Two very different businesses, and per Kroll, two very different multiples. Sell it. The AI consultancy is building an agentic workflow app for a multi-location consumer services business with thousands of customers and a mix of full-time and part-time staff that makes scheduling and billing a manual mess. The new system takes those processes from weeks to hours. That's an Output Multiple well past ten. The question worth asking is who keeps it. The client keeps every hour it saves. The consultancy keeps only what it priced for. Bill that build as a project and the consultancy is giving it away on its client's behalf.The Solo Scale version of "build" is my own. I built Uphoff Advisory's client tools, the Uphoff Value Index, the Sales Talent Matrix and the Marketing Talent Matrix, along with the website and the system I run engagements on, at roughly a ten-to-one ratio of output to time against what the same work would have taken before agents. I did about 90% of it myself. For the last 10%, authentication, secure data storage and wiring it all into one client portal, I brought in a developer. AI didn't replace him. It made me a much better client, because I handed him a working spec instead of a rough idea.The judgment didn't get a discountScaling yourself and automating yourself are not the same thing.Automating hands execution to something that isn't you. Scaling extends judgment that is still entirely yours to more clients, more often. Every story above is a scaling story. The research, the drafts, the versioning, the scheduling, the reporting all got cheaper. The judgment about what the work should be didn't get cheaper at all.My Pipeline-vs-Forecast tool weights a strong verbal awaiting signature at 90%, a confirmed proposal at 50%, a contact made at 10%, and an assigned account nobody has touched at zero. No agent came up with those

B2B buyers still build shortlists. They just don’t build them alone anymore.Google “Tony Uphoff” and you get about 86,000 results. Thirty-five years of bylines, press releases, bios and CEO announcements make me discoverable, and by extension the two businesses I started in February. As long as someone is searching for my name.AI is a different story. This month I asked AI the questions my audience and prospective clients ask: who to hire, who to call, what to read. 19 prompts across ChatGPT, Gemini, Perplexity and Google AI Overviews, tracked in Ahrefs Brand Radar from October 10 to 17, produced 59 answers about my two businesses and me:• Uphoff on Media showed up in 0 of 12 category answers.• Uphoff Advisory showed up in 1 of 43.• “Who is Tony Uphoff?” Every engine that answered knew the career. The models know who I am. They almost never recommend what I’ve built. That gap is where B2B buying has gone, and most companies aren’t measuring it.Search didn’t die. It changed jobs It used to be where buyers found you. Now it’s where they check what the machine recommended. In a March 2026 G2 survey of 1,076 B2B software buyers, 51% said they now start their research in an AI chatbot more often than in Google, up from 29% a year earlier. In a Semrush survey of US B2B professionals fielded in March and April 2026, the most common workflow among those who use AI at work (41%) was to start with AI, then verify its answers through traditional search.I’ve been brought in to transform two businesses built on the last era. At Business.com and ThomasNet.com, helping B2B buyers find suppliers was the product. The function hasn’t changed: connect the buyer and the seller. The form has. The machine does the finding now, and the shortlist comes back already built.That’s the new sequence. The machine drafts the shortlist. The human validates it: checks the names, reads the reviews, asks a colleague, asks the model a follow-up. In the same G2 survey, 69% said a chatbot led them to choose a different vendor than they had planned, and one in three bought from a vendor they had never heard of before. The buyer still makes the call, but from the machine’s list. If you’re not on it, you’re counting on the buyer to go looking, and those numbers show how often the machine’s list wins.For two decades the scoreboard was discoverability. SEO taught every marketing team to rank, earn the click and get found. Most companies carried that scoreboard straight into AI. They ask a model about themselves by name, get an accurate answer and conclude they’re covered. That’s discoverability in a new channel. The scoreboard that matters now is recommendation: when a buyer asks for your category, does the model put you on the shortlist?For most companies, the pattern in my results is a false positive. A well-known founder, CEO or legacy brand makes the business feel discoverable. Ask the model about the leader and the answer sounds right. Ask it what a buyer would ask, and the company rarely makes the list. Measure only discoverability and you’ll never see the gap.Answer engines describe the company you used to beAll three engines that described me still call me CEO of Pipeline360. I led its carve-out and built it as a standalone business, and since February I’ve served as a board advisor. Only one of the three mentioned Uphoff Advisory, the