B2B buyers pick their early favorite 82% of the time

Why most deals are won before the first call, which AI agents visit your site, and what agents could do to SaaS margins.

Newsletter //

Issue 3, November 10, 2026

Featured story

B2B buyers end up choosing their early favorite 82% of the time

6sense chart of where buyers' experience with the winning vendor came from: 73% were previous customers, 59.2% because they had bought it before and 14.0% because they had used it at another company, and 27% had other experience, such as evaluating it without buying (13.8%), a known brand or content (6.6%) or a colleague's experience (5.9%)

6sense, which sells buyer-intent data, released its 2026 B2B Buyer Experience Report on November 2. Across 3,596 completed purchases worth at least $25,000 a year, the vendor buyers favored before first contact won 82% of the time. Buyers first spoke to a seller about 59% of the way through a ~10-month cycle. In 73% of purchases, the buyer was a past customer of the winner.

Our thoughts

No surprise that buyers renew, but a good reminder that excellent delivery and customer service is its own sales funnel.

Also interesting that ~14% were those who previously evaluated but didn't purchase. Higher than I'd expect.

The report is chock-full of other interesting data points. Suggest a read if you have time (it's dense).

No surprise that buyers renew, but a good reminder that excellent delivery and customer service is its own sales funnel.

Also interesting that ~14% were those who previously evaluated but didn't purchase. Higher than I'd expect.

The report is chock-full of other interesting data points. Suggest a read if you have time (it's dense).

– Chris Heysel, CEO

More this week

ChatGPT handles about a fifth as many searches as Google, OpenAI says

Pencil sketch of a curve rising steeply, with OpenAI's logo above it

On November 2, OpenAI's Adam Fry told Big Technology that ChatGPT's search-like queries now run at about a fifth of Google's, by OpenAI's own count. Weekly users rose from 1 billion to 1.2 billion in 60 days.

Our thoughts

Important to be skeptical of internally reported metrics that are self-serving, but interesting if true.

Related to this, last week, Microsoft announced that its free tool, Clarity, now allows deeper AI visitor analytics.

Microsoft also said it's seen weekly agent visits to sites running its Clarity analytics nearly double in three weeks between mid-August and early September, to almost 2 million. A trend we're continuing to watch.

Important to be skeptical of internally reported metrics that are self-serving, but interesting if true.

Related to this, last week, Microsoft announced that its free tool, Clarity, now allows deeper AI visitor analytics.

Microsoft also said it's seen weekly agent visits to sites running its Clarity analytics nearly double in three weeks between mid-August and early September, to almost 2 million. A trend we're continuing to watch.

– Chris Heysel, CEO

AI agents booked 4 Uber rides in Halloween week

An "Available on Uber" sticker on the window of a black car

On its November 4 earnings call, Uber said AI agents ordered 4 rides and 500 Uber Eats orders in Halloween week, out of about 320 million trips and orders. CEO Dara Khosrowshahi called it "very, very early and very limited traction."

Our thoughts

This is an interesting story next to the current bet Meta is making with Muse. Meta gives the agent away free to most people and plans to take a small fee on the purchases agents make.

This number suggests maybe those agent purchases aren't that high yet. Where it will go and if Meta's bet will pay off are things to be seen, but it does remind me of a common expression that "being early and wrong are often indistinguishable".

This is an interesting story next to the current bet Meta is making with Muse. Meta gives the agent away free to most people and plans to take a small fee on the purchases agents make.

This number suggests maybe those agent purchases aren't that high yet. Where it will go and if Meta's bet will pay off are things to be seen, but it does remind me of a common expression that "being early and wrong are often indistinguishable".

– Chris Heysel, CEO

Will the agent cloud dwarf the cloud that SaaS was built on?

Illustration of a robot at a laptop inside a cloud labelled with the parts of the agent cloud: sandboxes, persistent state, inference, router or gateway, RL and training, and compute, storage and network

On November 6, Altimeter's Jamin Ball argued that the agent cloud has to give each of millions of agents its own machine, sometimes for only seconds, so it will outgrow the cloud built for apps.

Our thoughts

An interesting article as more and more SaaS flips to API-based, MCP-based or agent-based pricing and business models.

