By Ryo Kaneko, Director of Innovations, NEC X 

I speak with many startups, and recently I have noticed a common challenge: cold outreach seems to be producing fewer and fewer results.

In the past, sending hundreds or thousands of carefully targeted messages might have generated a response rate of close to 1%. Today, some startups tell me that the response is nearly zero.
There are several possible reasons. One is that AI has made it inexpensive to generate large volumes of seemingly personalized outreach. As inboxes fill with automated messages, recipients have become better at ignoring them.

But I believe something else is also changing: how buyers discover solutions.

Buyers No Longer Wait to Be Educated

When I face a new problem in my own work, I rarely spend time reading newsletters or browsing blogs in the hope of encountering a relevant solution. I begin by asking AI. I describe the problem, explore possible approaches, compare solutions, and then investigate the companies that appear most relevant. Only after narrowing the field might I contact one of them.

In other words, buyers may be becoming more proactive. They no longer need to wait for the right sales email to arrive. They can ask AI for possible solutions at the moment a problem arises.

If this behavior becomes widespread, inbound may be returning—but in a different form.
Instead of finding a company through a search engine and filling out a contact form, customers may increasingly describe their problems to AI and ask it which solutions they should consider.

In the old model, startups searched for customers.
In the emerging model, customers describe their problems to AI—and AI searches for startups.

From SEO to AI-Mediated Discovery

This shift could make SEO and Answer Engine Optimization increasingly important, especially for startups that do not yet have a widely recognized brand.

A startup’s website can no longer be only a digital brochure filled with broad claims such as “transform your business” or “increase productivity with AI.” It must help both humans and AI systems understand precisely:

• Who has the problem
• In which workflow it occurs
• Why existing approaches are insufficient
• How the product changes the workflow
• What inputs, outputs, and results can be expected

This may require creating separate pages for specific industries, roles, problems, and use cases. The objective is not simply to publish more content. It is to provide concrete, useful, and credible answers to the questions prospective customers are likely to ask AI.

An Interesting—but Unproven—Experiment

One startup I work with is experimenting with a particularly ambitious version of this strategy.
It is using AI-generated synthetic data to create demo environments for many different industries and workflows. It then produces videos demonstrating the customer’s problem, how the product addresses it, and what the resulting workflow could look like.
The company is also experimenting with how the content is generated. Its hypothesis is that heavily promotional language is less likely to be cited or recommended by AI systems. It therefore prompts AI to focus on the user’s problem, workflow, and possible solution rather than conventional sales messaging.
The goal is to make the product easier for both prospective customers and AI systems to understand.

However, this approach carries risks.
Large volumes of synthetic content can become repetitive or reduce credibility. A simulated demonstration may imply outcomes that have not been validated with actual customers. Videos may also remain difficult for AI systems to interpret unless they include transcripts, structured explanations, and clear labeling of synthetic data.

Most importantly, we do not yet know whether this strategy will work.

AI systems that retrieve current web content may discover these pages relatively quickly. Influencing what future foundation models know about a company is much less predictable and may require waiting for future model releases—without any guarantee that the content will be included.
For now, this should be considered an experiment, not a proven playbook.

When Anyone Can Build, What Becomes Scarce?

Software development is becoming more automated every day. We are moving toward a world in which much of a product can be generated from a sufficiently clear specification. Synthetic demo data, demo environments, and even product videos can increasingly be generated as well.

If building and demonstrating software become dramatically cheaper, the scarce resource may no longer be the ability to produce software. It may be the ability to understand a customer’s problem deeply enough to determine what should be built in the first place.

Ideas will not be scarce either. AI can generate hundreds of possible solutions in seconds.
The real advantage will come from knowing which assumptions to test, learning from customers faster than others, and identifying the solution that creates meaningful and defensible value.

The bottleneck is shifting from building the solution to identifying the problem worth solving.
When anyone can build, the advantage belongs to those who know what is worth building.