firm I run now. The sources included org-chart aggregators, executive databases and old press releases.Answer engines default to what the web has said about your company most often, and most often usually means longest ago. If you’ve changed leadership, renamed a product, repositioned or entered a new market in the past two years, assume the models are describing your old company to your new buyers.The stale record costs you twice. The machine uses it to decide whether to recommend you. The human uses it to check the machine’s work. Fixing that is the job of answer engine optimization (AEO), and most companies haven’t started. A Corporate Ink study reported by eMarketer in May found only 29% of marketers actively track and measure AI visibility.The idea travels. The name stays behindAsk the same engines “What is Demand-as-a-Service?” and all three define it accurately, each one citing Pipeline360’s own pages. I helped create it as the foundational strategy when we built Pipeline360. None of the answers mention me, and for Pipeline360 that’s exactly right: the company owns the product and the category language.For any business that runs on a point of view, it’s a warning. The models credit whoever publishes the clearest explanation, not whoever had the idea first.AI shortlists are built from other people’s shortlistsThe one time Uphoff Advisory made a list, the question was narrow: “Who advises founders of B2B information and data businesses?” ChatGPT listed it second of four. The broad questions, the best strategic advisors for B2B media and who to call about a turnaround, returned nothing.The citations explain why. When the models answered category questions, they rarely cited the firms or newsletters they named. They cited roundups: “best AI newsletters” and “best B2B marketing podcasts” lists from Zapier, Content Allies, Breaking B2B, The B2B Playbook and Dupple. The AI shortlist is assembled from lists other people already published.A May 2026 study by Maryanna Franco and João da Silva, run on Friction AI’s platform, found the same mechanism at scale. They sent 14,140 queries about twelve activewear brands to five AI engines. Nike and New Balance carry the identical Google Knowledge Graph description: footwear company. Nike appeared in about 71% of athleisure recommendation answers. New Balance almost never did. In third-party content, Nike shows up alongside lululemon roughly 50 times more often than New Balance does. Equally discoverable. Only one recommended.In The Buyer Left the Building, I argued the shortlist gets built before a human enters the conversation. In The CPM Collapse, I called 2026 the agentic inflection for B2B research and vendor discovery. This is what both look like from the vendor’s side.Change the scoreboard firstIf you’re building, running, funding, or starting a B2B information, marketing or technology business, here’s the sequence. I’m running it on my own businesses now. Move 1 changes what you measure. Move 2 serves both sides, because the machine and the human read the same record. Moves 3 through 5 earn the recommendation.1. Change the scoreboard. Rankings, traffic and AI mentions tell you whether you’re discoverable. Add the number that tells you whether you’re recommended: how often you make the list when a buyer asks for your category. Report it first. Write 10 to 15 prompts the way your buyer asks them: the category, the problem, “who should I hire for.” Add two or three that ask about you by name. Run them across ChatGPT, Gemini, Perplexity and Google AI Overviews and track two numbers: how often you make the list, and whether the description is right. A tool like Brand Radar does this weekly. A spreadsheet and an hour a month works too.2. Fix the record before you chase the shortlist. Read the sources the models cite when they describe you. Update the pages you control: your site, LinkedIn, leadership bios, press page. Request corrections on the ones you don’t: org-chart aggregators, executive databases, directory listings. After any leadership change, rebrand or new pricing model, assume the models haven’t heard. This is also the record your buyer reads when they check the machine’s work.3. Publish the canonical version of your ideas. If you coin a term, a model or a method, publish the definitive explanation under your brand, with your name on it, before anyone else writes the explainer. Then make sure the people who write about it link back to yours.4. Win the narrow question first. Pick the most specific way a buyer describes the problem you solve and own that phrase before you chase the broad category. “Advisor for founders of B2B data businesses” made the list. “Strategic advisor for B2B media” didn’t. Narrow first, then wide.5. Get on the lists the models read. Find the 10 to 20 roundup and comparison pages the models cite in your category. The citations are visible in the answers. Earn a place on them, next to the names already being shortlisted. That’s a PR and analyst relations job. Give it an owner and a target, like any other pipeline source.If you run a B2B media or information business, read move 5 from the other side. The lists the models read are published by someone. In the discoverability era you sold traffic. In this one, you could own the inputs to the shortlist.If you’re funding these businesses, add one diligence question: “When a buyer asks AI for your category, how often are you on the list?” A company that’s missing there has a pipeline problem that hasn’t reached the forecast yet. If the answer comes back as a traffic chart, they’re measuring the last era.Here is an Uphoff Advisory Framework you can download and use in your business. Run both tests on yourself and your top three competitors, then find your square. Each one tells you what to run first, what the risk is and what to watch.Download Discoverable Is Not Recommended → I’ll rerun the test in JanuaryZero and one are where my two businesses start. I know which pages to fix, which phrase to own and which lists to earn. I’ll rerun the same 19 prompts in January and publish the numbers either way.The models are drafting your buyer’s shortlist right now, from the record the web already holds on you. Write that record on purpose, or a stale org chart will write it for you.Keep going: B2B Branding Has a New Audience. Most Companies Are Talking to the Wrong One and B2B Thought Leadership and Content Marketing Are BrokenThe views expressed in Uphoff on Media are entirely my own. They don’t represent the opinions of any company I’ve led, any board I’ve sat on, or any investor who’s had the pleasure

I’m a fan of radical thinking. It forces you to test assumptions you didn’t know you were making. You don’t always have to agree with it. Agreement was never the point.The point is whether it makes you think. Jack Dorsey and Roelof Botha’s essay, From Hierarchy to Intelligence, did that to me. It lays out how Block is rebuilding itself as a company organized as intelligence rather than hierarchy, and it came a month after Block cut its workforce by about 40%. Most of what gets written about AI and organizations stops at copilots, a smarter assistant bolted onto the same old structure. They go further and ask whether the organization itself still needs to exist in its current form.It pushed me past the easy cliché that AI just means fewer middle managers. Here’s where that thinking led.The oldest constraint in businessDorsey and Botha trace two thousand years of the same fix. The Roman legion built command around eight soldiers to a tent. The Prussian General Staff existed to route information up and down the chain. Daniel McCallum drew the first org chart in the 1850s to keep a 500-mile railroad from killing people.Every version since, from Taylor’s pyramid to the matrix, worked around one constraint: a human brain can only hold so much context, so organizations add layers, and layers slow down information. Spotify, Zappos and Valve tried to break the pattern. All three scaled back into hierarchy or hit a ceiling.Dorsey and Botha’s claim is that AI removes the constraint, not the org chart around it.What Block actually builtBlock is building a company world model (what the business knows about its own operations) and a customer world model (what it knows about the people transacting with it), and using both to compose solutions directly, without routing decisions through management. What’s left are three roles:* Individual contributors who build* DRIs who own outcomes across whatever teams a problem touches* Player-coaches who develop people instead of running status meetingsNo middle layer required.Money is the one signal nobody fakesHere’s the sentence that matters most in the whole piece.“Money is the most honest signal in the world.”Jack Dorsey and Roelof Botha, From Hierarchy to IntelligencePeople lie on surveys. They abandon carts. They ignore ads.But when they spend, that’s the truth. Block can build a company-as-intelligence because it sits on both sides of millions of transactions every day, through Square and Cash App. That gives it a signal almost nobody else has: honest, at scale, compounding.Where this breaks for the rest of usThat’s also where the model runs into trouble for most of the people reading this. Very few B2B information, marketing, or technology companies have anything close to a money-grade signal. What they have instead is engagement data: opens, clicks, content consumption, lead scores, intent signals bought from a vendor.I’ve run enough pipeline reviews to know how much of that gets gamed on purpose, by the same marketing teams reporting on it. The essay’s point about people lying on surveys is more true of modern B2B marketing than its authors probably intended.A lead score isn’t a fact about someone’s business. It’s a fact about how well your content ranks.Most B2B signal sits at the wrong end of the spectrumThat gap is the actual work before anyone in B2B tries to run Block’s playbook. Think of it as a spectrum, honest signal on one end and performative signal on the other. Self-reported data and form fills sit at the weak end: cheap to produce, cheap to fake.Here is an Uphoff Advisory Framework you can download and use in your business. Put every number your leadership team reviews each week on it, and see how many sit right of center.Download The Signal-Honesty Spectrum → Engagement metrics sit a notch above: harder to fake outright, but easy to optimize for the metric instead of the buyer. Usage depth and renewal behavior sit higher still, because pretending costs the customer something real. Contract value, expansion revenue, and churn sit closest to money itself.The businesses already living near that end of the spectrum are the ones actually positioned to build a world model worth trusting. Everyone else is redesigning an org chart around a signal that isn’t honest yet.Everyone else is redesigning an org chart around a signal that isn’t honest yet.This isn’t just Silicon Valley talkingIt would be easy to dismiss this as some crazy idea from the tech elite. Don’t. The data says otherwise, and it’s not subtle. The number of managers and executives at public companies fell 6.1% between May 2022 and May 2025, according to Live Data Technologies.Gartner predicted in October 2024 that through 2026, one in five organizations would use AI to flatten their structure, eliminating more than half of current middle management positions. We’re in the last quarter of that window.I’ve been the skeptic on AI-attributed layoffs. In June I called most of it a misattribution problem: pandemic correction and PE margin engineering, relabeled as AI. That’s still true of last year’s numbers. But manager headcount specifically, not total headcount, is the number that’s moved. Nobody announced that. They just stopped backfilling the role.This isn’t theoretical, and it isn’t just Block. GitLab and Coinbase both cut management layers this year and pointed to AI when they did. Cloudflare’s cuts reached middle management too. Google said last year it had eliminated 35% of its managers overseeing small teams. Intuit, Meta, Cisco, Atlassian, Salesforce, and Oracle all paired 2026 layoffs with AI investment or efficiency claims. Block didn’t start this. It’s just the company doing it with the strongest claim to a signal that actually justifies it.Revenue per employee is how the market keeps scoreBlock’s reorg came with a scoreboard, and it isn’t the org chart. Run Block’s own numbers:* 2025: $10.36 billion of gross profit with 10,205 people. About $1 million per person.* The plan: $12.2 billion for 2026 with just under 6,000 people. The stock jumped more than 20% after hours.* Now: guidance raised twice, most recently in August to $12.51 billion, after second-quarter gross profit grew 25%. About $2.1 million per person, double where it started the year.The market wasn’t paying for the layoff. It was paying for the output per head.Then: a vanity stat. For most of my career, revenue per employee was a number nobody built a valuation around. A 2013 analysis of public software companies found it had almost no relationship to market value. Hiring waves, sales ramp times and company age moved it more than strategy did.Now: the cleanest read on scale. AI changed what the number measures. It shows whether technology or headcount is doing the work, and that’s the question every buyer and investor is asking. Here’s where the bar sits:* Marketing agency average: about $163,000 per employee* Median public software company: about $395,000* Median AI-native software company past $100 million of ARR: $655,000 (ICONIQ, September 2026)* Top quartile of that group: $890,000Download The Revenue per Employee Ladder → Why it’s hard to fake. Anyone can say AI made their team more productive. Revenue per employee shows whether the productivity reached the P&L. That’s why I added it to the Uphoff Value Index this quarter as a driver in all eight categories we benchmark. A services firm approaching the software standard earns up to a full turn of EBITDA. Two agencies with identical P&Ls, one running with 68 people and one with 40, now land nearly $1 million apart.One caution from the operator’s chair. Revenue per employee can be gamed too. Push work to contractors or offshore partners and count only full-time staff, and the number looks better than the business is. Expect buyers to start asking for revenue per dollar of total labor cost, not just per head. Measured honestly, it’s the closest thing most B2B companies have to a money-grade signal about themselves.Marketing and sales break firstIf you’re running marketing or sales, none of this is abstract. These are the two functions most exposed, and the exposure isn’t only about headcount. It’s about who you hire, how you coach them, and what “good” looks like once the AI is closer to the honest signal than your team is.Marketing: the funnel was never honest. Marketing has been running on performative signal for years, and the funnel proves it. Forrester has found that fewer than 1% of leads make it from the top of the funnel to a closed deal.The industry knows the model is broken, and vendors are already pitching a replacement. Docket calls it the Agent Qualified Lead: a lead qualified by an AI agent in a live buyer conversation, capturing real intent instead of inferring it from a click or a form fill. Hiring for that looks nothing like hiring for lead-gen volume. Coaching it looks nothing like reviewing MQL counts in a Monday pipeline meeting.The SDRs who survive aren’t dialing more. They’re coaching AI performance.Sales: the job is splitting