ServiceNow Paid Hundreds of Millions to Kill Weekly Reports

So, What Actually Happened?

So I spent part of Sunday reading a pricing page, which tells you something about my weekend. Monday.com dropped per-seat pricing for AI credits, and I sat there trying to work out what a bill like that looks like twelve months from now. Could not do it. We scanned 190,000 articles this week so you don't have to. Same weekend, the reasoning behind ServiceNow's purchase of Pyramid Analytics finally got spelled out in public, and it comes down to one impatient sentence: nobody waits a week for an answer anymore. Then Persistent posted a record $1.15 billion in contract value, one deal of which runs six and a half years. Three different desks filed those. None of them filed the same story.

The Bottom Line: Enterprise software quietly stopped being priced per person this weekend, and nobody sent a memo. Your 2027 budget is still built on headcount.

The Tracks That Matter

1. ServiceNow Paid Hundreds of Millions to Kill the Weekly Report

Enterprises will not wait a week for an answer. Stripped of the press release, that is the whole reason ServiceNow bought Pyramid Analytics, and it is a strange thing for a workflow company to spend that kind of money on until you look at what it deletes. The old shape was a request to the BI team, a queue, a dashboard, then a meeting to argue about the dashboard. The platform acquired the analytics vendor back in February, so this is not fresh news, it is the reasoning arriving months after the receipt. ServiceNow now describes the result as data-driven AI expanding inside the platform, which is a polite way of saying analytics is no longer a place you go.

Here's what works: Count how many decisions this week waited on a report request. That queue is what vendors are now pricing, and you are already paying for it twice.

2. Monday.com Deletes the Seat and Charges for the Work

Monday.com scrapped per-seat pricing in favour of a hybrid model built on AI credits, and the honest first reaction from anyone who has ever defended a software line item is: how do I forecast that? A seat is a person. You know how many you have, you know when you hire, and you can defend the number in a planning meeting without sweating. A credit is a unit of work a machine performs, and how many you burn depends on how messy your data is, how ambitious your prompts get, and how enthusiastic one team becomes in March. The risk used to sit with the vendor: sell the seats, cover the cost. It just moved to your side of the table, dressed as a product update.

Here's what works: Meter one month of real AI usage per team before your next renewal. Without that baseline you negotiate a credit price blind, and the vendor does not.

3. Hospitals Are Piloting Agents Everywhere and Deploying Them Almost Nowhere

Healthcare is the cleanest picture of pilot purgatory in the data right now: 43% of US healthcare organisations are testing agentic AI while 3% have one running in a live workflow, and 81% of physicians already use AI professionally in some form. Appetite is not the blocker. The deals that actually move are narrow and contractual, which is the tell. ConcertAI put agents inside clinical trial timelines, a bounded problem with a deadline attached, and CytoReason expanded its collaboration with Pfizer on disease modelling, where the output has a named owner. The gap between 43 and 3 is not technology. It is whether somebody's name is on the outcome when the agent gets it wrong.

Here's what works: Pick one workflow where being wrong has a named owner and a visible clock. That is the only kind of agent pilot that survives a hospital.

Quick hits:

• Korea keeps funding models it can shape itself. Upstage raised $72 million for custom enterprise LLMs, and the appetite for models you host and control is not fading anywhere outside Silicon Valley.
• Israel had a July. Israeli startups pulled in $1.5 billion during the month, heavily weighted toward securing AI systems, proof that geographic clustering still beats capital efficiency.
• The UN put a science floor under the argument. The Secretary-General welcomed the first global scientific assessment of AI, which matters mostly because regulators now have something to cite that is not a vendor deck.

Signal vs. Noise

🟢 Signal: forecasting. Predictive analytics had the strongest week of any concept in the industry, climbing in both attention and real weight, and the ServiceNow deal explains why: once analytics sits inside the workflow, someone has to answer ”what happens next” at the moment of the decision, not the Friday after. Most coverage is still busy scoring model releases against each other.

🔴 Noise: the word ”automation.” It pulled heavy volume again while steadily losing its grip on what actually attaches to it. The tell is in this weekend's pricing news: nobody sells ”automation” anymore, they sell metered units of work at a credit price. When the buying language gets more specific than the coverage language, the coverage is behind.

From the 190K

We scanned 190,000 articles this week. Here's what no one's talking about:

ServiceNow spent heavily to fold analytics into the workflow, Monday.com replaced per-seat pricing with AI credits, and Persistent booked a record $1.15 billion in contract value including a six-and-a-half-year agreement, all inside one weekend.

