Is AI delivering measurable ROI inside large enterprises today?

Harry Stebbings · Oct 28, 2024

Outside of customer service, building with AI has not yet meaningfully changed how people in most roles work

The visible example is an 80% reduction in customer service cost; beyond that little has actually changed for different roles

Scope: as of today

9:03 20VC: Why SaaS is Dead | Why AI First Companies Will Win | We are in the Middle of a Cold War for AI Talent | Why Europe is F******* and We Need to Stop Whining with Daniel Khachab, Co-Founder @ Choco

Harry Stebbings · Apr 4, 2026

AI's impact is real in engineering and customer support but overexaggerated in other functions

11:31 20VC: Hims & Hers: $4.3BN Market Cap on $2.3BN of Revenue: The Comeback | Why Being Public is 10x Better | The Death of the "Strategy" Hire | Why Performance Marketing is Worse than Brand Marketing with Andrew Dudum

Harry Stebbings · Jun 20, 2026 · hedged

Claims about AI's impact are being extrapolated too far — the tools can't even do basic tasks like making clips, and Uber and Microsoft say productivity gains are questionable

He can't even get AI to cut clips for his show, and large operators are publicly sceptical.

19:45 20VC: Why Remote Work is White Collar Fraud | Why Revenge and Patriotism are the Best Founder Traits | Two Questions Every Founder Needs to Ask | The Wild Story of Raising $1BN from Masa Son in an Hour Long Meeting with Ryan Peterson, Founder @ Flexport

Arvind Jain · Jul 11, 2026

AI value realization in enterprises today exists in pockets, clearest in customer support where productivity is directly measurable

Support work is largely reading knowledge and summarizing it to customers — a task AI is good at — and case resolution counts give a concrete productivity metric

Scope: other use cases are more complex

21:17 20VC: Why OpenAI and Anthropic Won't Win the App Layer | Why Teams Will Get Bigger Not Smaller in a World of AI | Why AI Removes Incumbents Advantage of Bundling | China vs America: Who Wins the AI War with Arvind Jain, Co-Founder @ Glean

David Frankel · Aug 8, 2026

AI has had far less real impact on software, enterprise and SMB than the hype implied, and consumer AI in particular is nowhere near played out

Beyond the models themselves and improvements in human interface like voice, he has not seen the transformation he expected

Scope: exceptions: the models themselves, voice and human-interface improvements

73:46 20VC: The AI Boom Will Create Enormous Roadkill: Who Wins & Loses | Why Founders Should Never Take Multi-Stage Money at Seed | Why Triple, Triple, Double, Double is Good Enough

Harry Stebbings · Aug 17, 2026

Outside of coding and customer support, AI is still not material to large enterprise P&Ls

Anyone who runs a budget in a large enterprise would say so

Scope: excludes coding and customer support

36:53 20VC: Uber President on The Untold Uber Stories: Travis, China and Self-Driving | Why Autonomy Is Existential | How to Beat DoorDash to #1 in Food with Andrew MacDonald

Harry Stebbings · Nov 15, 2023

Startups founded in the last twelve to eighteen months selling to incumbents are experiencing experimental budget evaporation.

Buyers ran a couple of interesting pilots, and then nothing went anywhere.

29:28 20VC: How to Survive and Thrive in a World of OpenAI, Are LLMs Being Commoditised, Where Does the Value Lie; Infrastructure or Application Layer, How Apple Could Win in a World of AI, How Amazon Could Threaten OpenAI and Why Google Struggle with Des Trayn

David Luan · Jun 24, 2024

The majority of enterprise AI spend today is still extremely experimental budget rather than core budget

Enterprises still run huge amounts on prem and on mainframes in 2024, and even cloud lacks full enterprise adoption

Scope: some specific use cases are clearly hitting PMF inside enterprises; 'AI' is too broad a category to answer for uniformly

40:09 20VC: Why Foundation Model Performance is Not Diminishing But Models Are Commoditising, Why Nvidia Will Enter the Model Space and Models Will Enter the Chip Space & The Right Business Model for AI Software with David Luan, Co-Founder @ Adept

George Sivulka · Jan 22, 2025

About 90% of the enterprise AI market is still in the experimental budget phase, though early signs of repeatable real value are emerging

Scope: his own business focuses only on the repeatable use cases

51:43 20VC: Why All AI Companies Are Under-Valued | The Future of Foundation Models: Scaling Laws, Generalised vs Specialised, Commoditised? | From Unable to Afford Rent to Raising $130M From Index and Peter Thiel with George Sivulka @ Hebbia