in two. It’s happening right underneath your comp plan. Regie.ai’s analysis of LinkedIn data found full-cycle AE seats up 10.8% while SDR headcount fell 22.7%. Two-thirds of teams cut SDRs. Fewer than one in five added them.The SDRs who stayed are running AI on the accounts that matter, not working the phones. That’s a different hire, managed a different way, measured against a different definition of quota progress.The playbooks most of us built our marketing and sales organizations on assumed a human was doing the qualifying, the outreach, and the judgment calls. Agentic AI is already doing large parts of all three. Keep running the old playbook and you’ll be coaching people to compete with a system that’s already faster and can’t be gamed.The real work comes before the reorgThat’s the mistake I expect to watch a lot of companies make over the next two years: skipping straight to flattening the org before doing the harder, less glamor

Somewhere in your organization right now, a team is piloting technology nobody can explain the job of. They can describe the vendor. They can describe the demo that got everyone nodding. Ask them what specific job that technology was hired to do, and watch the pause.That pause is the whole problem. Most AI pilots that fail were never given a job. Nobody wrote down the job description the technology was hired to do, who does that work today, and what would prove it got done better. Writing it down costs nothing. Skipping it costs the pilot.Procurement gave this mistake a name. It doesn’t stay in procurementOn a recent episode of the Art of Procurement podcast, Darshan Deshmukh, President of ProcureAbility, gave this pattern its name: “death by a thousand pilots.” His diagnosis was just as plain. Organizations keep leading with the technology instead of the outcome. Philip Ideson, who co-founded Art of Procurement with Kelly Barner and built it into the source procurement leaders actually rely on, put the consequence best in his write-up of the episode: “A compelling platform or use case can be a source of inspiration, but it cannot define the destination.” (Disclosure: Art of Procurement is an Uphoff Advisory client.)None of this is new. Procurement made this mistake with ERP, then with e-sourcing, and every other function made it with CRM. Same mistake, different software.The ERP era supplied a half-billion-euro example. In 2018, Lidl walked away from a seven-year SAP project after spending roughly €500 million on it, according to Handelsblatt. The sticking point was not complex. Lidl valued its inventory at purchase price. The software was built around retail price. In my read, nobody settled up front which way the business actually needed that job done, and hundreds of consultants spent years trying to bend one to fit the other.I’ve never run a procurement organization. I have spent three decades running media, marketing, and technology businesses, and every one of them had its own version of this: the CRM migration that died in month four, the martech stack nobody fully turned on, and now the AI copilot every vendor in your inbox is pitching your team. The martech version is measurable. Gartner found marketers used 58% of their martech stack’s capabilities in 2020, 42% in 2022, and 33% in 2023. The spend kept going up. The use kept going down.The failure point is identical everywhere it shows up. The technology got approved. The job never got defined.Why do most AI projects fail? Rarely because of the technologyThe failure rate for this generation of pilots is well documented. S&P Global Market Intelligence found that 42% of companies abandoned most of their AI initiatives in 2025, up from 17% the year before, and the average company scrapped 46% of its proofs of concept before they reached production.The more useful finding is why. When BCG surveyed 1,000 senior executives in 2024, it attributed roughly 70% of the challenges in getting value from AI to people and process, 20% to technology, and only 10% to the algorithms themselves. The part every demo is built to sell is the smallest slice of the problem.Now look at the budget for your last pilot. If most of it went to the vendor, you funded the 30% and left the 70% to chance.McKinsey’s August 2026 State of AI survey points the same direction from the other end. The companies getting real earnings impact from AI are a small group, about 6% of respondents. Nearly three-quarters of them say they have fundamentally redesigned workflows around it. Among everyone else, the figure is about a quarter. The winners started with the work. Everyone else started with the tool.The business case already exists. It was built backwardsAlmost every one of those pilots cleared some kind of approval process. Somebody wrote a business case. There were ROI projections, a vendor comparison matrix, maybe a proof of concept demo that went well. This is not usually an absence-of-diligence problem. It’s a diligence-aimed-at-the-wrong-target problem.The business case documented what the technology could do. It rarely documented, in plain words a non-technical person in the business could repeat back accurately, the job the organization actually needed done. Capability and