Read separately, each one belongs to a different beat. The M&A desk takes the ServiceNow rationale, the SaaS desk takes the pricing change, the services desk takes the earnings release. Put them on one morning and they are three views of the same move: the unit enterprise software is sold in is being rewritten. Not the product, the unit. A seat used to be a person you could count, and now it is a bundle of consumption nobody can forecast. A dashboard used to be a place you visited, and now the answer arrives where the work already sits. Persistent's six-and-a-half-year deal says the same thing from the services side, where a project has stopped having an end date.

Every procurement process you own runs on the old units. Headcount times seat price, annual renewal cycles, a BI budget kept separate from the workflow budget. None of that survives the change cleanly, and the vendors know it, which is why this is arriving as product news instead of pricing news.

Monday's move is small and unglamorous: find out what one team actually consumes in a month, measured in whatever the new contracts count. Not seats.

By The Numbers

• Persistent booked a record $1.15 billion in total contract value — including a single agreement running six and a half years, at $452.4 million quarterly revenue and 16.1% growth.
• Only 3% of US healthcare organisations run agentic AI in a live workflow — against 43% piloting it, which is the widest gap between intent and production in any regulated sector right now.
• Israeli startups raised $1.5 billion in July — concentrated in securing AI systems rather than building them, in one month, from one small country.
• Glow raised $180 million at a $1.2 billion valuation — for endpoint security aimed at AI agents, a category that did not have a budget line eighteen months ago.
• Upstage raised $72 million for custom enterprise language models — money betting that plenty of buyers want a model they can own rather than rent.
• See what's rising in our 190K-article corpus this week →

Deep Dive: When the Album Became the Stream

I still have the shelf. Around two thousand records, each one paid for once and owned forever, and I can tell you roughly what any of them cost me. Then streaming arrived and the money changed shape. Nobody buys the album, everybody pays per play, and not one artist I know can tell you what next month's cheque will be.

The seat was the album
A per-seat licence is a purchase you can count on your fingers. Headcount times price, signed once a year, defensible in front of a CFO who has never opened the product. It was never accurate (half those seats never logged in) but it was predictable, and finance departments will trade accuracy for predictability every single time.

Nobody knows what a play costs
Credits are per-play economics landing in enterprise software. Consumption depends on data quality you have not fixed, on prompts nobody standardised, and on one enthusiastic team in March. The vendor stops carrying the variance and you start. That is not a scandal, it is just a transfer, and transfers are usually announced as improvements.

The catalogue moved into the player
Buying an analytics company to bury it inside a workflow is the same move Spotify made on the record shop. You do not visit the collection, the track finds you mid-task. Convenient, and it quietly removes the one moment where somebody used to stop and ask whether the number was right.

What Actually Works

1. Meter before you negotiate: one month of real consumption per team beats any vendor estimate at the table.
2. Put a variance band in the budget: assume consumption swings 40% either way in year one and say so out loud in planning.
3. Keep one human checkpoint on the answer: when analytics moves into the flow, the review step vanishes unless you deliberately keep it.
4. Renegotiate on units, not on discount: the price per credit matters less than what counts as a credit, and that definition is where the money hides.

Nobody misses the CD. Plenty of people miss knowing what the cheque would be. Learn to read your meter.

What's Coming

Seat-Based Pricing Cracks Across the Stack

Once a name like Monday.com moves off per-seat pricing, the rest of the category gets permission. Expect two or three of your existing tools to introduce a credit or consumption tier before year end, presented as flexibility. Read the unit definitions before the discount.

The BI Line Item Disappears Into the Workflow Line Item

With analytics folding into the workflow platform, standalone BI stops being a separate purchase and becomes a feature you inherit. That is convenient until renewal, when your leverage turns out to be attached to a contract you no longer negotiate separately.

Healthcare Agents Move From Pilot to Contract

The bounded, deadline-shaped deployments like agents inside clinical trial timelines are the ones that will convert first. Watch for the 3% figure to move on narrow contractual use cases, not on broad clinical ambition.

For Your Team

Tuesday's meeting prompt: ”If our biggest software vendors stopped charging per person and started charging per unit of work, could we forecast next year's bill? If the answer is no, who owns finding out, and by when?”

Share-worthy stat: 81% of US physicians already use AI professionally, and 3% of healthcare organisations have an agent running in a live workflow. The bottleneck was never enthusiasm. It is that nobody wants their name on the outcome when the machine is wrong.

Go deeper: Track where enterprise AI pricing and adoption are moving, in real time →

The Track of the Day

”As enterprises increasingly look to scale AI across their businesses, we believe the differentiator will not be the model itself, but the ability to create a unified Enterprise Context from business logic, data and enterprise experience embedded across the organization.”
Sandeep Kalra, CEO of Persistent Systems

Everyone else is still shopping for the model. He is describing the part you cannot buy in credits.

We scanned 190,000 articles this week so you don't have to. Data Pains → Business Gains.

Published: August 3, 2026 | Curated by Yves Mulkers @ Ins7ghts