The Intelligence Grind: Why Tech Scouting Runs on Chrome Tabs and Luck

In 2012, Kodak filed for bankruptcy. This is not a surprising fact — it has been told so many times it has become a parable, a shorthand for missing the future. What’s less discussed is that Kodak didn’t miss digital photography through ignorance. Their own engineers had invented one of the first digital cameras in 1975. The intelligence existed. The problem was that it lived in the wrong place, in the wrong format, connected to nothing that mattered. By the time it reached anyone with the authority and context to act on it, a decade had passed and the window had closed.

This is still how most technology intelligence fails. Not through absence of information — there’s more of that than anyone can use — but through structure. Or rather, the lack of it.

Meet Marta. She leads technology intelligence for a mid-sized industrial manufacturer, a company serious enough about innovation to have created her role, but not yet serious enough to have given her the tools to do it properly. Her mandate is to track what’s happening in battery technology, in advanced materials, and in the adjacent chemistry domains that might reshape both within five to ten years. She is skilled. She has a team of three. She has a budget. She does not have infrastructure.

What she has, instead, is a system she built herself: a nest of RSS feeds, a shared Google Drive stuffed with PDFs and half-finished Excel spreadsheets, a Notion board that was optimistic in its conception and chaotic in its execution. She has a group chat with two researchers at a university who sometimes flag things they think she’d find interesting. She has a Slack channel that she really hopes someone reads. She sometimes uses her personal ChatGPT account for low risk questions (but remains skeptical of its confident outputs), she stopped using the company’s in-house AI tool because it was faster to search herself. She has, most importantly, three people whose brains hold the majority of what her function actually knows — connections, context, intuitions built up over years — which means that when one of them leaves, roughly a third of the company’s technology intelligence walks out the door with them, even if they “brain dump” into Notion for their last two weeks.

Knowledge workers already spend close to two hours a day, on average, just searching for information¹ — and that’s for information they already know exists. Marta’s team is doing something harder: searching for signals they don’t know to look for yet, in domains they can only partially map, using tools built for a different problem entirely.

They manage. The best teams always manage. They develop workarounds, rituals, shared intuitions that substitute for proper structure. But managing is not the same as having a system that scales, that doesn’t break when someone goes on parental leave, that can surface a weak signal from a materials science preprint and connect it — without a human having to hold both things in their head simultaneously — to a trend in a patent filing from a company in a completely different sector.

This is the gap. And Chrome tabs are not going to close it. And there is a wave coming through the gap. A wave of new information, new papers, new science and new breakthroughs. How are they ever going to keep up?

The conventional response to this problem has been one of two things. The first is generative AI: throw your PDFs at a large language model, ask it to summarise, synthesise, report. This is better than nothing. It’s faster than reading everything yourself. But predicting the next token based on whatever document you happened to upload is not intelligence — it’s retrieval with a confident tone. It will tell you what the document says. It will not tell you what the document means in the context of something it hasn’t read. It will not catch the cross-domain signal. It will not warn you when its synthesis is six months out of date. And it will do all of this with the same assured fluency whether it’s right or wrong.

The second response is bespoke tooling - you RAG it. The best teams — the ones who really care — eventually build something. A custom pipeline, a structured knowledge base, an internal tool that actually does what the commercial options don’t. Technology scouting draws on the broadest possible data landscape: global patent filings, academic preprints, conference proceedings, startup funding signals² — and the best teams find ways to pull from all of it into something coherent. But now they have a software product to maintain. When the engineer who built it leaves, the technical debt walks out with her. The insight is right. The execution is unsustainable.

The tools that exist at scale — Gartner reports, analyst subscriptions, commercial intelligence platforms — are built for a different buyer. They’re designed for fast-moving, software-adjacent industries where the landscape shifts quarterly and the vocabulary is relatively standardised. Industrial technology doesn’t work like that. The replacement cycles are long. The standards are domain-specific and frequently opaque. The failure isn’t in technology awareness or research capability — it’s in scouting systems that track obvious technologies while missing the weak signals and emerging patterns³ that determine where an industry is actually going.

What Marta needs isn’t more content, more tabs and more tools to maintain. She needs structure that persists, that connects, and that is honest about what it doesn’t know.

Here’s the argument in its plainest form: technology intelligence should be deterministic in method and Bayesian in output.

The first part means that the process of ingesting, structuring, and connecting information should be systematic and reproducible. Not dependent on who happens to be in the office. Not lost when someone leaves. Not siloed by domain because that’s how the data happened to arrive. You build a knowledge graph — a structured map of entities, relationships, and evidence — from primary sources: academic papers, patent filings, technical specifications. The graph is the asset. It accumulates. It connects things that weren’t explicitly connected in any individual document. It can be interrogated, versioned, audited.

Research from Youn et al. (2015) showed that most breakthrough innovations emerge not from novel discoveries but from novel combinations of existing knowledge — recombinations across domain boundaries that weren’t obvious until someone held both pieces at once. This is exactly what a well-structured cross-domain knowledge graph does systematically, what Marta’s spreadsheets do accidentally on a good day, and what most enterprise intelligence tools don’t do at all.

The second part — Bayesian in output — means being explicit about uncertainty rather than hiding it. A technology pathway assessment is not a prediction. It’s a probability distribution over possible futures, updated as evidence arrives. This is more useful than a confident-sounding analyst report, because it tells you where the uncertainty actually lives, and therefore where to focus your attention and your hedging. When Marta presents to her leadership team, she shouldn’t be saying “solid-state batteries will reach cost parity by 2028.” She should be showing them the probability curve, explaining what would shift it, and telling them which signals to watch. That’s a different conversation. It’s also a more honest one.

Back to Marta. Six months after her team adopts a structured approach — a knowledge graph built from patent filings and preprints, maintained systematically rather than ad hoc — something changes. Not dramatically. Not overnight. But a pattern emerges in the graph that none of them had flagged consciously: a cluster of activity in ceramic manufacturing techniques, coming from a domain none of them were primarily watching, that intersected with materials approaches they were tracking in solid-state battery development. The connection existed in the literature. It just had never been connected, because the literature is vast and human attention is finite and the two domains didn’t share a conference or a citation network.

That signal is now visible. Marta’s team debates its significance, interrogates the evidence, updates their probability estimates. They bring a recommendation to their R&D director that is specific, hedged appropriately, and traceable back to primary sources. The director can follow the chain of reasoning. She can challenge it. She can take it to the investment committee with confidence, not because the outcome is certain, but because the process that produced it is sound.

Strategic foresight, ecosystem engagement, and cultural adaptability⁴ are what the post-mortems always identify as the things Nokia and Kodak lacked. But foresight isn’t a personality trait. It’s an output of a process. And right now, for most organisations trying to navigate industrial technology change, that process runs on people’s heads, scrappy drives, and a lot of Chrome tabs.

That’s not a people problem. It’s a tooling problem. And it’s solvable.

References

ArticleCube — cites research that knowledge workers spend close to two hours a day, on average, searching for information they already know exists. Source: ArticleCube. ↩
PatSnap, “Technology scouting vs competitive intelligence” — on the breadth of data a scouting function must draw on: global patent filings, academic preprints, conference proceedings and startup funding signals. Source: PatSnap. ↩
Fragments, “Technology scouting” glossary — the failure mode of tracking obvious technologies while missing the weak signals and emerging patterns that decide where an industry is actually heading. Source: Fragments. ↩
“Top Business Failures: A Comparative Case Study of Nokia and Kodak” (Academia.edu) — its post-mortem identifies strategic foresight, ecosystem engagement and cultural adaptability as what both firms lacked. Source: Academia.edu. ↩