job are different questions. Most business cases answer the first and assume the second.Pilots multiply for another reason nobody puts in the business case. A pilot is the one way to say yes to the vendor and no to accountability in the same meeting. Nobody’s name is on it. Nobody set a date to decide. It costs less than the argument about killing it would.Death by a thousand pilots is death by a thousand decisions nobody made.A job is not a use caseI’ve made a version of this argument before, about customers rather than internal technology decisions. When I was brought in to transform ThomasNet.com, my team thought we understood the dozen or two things our customers were trying to accomplish on the platform. The real number, once we did the research, was over 350. We had been building for what we assumed the job was, not what it actually was. I wrote about what that discovery did to the company, and to my own thinking, in the post You Don’t Know What Your Customers Are Actually Trying to Do.The same blind spot shows up on the buy side, and it may be more dangerous there, because nobody outside your own walls is going to do that research for you. A vendor’s use case library describes what other companies did with the tool. It does not describe the job inside your organization, under your specific circumstances, that this technology is being asked to do. If you cannot state that job without naming the technology, you haven’t defined the job. You have named a category.A better demo makes the mistake easier to makeAn ERP demo in 2004 was a slide deck and a promise. An AI demo in 2026 is the product working on your own data, in your own language, in front of the people who will approve it. The gap between the demo and the job has never looked smaller. It has rarely been larger.A demo shows one job done well, under conditions the vendor chose. Your organization runs hundreds of jobs, under conditions nobody chose. That was the lesson of the 350 at ThomasNet, and no demo is built to show it to you. A good demo makes defining the job feel unnecessary. That is what it was built to do.So I have a rule, and vendors do not love it: no demo until the job is written down.Two companies wrote two different sentencesKlarna is the clearest recent example of what happens when the job gets written wrong. In early 2024 the company said its AI assistant was doing the work of 700 customer service agents and handling two-thirds of its customer chats. By May 2025, CEO Sebastian Siemiatkowski told Bloomberg the company was investing in human support again. His explanation was unusually candid: “As cost unfortunately seems to have been a too predominant evaluation factor when organizing this, what you end up having is lower quality.” The job got written as “cut the cost of support.” The job customers needed done was “get my problem resolved.” The pilot did the first job and failed the second.Morgan Stanley did the opposite. Before rolling out its GPT-4 assistant to wealth management, the firm started with a narrow job: help a financial advisor find the right answer, fast, in a library of roughly 100,000 research documents. Every use case was tested against real advisor tasks, with advisors grading the output, before it shipped. According to OpenAI’s case study, more than 98% of advisor teams now use it. Same class of technology as the pilots in the S&P numbers. The difference was a job defined before the purchase order was sent. I wrote the jobs down before I hired a developerWhen I set out to build the decision tools I use in my advisory work, the easy path was obvious: pick a fast AI development platform, sketch a rough concept, and let a good demo carry the rest. I did the opposite. Before I hired a developer, I wrote out the job to be done for each tool.The valuation tool’s job: when a CEO is weighing a major decision, show what that decision does to the value of the business, and on what timeline. Not “calculate what the company is worth.” That distinction decided what got built. A valuation calculator answers a question once. A decision tool gets checked before every big call, so it needed drivers an operator can actually move, benchmarked against real trading multiples.The talent tools’ job: when a manager has to hire, coach, or restructure a sales or marketing team, help them make that call on evidence, starting with telling a strong resume from a strong hire. Writing that down ruled out the obvious product, an employee-tracking system. The job is the decision, not the monitoring.The technology, the specific platform, the build sequence, all came after those jobs were written down, not before. That ordering is the entire difference between a tool people keep opening and one that stops getting used after the second demo.Agentic AI removed the friction that used to slow this mistake downThe old version of this mistake had natural speed bumps built in: long procurement cycles, an IT approval queue, a budget committee that met quarterly. A bad idea had time to die on its own before real money moved.Agentic AI tools mostly skip all of that. A capable pilot can now show up as a subscription one manager puts on a corporate card, no committee required. Zylo’s 2026 SaaS Management Index found expensed software spend up 267% in a year, with ChatGPT now the single most expensed application, and business units, not IT, controlling 81% of software spend. Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027, and its analysts describe most current projects as experiments “mostly driven by hype.”The old failure mode took a fiscal year to play out. This one takes a sprint. A twenty-year-old mistake now compounds a

