This post first appeared in The Upside-Down Funnel, our newsletter on Substack.
The last few weeks have been all about data for me. I ran the same 102 buying signals across 10 companies to see their impact, ran a wide analysis of how Upside enhanced GTM data and I even had my Dot count the ads in my LinkedIn feed every day for a week just to see if I am imagining my feed looking a lot staler.
I made an image recently about my feelings about AI - somewhere between dread and wonder, wonder at what AI can do with data and workflows, and dread at what this means for society, and even our critical thinking and anxiety levels. At the same time, working with multiple agents has started to feel like having a real team that never sleeps, so I decided to give the team a performance review to help with my own mental sanity: ranked them S to F by ROI - the impact on our pipeline vs the amount of human effort it takes.
What we’re trying: Ranking my AI workflows from S to F, embedding Upside Miniapps in Salesforce, testing what’s actually in our X and LinkedIn feeds + OpenAI business pages
What we’re pondering: Why the same buying signals don’t hold from one company to the next + how messy GTM data really is
Worth your attention: CRM conversion events for ChatGPT, Google call conversions + the cost of running GTM agents
Stranger than fiction: People holding ads + the intelligence explosion forecast is back to 2027
What we’re trying
My AI workflows, ranked
Here is the full performance review - every workflow I run with agents right now, with the model, harness and MCP behind each one. Research reports, the children’s book, our website and webinars got an S. The arrows mean I switched from one tool to another and I also wrote a full breakdown of all my workflows if you want more detail.
Is LinkedIn content stale?
I asked my Dot to read the first 100 posts in my LinkedIn and X feeds every day for a week and log what I was actually getting. On LinkedIn, one in four posts was an ad, every single day. About half were less than a day old, and on the worst day I got twice as many posts older than a day (37) as fresh ones (18). LinkedIn kept serving me posts up to three weeks old; the oldest thing X showed me was under two days. On X (four of the same days), 13% were ads (but I have a premium subscription which limits them) and 82% were from the last 24 hours.
Upside inside Salesforce
This one is simple and I love it: Upside Miniapps and Campaign scorecards now work inside Salesforce record pages, so the team sees interactive analysis where they already work instead of opening one more tab - the possibilities are endless, from showing a full interactive account story on every account, to campaign metrics and ROI on campaigns, to each contact’s history, to account health and more.
OpenAI business pages
OpenAI launched Businesses in ChatGPT on Tuesday, so of course we went to claim Upside. You find your company in Business Hub, prove you actually work there (business email or website domain, and then a full ID check before you can publish anything), and you get a blue check on your ChatGPT profile plus a description, logo and category that you control.
I’m pretty excited to see how the insights tab shapes up: mentions, website clicks and the topics where ChatGPT cited you, straight from OpenAI, for the last 28 days. We have been piecing AEO impact together from sales calls, forms and URL parameters, so even a little first-party data feels like a gift. Two things to know before you get too excited (like I did): a mention only counts when ChatGPT references your website, so your brand name alone doesn’t count, and OpenAI is very clear that claiming your page does not move you up in the answers.
What we’re pondering
Buying signals don’t travel
I went into buying signals research with a hunch that these signals are not the same for every company, and it turns out they are less consistent across companies than I imagined. We ran the same 102 signals at 10 B2B companies, across 19,459 new-business deals and 205,036 contacts. Even a demo request (the signal everyone trusts!) signaled 15.2 times the opportunity rate at one company but no lift at all at another.
Another interesting finding, across 17 contact behaviors, 74% of the company tests showed a clear link to becoming an opportunity, but only 7% predicted an increase in winning likelihood. What did help with wins was what the buying group does after Discovery - if the buyer goes silent for 14 days in the first month, win rates were lower at 7 of 8 companies (about a third of the win rate at the typical company).
How messy is messy for GTM data
Everyone says their GTM data is messy, so we went and measured how messy. For our messy GTM data report we looked at 29,367 new-business deals at 12 customers. We found that 2/3 of the buying group on won deals is missing from the CRM opportunity, 13% of one-to-one emails don’t exist in the CRM as a touchpoint (we extract them from forwards and replies), and 11% of activity records were extra copies of something already logged.
Worth your attention
HubSpot can now send conversions back to ChatGPT. Its November 2 roundup adds a ChatGPT pixel that sends lifecycle-stage changes and form submissions back to ChatGPT (Marketing Hub Starter and above). This is big, and paired with the business pages above it shows OpenAI making real strides toward monetizing ChatGPT.
Google adds call conversions and better diagnostics. The November 4 Data Manager API release adds Google Ads call conversions and finally tells you which uploaded records had errors or warnings.
Measure the cost of a usable result. In Cloudera’s survey of 1,500 IT leaders, 89% said rising costs made them limit or redesign AI use cases, but only 19% continuously monitor token and inference costs. For GTM I think we are going to start seeing this affect limits and use cases more in the short term.
GTME jobs and trends
This week’s pattern on the GTME tracker: GTM engineering is evolving to have a career ladder. Tremendous, Nooks and 6sense are hiring managers or directors to run these teams. Two cool roles:
Tremendous, GTM Engineering Manager: Remote, US. $190K to $230K, plus equity.
LiveKit, Staff GTM Systems Engineer: Remote, US. $216K to $292K, plus equity. Owns the architecture across Salesforce, HubSpot, enrichment and billing.
Stranger than fiction
For everyone worried AI will take our jobs, Silicon Valley has a suggestion: hold an ad for a software company. My co-founder Dan took this photo in South Park of someone holding an ad for a startup: “pissed at posthog? try sazabi.com now in g.a.”
And then one that is squarely on the dread side of dread and wonder: the AI Futures Project (the team behind AI 2027) published an update on November 10 called Looks Like the Intelligence Explosion Will Happen in 2027 After All. All five authors now expect an AI that out-researches the best human research teams to arrive in 2027, and they put a 35 to 65% chance on AI that can automate all white-collar work arriving before 2028. I ranked my AI workflows this week, by next year they will be ranking me.