SaaS METRIC OF THE WEEK
ARR - it ain't the trophy metric it once was - that and other traditional SaaS metric takes from current VCs here.
COMPUTE
The AI price war and the AI debt boom are hanging out at the same time. $500B of other people's money, lent against GPUs. That's Nvidia's new plan with Apollo, BlackRock, Blackstone, Brookfield, Goldman, and KKR - announced at the same time token prices fell up to 80% and memory started building toward its next glut.
MAGICAL THINKING
It's a trap! Check these 5 magical thinking traps that are common for new founders to fall into - Build it, and they'll come; launch it, and they'll buy; add more, and they'll convert; move faster, and my fav - a few early customers means a viable business.
SECURITY
I find the threat of AI-enabled cyberattacks terrifying (and they are inevitable) - the president at OpenAI agrees - he lists 10 things to do ASAP against AI-enabled cyberattacks (remember his own agents hacked Hugging Face, so he'd know - see #10 here). Give your security team an agent today, read-only first, triage the backlog, security review in CI. His phrase is "turbo speed."
VENTURE DEBT
A £10M facility quoted at 10% actually costs 14.3% in cash and gives the lender 16.8% once the warrants land. Every point of the gap is disclosed somewhere in the documents - the 2% fee on day one, PIK quietly compounding the balance to £11.3M, a 3% exit fee someone agreed to as a rounding error.
CAPITAL
Every private capital strategy is down this year except one. Private debt raised 27.3% more in H1 than last year and is tracking its second-best year ever. $5 trillion is now stuck in private funds seven years or older - 39% of everything. Fundraising dropped to $1.35T while the number of funds fell 37% to 3,763, showing capital is concentrating - 78.2% of capital went to $1B+ funds, which is up from 59.1% in 2021.
PITCHING
Check this report from Papermrak (they tracked 24,541 decks and 358,672 investor views) - Investors give your deck 4 minutes, and fewer than half of them ever reach the last slide - which is where we all put the ask. Put your best facts in the first three slides, ask early, and the appendix goes behind everything. Super helpful report. BTW - TEAM is the most-read page (at 5.7 seconds).
GRAPH ENGINEERING
New one for our tech dictionaries - A graph is a loop with something that can contradict it. It's worth it for the phrase "a larger hallucination with better project management" alone. 20 AI agents reviewing each other's work ignores the same blind spots and agrees with itself twenty times.
CMO
Founders are killing the CMO position - brand goes into product, demand goes into sales, problem solved. You are better off with no CMO than the wrong CMO. Good idea from the article - hire the person who could run marketing in 18-24 months, and pay a great CMO to advise them for a year. Cheaper than the mis-hire, and the mis-hire builds a team in their image for years.
CASE STUDY
DETOX: 3 days without your phone and your brain starts showing the same activation patterns researchers see in substance withdrawal. Check this study - Researchers took smartphones off 25 young adults for 72 hours. The participants reported no extra craving and no worse mood - which seems fine, but their brain scans said otherwise.
Get the top 10, every Friday
Curated SaaS and tech insight from around the web, repackaged for people to put to good use — free.
Done - check your inbox to confirm.
SaaS METRIC OF THE WEEK - MMM
This is a new acronym for your tech dictionaries - MMM, or Marketing mix modeling, is used to measure the cross-channel impact of all the various marketing and non-marketing drivers on your startup's sales or business efforts and performance. Google's Meridian is its open-source take on it.
PRINCIPLES
I love me some laws and principles - here is a new list of 10 of the lesser known ones. Hanlon's Razor is a goodie: rather than questioning people's intentions, question their competence.
AI THESIS
Sarah Guo built Conviction around one bet: the biggest AI opportunities sit above the foundation models, not inside them. Her logic is simple - the labs become infrastructure, while startups win by owning the application, workflow, and customer.
VERTICAL INTEGRATION
Counterpoint to Sarah Guo above (#3). Akash Bajwa reckons model labs and agent companies are converging the stack, not separating - Anthropic is building products while Harvey is training models. The endgame may be owning the model, the application, the workflow, and the customer around it.
WATERMARKS
Last week Anthropic/Claude announced that it will now embed an invisible watermark into the text it generates - and it can survive copy-paste and some editing. The weird bit: even using Claude to proofread or translate your own writing can leave the mark. Heavy rewriting can break it, but a third-party detection tool is coming. (not necessarily by choice - this is part of Anthropic's EU AI Act compliance).
BENCHMARKS
Stop asking if you're above average. This article argues the more useful benchmark question is whether you're improving faster than everyone or anyone else. Your own trend shows change; the market trend tells you whether that change was actually yours or just the tide coming in for everyone.