Paramount Skydance closed its acquisition of Warner Bros. Discovery on October 6. The combined company is now simply called Skydance: roughly $70 billion in revenue, more than $30 billion a year in content spend, and about $79 billion in debt.Most of the coverage will treat this as a Hollywood story. Studio lots, franchise slates, who runs HBO. That misses the point. This is a technology bet with a studio attached, and David Ellison has made that his whole argument: “Unless you can build a tech product truly competitive with Silicon Valley, you can’t compete.”You might be asking how the merger of Warner Bros. and Paramount is a B2B story. It's the clearest case study I've seen this year of a legacy business adapting to technology disruption. The real test isn't the deal. It's whether the company can change how it operates, and the order it has always done things in.Ellison is right about the problem. Whether he can fix it comes down to a sequence the traditional studios have gotten backwards for most of the streaming era.Netflix won on the how, then the whoNetflix didn’t take over filmed entertainment because it had better content. It didn’t produce award-winning content until 2013, when House of Cards won its first Emmy. Netflix was 16 years old. It won because it started with the how: consumer choice, ease of access, and a user experience built by technologists, not bolted on by a distribution department.Only after it owned the how did Netflix go after the who: world-class creators on screen and behind the camera. By then it had something no studio could offer talent. It knew exactly who was watching, what they finished, and what they’d watch next. The how made Netflix the most attractive place for the who to work.That order matters. Netflix still doesn’t own a cloud. It rents Amazon’s, and still spent $3.4 billion on technology and development last year, because the advantage was never the servers. It was the product organization that decided what got built.The studios bought the business model but kept their old operating modelThe traditional studios watched this happen and responded by launching streaming services. Subscription pricing, original programming, a global app. They copied the business model.They never changed how they worked. The operating model stayed talent first and talent last, with user experience treated as somebody else’s job. The apps launched feeling like 2009 enterprise software draped over a content library. Warner Bros. Discovery renamed HBO Max to “Max” in 2023, erasing one of the most valuable brand names in media, then put it back two years later. That isn’t a technology company making a product decision. That’s a studio treating the one of the most valuable consumer brands as an afterthought.Every prior owner of Warner Bros. made a version of the same mistake. I traced that history in Bundling Didn’t Die. It Moved.: AOL, AT&T and Discovery each arrived with a distribution or cost thesis, and none of them changed how the place actually worked.Skydance is the first studio organized how-firstLook at what Ellison built before the deal closed, and almost none of it is about movies.Paramount+, Pluto TV and BET+ are moving to a common Oracle backend. Skydance Animation already runs Oracle’s “Studio in the Cloud” production model, and Ellison intends to scale it across every production workflow. Safra Catz, Oracle’s former CEO, joined the board. Dane Glasgow, a former Meta and Google product executive, now runs product, digital platforms and AI. Larry Ellison personally guaranteed roughly $45.7 billion of the equity.The closing announcement reads the same way. More than $6 billion in run-rate synergies over three years, with technology the first source named. HBO Max and Paramount+ unified “into a single service over time.”And the org chart. Ynon Kreiz, the former Mattel CEO, became co-CEO on October 5 and owns day-to-day operations and integration. Ellison kept strategy, creative, talent, technology and capital allocation. Other than Netflix, no studio has made technology the centerpiece of its strategy the way Skydance is.Cloud + AI + IP is the how. It still needs the who.Ellison’s formula is Cloud + AI + IP. The first two terms are the how. The third is what the who produces. And the plus signs hide the real math: this works as multiplication, where a zero anywhere zeroes everything.Cloud is the easiest term to buy. Moving to Oracle can lower the cost to serve. On its own it builds nothing a competitor can’t rent next quarter. One catch: Oracle makes money when Skydance consumes more cloud, so watch whether that relationship gets priced and measured like any other vendor’s.AI is only as good as the data under it. Skydance today has two flagship streamers