Vlad Tenev · Jul 14, 2025

AI has had a huge impact on Robinhood's largest internal teams, from software engineering to customer support, and is a big part of the cost reduction story

6:05 20VC: Vlad Tenev on Robinhood's $85BN Resurgence | Tokenizing SpaceX & OpenAI | Building Nine Revenue Lines Over $100M | Why Crypto Will Be Robinhood's Biggest Revenue Line | Why Stablecoins Will Replace Banking Rails

Andrew Ng · Nov 17, 2025 · hedged

AI coding assistants are the clearest current bucket of vertical value and a harbinger of what will happen to marketing, recruiting and finance functions as their tools improve

Coding tools are already making developers dramatically more productive, so demand explodes; the same dynamic should follow in other job functions as their tooling matures

Scope: framed as possible foreshadowing

6:10 20VC: Andrew NG on The Biggest Bottlenecks in AI | How LLMs Can Be Used as a Geopolitical Weapon | Do Margins Matter in a World of AI? | Is Defensibility Dead in a World of AI? | Will AI Deliver Masa Son's Predictions of 5% GDP Growth?

Markus Villig · Nov 13, 2024 · hedged

AI's usefulness extends well beyond customer support into CRM and sales/marketing productivity, internal tooling, information discovery and document summarization for better decisions

Customer support automation is already clearly workable and improves customer experience, and the same pattern is becoming obvious in adjacent internal domains

Scope: framed as growing optimism rather than proven

82:28 20VC: Bolt; The Most Insane Story in Startups | Turning a $5K Loan into an $8BN Company | Why Every VC Turned Down One of Europe's Biggest Winners | Competing with Uber & The Future of Micromobility and Self-Driving

Yamini Rangan · Mar 18, 2025

Customer support is the clearest area where AI agents are already delivering value, with marketing personalization second.

Across 1,500 customers the customer agent deflected 42-45% of tickets within weeks, HubSpot internally resolved 35% of tickets with AI in year one and is heading to 50%, and personalized emails have lifted conversion rates.

Scope: still early in the technology; looking for signals rather than proof

32:34 20VC: HubSpot CEO on Where Value Accrues in SaaS AI | How HubSpot Competes Against Salesforce | Why B2B Is Not a Winner Take All Market | How to Go From SMB to Enterprise an Win | How SEO Dying Changes HubSpot's Business with Yamini Rangan

Brad Lightcap · Apr 15, 2024

Enterprises criminally underrate the return on simply giving employees broad access to AI, over-indexing instead on throwing it at a specific business process with quantifiable ROI.

Someone who used to spend two days on a task and now spends two minutes is freed for 85 other things — that doesn't show up in enterprise ROI accounting, but multiplies across 10,000 or 100,000 employees

Scope: targeted ROI projects are still worth doing

31:38 20VC: OpenAI's Sam Altman and Brad Lightcap on The Future of Foundation Models: Will They Be Commoditised | How to Solve the Problem of Compute | Open vs Closed: Which Dominates and Why | Which Companies and Verticals Will Be Steamrolled by OpenAI

Andrew Macdonald · Aug 17, 2026

AI ROI at large companies is real but cannot be precisely quantified, because the hours freed up by AI get refilled with other presumably high-value work rather than showing up as headcount you can remove

You can't draw a direct line from transforming a process to needing two fewer operations analysts; the excess time is absorbed by other activity that may not have been done before or done as precisely

Scope: applies to pure OPEX/efficiency framing; he cites concrete process wins: capital allocation, forecasting, marketing QA

37:14 20VC: Uber President on The Untold Uber Stories: Travis, China and Self-Driving | Why Autonomy Is Existential | How to Beat DoorDash to #1 in Food with Andrew MacDonald

Omer Shai · Jan 31, 2026

AI collapses the marketing cycle from ideation to execution to A/B test from months into days or hours without degrading the results

Previously a brief took weeks of internal iteration and up to six months to land; now he can express his thinking clearly, hand it over and get back something better within hours

Scope: some AI output still needs human insight to reach the quality he wants; expects near-full compression within four to twelve months

48:14 20Growth: How Wix Built a $100M Marketing Machine | Why LTV is BS and Why Time Return On Investment is the Most Important Metric | How to 10x Your Growth: What is the Next Great Channel with Omer Shai, CMO @ Wix

Andrew Dudum · Apr 4, 2026

Design and creative production is one of the most tangible areas of AI leverage inside their organization, alongside engineering and customer support