SPEND
Speaking of Benchmarks, the SaaS Capital Spending Benchmarks for B2B SaaS Companies is out - bookmark this when it's time to benchmark your wallets y'all! Bootstrapped B2B SaaS companies spend 96% of ARR (hey - I do better than that!!), while VC-backed ones burn 101% (last year it was 95% vs 106%). Equity-backed teams in turn grow at 25% average vs 20% if your boots are strapped.
RETENTION
Meta laid off 10% of staff and reassigned 20-30% of software engineers to AI whatever - now they are reportedly offering $400K-$1M+ in equity to senior engineers trying to resign after layoffs/AI reassignments. Some still left for Google, Anthropic, and OpenAI.
LIQUIDITY
Startup liquidity is quietly becoming its own market. Carta administered 71 tender offers worth $3B in H1 2026 - the highest first-half total it's logged in at least six years. With IPOs taking longer, tenders are becoming the liquidity pressure valve for employees and investors.
CASE STUDY
WISE: Wise's original Seed deck is basically the opposite of most pitch decks today. No giant TAM slide, no AI-generated design theatre. It explains the problem, shows the before/after economics, proves people are already using it, then asks for $1.3M.
This issue uses the long-form layout — read it on its own page →
SaaS METRIC OF THE WEEK
Churn.: See #2 for more on Churn\*\*.\*\* According to CatchJS, though, we're all calculating churn rates wrong. If you love Statistics, the article is well worth reading, and it even gives some Python code to perform the more complicated probability-based equation they recommend. You can then check out this tool (as a handy Google Sheet) from Newfund as a way to analyze the strength of revenue streams for any B2B startup. A complimentary article outlining the methodology behind the tool is here (and you should read it first).
CHURN
See #1 above\*\*;\*\* it's the ultimate leaky bucket. BVP's guide on tackling customer churn explains how to identify root causes and implement strategies to reduce attrition (that can be terminal). Analyzing churn data, improving customer onboarding, and enhancing product value to retain users are all in there. The guide also provides actionable steps for creating a comprehensive churn action plan to plug those leaks.
INVESTMENT
Hey founders - check yourself (at least your metrics) - Investors have an experienced nose for this stuff, and generally, legitimately good metrics only mean something in pairs. NRR next to GRR tells you if retention is real or propped up by your upsells. Burn multiple at 1.1x - nice, that's top quartile. Now look at Rule of 40 next to it, down from 38% to 22% in four quarters? That's not efficiency, that's a company that stopped spending and stopped growing at the same time.
ONBOARDING (STAFF)
Only 12% of employees reckon their company onboards them well, so check out these 7 steps for onboarding a new team member (and they start two weeks before day one). Real handy if your onboarding currently means a laptop, slap on the back, and a Slack invite.
VENTURE
Carta's VC Fund Performance Q1 2026 is out. $3.9B raised across 86 new funds, the strongest start to a year since 2022, and TVPI (Total Value to Paid-In) is up for nearly every fund vintage. All on paper though. The median 2017 fund has returned 0.31x to its LPs after nine years, and 2018 sits at 0.15x.
AGENTS
Computer-using agents went from 42% on OSWorld a year ago to 85% today (which is past the 72% humans score). Before you get outraged and/or excited - 85% means 15 of every 100 tasks failed, and a business process only finishes if every step does. $6-8 an hour is what a16z reckons it costs to run a computer-using agent, against $10 for offshore people and $30-45 for a US back-office worker.
CPU
Turns out the agent boom isn't only a GPU story - AWS engineers were told in May to cut CPU waste and now reportedly wait days for an EC2 instance that used to take hours - agentic workloads are making CPUs scarce. Apparently, Intel, AMD, and others say customers will take whatever they can get.
VISIBILITY
Your sales are flat so you post more, add a platform, say yes to another podcast......etc. Sylvia Huang calls it "the visibility treadmill" - more output, same revenue, tireder founder. The real question underneath: have enough of the right people actually seen this yet? Until they have, you can't tell a reach problem from a product problem.
TRUST
Report time and ignore the Plaid bias evident in this - rest of it is good. 44% of Americans would let an AI agent execute trades, it goes up to 54% for Gen Z and Millennials. But on the other hand, 80% of Americans think companies should reimburse them when AI makes a mistake with their money, and 78% say AI should be held to the same accountability standards as a human advisor. Can you have your AI cake and eat it?
CASE STUDY
M&A -: Airtable just sold to Bending Spoons for $1.3B, and they last raised at $11.7B in 2021! CJ Gustafson pulled 10 lessons out of it, and they're all worth your time, but here's the one that stings: back-of-the-envelope math from 20vc says nobody who invested after March 2018 made money (that includes employees).
▌ Older issues Full archive (387) →