The trend to watch: how the cost underneath that pricing moves, and whether it brings variable costs into what was historically a more fixed-cost base for SaaS, which drove the historical SaaS margins.

An interesting article as more and more SaaS flips to API-based, MCP-based or agent-based pricing and business models.

The trend to watch: how the cost underneath that pricing moves, and whether it brings variable costs into what was historically a more fixed-cost base for SaaS, which drove the historical SaaS margins.

– Chris Heysel, CEO

GPT-6 Astra raised a small model's math score from 7.9% to 48.1% by editing its weights

Three panels from the MIT post: GPT-6 Astra names the game a network was trained on from its weights alone, fixes a car-control network so it succeeds every time, and lifts Qwen3.5-0.8B's MATH-500 score from 7.9% to 48.1%

MIT researchers reported on November 2 that GPT-6 Astra edited 0.016% of the weights of Qwen3.5-0.8B, a small open model, with no training. The gain came from stopping the model from repeating itself until it ran out of room, and the method couldn't improve Gemma 3 270M.

Our thoughts

Even though this is more general AI news, I think it's very interesting. It was a small model, so the caveats matter.

I think it ties to METR's review of AI research at Anthropic, and how heavily Anthropic redacted it. METR found Anthropic is likely improving its models on well-specified tasks about twice as fast as before, and said faster growth in training compute wasn't the cause.

The trend I'm watching is whether AI capabilities begin to scale detached from compute, in a larger way than in the past.

Even though this is more general AI news, I think it's very interesting. It was a small model, so the caveats matter.

I think it ties to METR's review of AI research at Anthropic, and how heavily Anthropic redacted it. METR found Anthropic is likely improving its models on well-specified tasks about twice as fast as before, and said faster growth in training compute wasn't the cause.

The trend I'm watching is whether AI capabilities begin to scale detached from compute, in a larger way than in the past.

– Chris Heysel, CEO

US first-time venture funds fell from 479 in 2022 to 81 so far this year

Kyle Harrison's chart "The money came back. The new firms didn't.": US venture fundraising indexed to 2016, with capital raised at 214 ($108.5B through Q3 2026) while first-time funds fell to 38 (81 funds through Q3)

Contrary's Kyle Harrison argued on October 31 that venture is going the way of McDonald's, where each rational choice adds up to big funds backing the same bets. Citing PitchBook, he showed that established firms took a record 88.2% of US venture capital raised this year, and AI took 82.7% of deal value.

Our thoughts

Last week I asked whether VC consolidation is speeding up, and why. Harrison's numbers say it is, and his why is incentives, since "nobody gets fired for writing a big check into a big brand."

For founders outside AI, the pool hasn't grown. Non-AI venture has sat at about $100B to $120B a year since 2023. The partners who left big firms this year to start smaller funds may be the likeliest to back something off-consensus.

Last week I asked whether VC consolidation is speeding up, and why. Harrison's numbers say it is, and his why is incentives, since "nobody gets fired for writing a big check into a big brand."

For founders outside AI, the pool hasn't grown. Non-AI venture has sat at about $100B to $120B a year since 2023. The partners who left big firms this year to start smaller funds may be the likeliest to back something off-consensus.

– Chris Heysel, CEO

The Office Buzz

Personal recommendations and thoughts from the team.

Parallel, a web search tool for AI agents

Parallel's banner: its logo and the words "Infrastructure for intelligence on the web" over a pale blue field of ones and zeros

We have no affiliation with parallel.ai. We just use it and like it.

Plain web fetch can be slow and not as complete as you need, depending on how a website's data is structured.

There are lots of tools out there. Highly suggest trying one of them if you do a lot of AI-driven research or web search.

– Chris Heysel, CEO

Newsletter: Last Week in GTM

Get the latest on AI and GTM.
Developed for ourselves, now live for everyone.

Email address

We’ll respect your inbox. Because we don’t like spam either.

Newsletter: Last Week in GTM

Get the latest on AI and GTM.
Developed for ourselves, now live for everyone.

Email address

We’ll respect your inbox. Because we don’t like spam either.

Newsletter:
Last Week in GTM

Get the latest on AI and GTM.
Developed for ourselves, now live for everyone.

Email address

We’ll respect your inbox. Because we don’t like spam either.