on two technology stacks, a free ad-supported service, a stack of cable networks and two newsrooms. That’s a lot of customer data and no single view of the customer. Netflix’s how was built on one identity, one product, one recommendation layer. Until Skydance has that, AI gets applied to fragments.The who is the term studios think they already have. They don’t, not anymore. HBO’s value was built on creators believing they’d be protected, and the Zaslav years eroded a lot of that trust. That’s why the leadership moves on the eve of closing matter. Casey Bloys, the steward of the most trusted creative brand in the building, is set to run all of streaming. And Kreiz is the executive who turned Mattel from a company that sold plastic into one that monetizes characters. The Barbie movie grossed more than $1.4 billion worldwide. Mattel's direct take was about $125 million, and Barbie toy sales rose 27% in the fourth quarter but just 3% for the year. The ticket sales were never the point. The movie proved Mattel's characters could carry a franchise, and that's the asset Kreiz knows how to build.Here’s the tension Ellison has to manage. The how-first instinct, applied by a company carrying north of 6x EBITDA in debt, points straight at cost. Fitch cut Paramount to junk in March, and the stated target is 3.0x leverage by the end of 2029. Every quarter between now and then will reward using technology to cut production cost. Netflix used technology to make the experience better, which is what made it the place creators wanted to be. Same tools. Opposite outcomes.The line between a technology company and a tech-enabled one is disappearingNetflix disrupted Hollywood as a technology company. It built its own product, its own recommendation engine and its own engineering culture, and it got valued like a tech company. Content was what the platform needed.Skydance is something else: a tech-enabled studio. It owns the IP, rents the infrastructure from Oracle, and hired product leadership instead of building an engineering organization from scratch. A few years ago, that structure all but guaranteed second-place software. Agentic AI changes the math. When agents write the code, run personalization and handle the workflow, capabilities that once took a large engineering organization can be rented and managed by a much smaller team.That’s the real experiment inside this deal. Skydance doesn’t need to become Netflix. It needs Netflix’s results from a different structure: IP it owns, technology it rents, and agents doing work that used to take an engineering department.Boards and investors are asking every B2B information CEO a version of the same question: do we need to become a software company? In the agentic era, the answer is increasingly no. You need to operate how-first. Who builds the stack matters less every year. Who decides what it’s for matters more.The same sequence error runs through B2BI’ve covered Oracle as an analyst for Cloud Wars, and I’ve spent most of my career running businesses that owned valuable content and rented the technology. B2B made the studios’ mistake first.B2B information companies digitized the business model years ago. Print subscriptions became digital subscriptions, then SaaS pricing, then data licenses. Most of them never changed how they worked. Editorial first, sales first, product experience somebody else’s problem. Plenty of B2B companies still have world-class experts and analysts sitting behind a product their customers tolerate rather than use.When we rebuilt ThomasNet from a 120-year-old industrial directory into a data platform, the infrastructure decision was the easiest one we made, and in the end, the step that opened the door to growth and value. We moved legacy, on-premise systems to the cloud and went all in on AWS, the same cloud Netflix rents. The hard part was changing the focus of the business: from obsessing about a declining base of advertisers to building the product and user experience around how buyers actually searched and sourced. Then we put a century of supplier content to work inside that experience.Agentic AI is about to make this sequence error expensive. When the buyer is an agent, the how is the whole front door. If the machine doing the buying can't find, read and use your content, it never gets considered. The who doesn't matter if the how can't get it through the door.Here is an Uphoff Advisory Framework you can download and use as a diagnostic tool in your business. Plot where you are on the how and the who today, and which one you’re actually investing in.Download The How/Who Matrix → Four signals will show which sequence Skydance is running1. Does the Oracle migration show up as a better product or only a cheaper one? Cost to serve is half the story. Watch the app.2. When does one subscriber identity span HBO Max and Paramount+? “Over time” isn’t a date.3. Do the first AI announcements cut production cost, or create new value from existing characters and catalog?4. Does HBO’s talent stay? The creative deals signed over the next 18 months will answer that before any press release does.Ellison has the capital, the cloud relationship and a credible division of labor at the top.