They deploy about a billion dollars of marketing spend a year and produce thousands of variations of TV commercials and Facebook and Google ads; AI removes the cost and slowness of live photoshoots and dramatically speeds iteration

Scope: based on what he has observed across his own organization

11:45 20VC: Hims & Hers: $4.3BN Market Cap on $2.3BN of Revenue: The Comeback | Why Being Public is 10x Better | The Death of the "Strategy" Hire | Why Performance Marketing is Worse than Brand Marketing with Andrew Dudum

Joelle Pineau · Nov 3, 2025

A 10x productivity gain from AI is realistic for a lot of current work within the next couple of years

For well-specified tasks the output arrives in seconds where it used to take weeks or months — Hollywood-quality productions made in hours, long-form machine translation going from hours to seconds

16:54 Ten x output per worker is achievable within two years

David George · Dec 15, 2025

There are already real green shoots of AI ROI in public company earnings — C.H. Robinson's disclosed 40% productivity increase and 680bps operating margin gain show effective AI implementation and answer the 'is there real usage / are we in a bubble' question

C.H. Robinson replaced football-field-sized call centers with AI and disclosed shipments per person per day up 40% since end of 2022, with operating margin up 680 basis points

26:57 Public company disclosures already show hard productivity gains

Des Traynor · Nov 15, 2023

AI budgets for speculative, open-ended use cases will be cut to zero, while budgets for AI products delivering demonstrable value will stick and massively increase

ROI is self-evident for real products — during OpenAI's 45-minute outage support teams immediately felt the volume spike when Fin had to hand over to humans, whereas 'generate ideas for a workshop' style uses are box-ticking future-focus spend

29:58 Speculative ai budgets get cut to zero while demonstrable value budgets increase

Victor Riparbelli · Jan 15, 2025

Renewal, not signing new contracts, is the only real signal of value in AI, and many AI startups have optimized for the wrong one and will hit a wall of churn

Buyers are unusually willing to part with money right now, so a signed contract and even positive feedback can be false signals from someone justifying their spend; the value failure only shows up at renewal

19:20 Renewal not initial contract is the real roi signal

Aaron Levie · May 22, 2024

Enterprise AI spend is a mix of experimental and production budget — it is neither all experimental nor all production — and the exact split is currently impossible to diagnose

From about a dozen AI events across the US this quarter, the vast majority of companies have many experiments running but also areas already in production; it all gets lumped into one AI bill for the CFO

19:44 Spend is a mix of experimental and production budget impossible to cleanly diagnose

George Sivulka · Jan 22, 2025

Enterprises are not actually happy with RAG; about 90% of enterprise AI today is vaporware that demos well and fails in real-world use

The impressive demos ask questions like 'what does the CEO say about the investment', but the moment buyers try a real-world example the system completely fails, so usage statistics are unreliable

24:34 Most enterprise ai is vaporware that demos well but fails in real use

Matan Grinberg · Jun 13, 2026

Enterprise AI adoption is moving through three phases — board pressure on the CEO for an AI strategy, then 'AI at all costs' token-maxing with adoption written into performance reviews, and now a hangover phase of examining the bill and questioning ROI — with most companies currently in phase three

He is observing this sequence directly across the enterprises Factory works with, and phase two ran faster than expected

18:16 Adoption passes through token maxing then an roi hangover

Ryan Petersen · Jun 20, 2026

Most of the economy looks like freight forwarding — humans emailing PDFs and moving data between systems — not like already-automated businesses such as Uber, so the automation upside is enormous.

Freight forwarding is really 'freight email forwarding': huge volumes of manual human labour moving data around.

20:05 Most of the economy is still manual data shuffling so upside is huge

Ryan Petersen · Jun 20, 2026

Flexport's AI agent program only counts as successful if at least 80% of its ~100 core workflow agents actually go live and deliver savings; otherwise the AI spend is wasted

Only five of roughly a hundred agents are live and saving money today, so without conversion of the remainder the company is spending without return

74:22 Roi hinges on how many built agents actually go live

Andrew Macdonald · Aug 17, 2026

Uber's two widely-reported AI comments were misinterpreted by both AI skeptics and evangelists: the CTO's remark about blowing through the AI budget reflected hard-to-predict usage growth rather than runaway spend, and the remark that it was hard to draw a direct line from AI spend to useful consumer features was not a claim that there is no return on AI

Time saved gives people days back and lets forecasts carry more nuance; the results come from pairing business and GNA people with a pod of 30 top AI engineers going process by process

33:49 Rising spend reflects real usage not absent returns