Once in a while, someone says something, and I think to myself: 'Dang, I wish I would have come up with that!' I had one of those moments when I listened to Yamini Rangan, CEO of HubSpot, deliver her 2026 UNBOUND keynote. She explained that most companies (90%) use AI, but don't see transformational results. In fact, according to HubSpot research (more of this below), only 6% of companies see transformational results.
There were two things in her keynote that I think deserve a deeper dive: first, the fact that popular AI use cases are also the ones with low impact, and second, that "Bad context is worse than no AI." (Honestly, that sentence deserved a🎤drop... and I will dedicate the next article to that. In this article, I want to take a closer look at the data Yamini presented and peel back the layers of what's underneath the negative correlation between popularity and impact of AI use cases, because there is a fascinating reason for why that is the case. But first, if you haven't seen her keynote yet, make sure you watch it below. It is worth every minute.
In HubSpot's UNBOUND Opening Keynote on 16 September 2026, Yamini Rangan shared the findings of a survey of more than 6,000 respondents and analysis of customer usage. Not surprisingly, 90% of the companies said that they used AI. However, only 6% of the companies interviewed saw transformational results. These 6% were 4x more likely to achieve their revenue targets and 3x more likely to hit their efficiency targets.
She also shared that HubSpot analyzed about 50 GTM AI use cases across the three Go-To-Market dimensions: creating demand, winning deals, and delighting customers. For each of the use cases, they measured popularity and impact (meaning how did it affect the company's bottom line, not just activity).
Some of the lowest impact use cases were:
On the other hand, the highest impact use cases were:
However, when we look at what companies actually use AI for (popularity), we notice that they use it for producing content, sending emails, and preparing meetings (which are all low-impact), but rarely use it to analyze campaigns, prioritize pipeline, and analyze customer feedback (all high-impact). In fact, the Scatter Plot Chart of the 50 use cases showed a negative correlation between Popularity and Impact of the AI use cases.
Credit: Screenshot of the HubSpot Spotlight Live from UNBOUND 2026 video.
The question is: what makes them have a low versus a high impact? What do activities such as "producing content, sending emails, and preparing meetings" have in common, versus what do "analyzing campaigns, prioritizing pipelines, and analyzing customer feedback" have in common? And why are some more popular than others? The connective tissue between all of these questions is context. What does your AI know about your business, your customers, your market, and your team?
In all low-impact uses (produce content, send emails, prepare meetings, etc.), AI generates a draft that a human then reads and passes judgment on. Does it need tweaks, and if so, which tweaks does it need, or can it be published/sent? These use cases are measured by the quantity of activity (# of articles published or emails sent), and they have no defined standards. What good looks like is subject to the human's objective opinion. In other words, a mediocre article still counts as done as long as it's published.
Failure is relatively cheap and can be caught by a human easily. A bad email draft gets deleted or rewritten. The expectation is that AI drafts, human rewrites. In other words: the AI decides what good looks like and the human decides if they agree. In addition, they all tolerate thin context because humans compensate for the lack of context, meaning the human must always be in the loop to make up for the thin context. Therefore, the number of humans reviewing will always be the ceiling. Gains here are linear, and impact is limited.
High-impact use cases, such as analyzing campaigns, prioritizing pipeline, and analyzing customer feedback, are different because they are translating well-defined judgment at scale. AI is not deciding what good looks like. It applies the standard, definition, or guardrail the company already decided on without a person in between. Notice how "At-risk deal," "priority lead," "resolved ticket," and "offboarding managed" are each a definition.
These use cases are high impact because judgment must exist as a written standard before the AI can run. Context is not an input that improves the task. Context is the prerequisite to the task. Once codified, the standard runs on every case, so returns scale with case volume, not headcount. Ten thousand leads scored on one decision. These tasks have a checkable right answer (the deal was or was not at risk), so the standard can be measured against outcomes and improved each cycle. This is the compounding curve that the low-impact use cases will never see.
Having rich context means that someone sat down and explicitly wrote down what AI must know about your business, your customers, your processes, your values, your standards, etc. This means they had to draw a line and say yes to one thing and no to another. They defined what your company's version of a Marketing Qualified Lead (MQL) is versus a Sales Qualified Lead (SQL). Who fits your Ideal Customer Profile (ICP) and who doesn't. The list goes on, but you can see that these are necessary, although not easy, decisions.
I would go so far as to say that context is the real reason why high-impact use cases are less popular: You simply cannot run it until the judgment is written down, and most companies never wrote it down. So they picked the twenty use cases that do not require judgment, call it AI adoption, and then complain "that AI doesn't work." And that is exactly the difference between the 6% and everyone else: they have defined their context that a competitor cannot copy.
Now you know why context determines impact, but how does this translate into real-world results? HubSpot shared amazing findings from their customer data across marketing, sales, and customer service.
Credit: Screenshot of the HubSpot Spotlight Live from UNBOUND 2026 video.
Companies with good context achieved:
Who doesn't want to increase their win rate by almost 200%? Well, it makes sense to invest in good context. But here is a harder question: What happens when a company has bad context? Whether the culprit is outdated information on your website, AI filling in the gaps because it cannot find any information, or your employees are all working on different versions of the truth. AI using bad context is often worse than having no AI at all. But that's the next article. Stay tuned.
And if you want to build an authentic AI advantage through context, pre-order my upcoming book, Scattered to Scaled, to learn how you build and maintain your very own Context Layer. Then come back and claim $148 worth of exclusive pre-order bonuses and get access to the manuscript immediately.
Sources: