In August I wrote a piece for Reworked about the extraordinary afterlife of Spirit Airlines. The airline had gone bankrupt, leaving behind 100 million emails, 500 million Microsoft Teams messages, and the files, calendars and other records of its working life. Google bid $10 million for the archive. Then micro1, a startup building AI agents designed to work inside simulated companies, made a late offer of $12.5 million.

The airline was dead. Its employees had lost their jobs. But the corporate corpse still had a few organs worth harvesting.

My argument was that an archive holds the residue of work, not the work itself. And that you don’t need to believe the residue is valuable for it to change how people behave. You only need to believe that someone else does.

This week, Diane Coyle, Bennett Professor of Public Policy at Cambridge, published a working paper that arrives at roughly the same place from the opposite direction. I came at it from twenty years of watching people decide what to put in writing. She comes at it from the economics of intangible assets.

Coyle explains why relationships become more valuable as AI makes everything else easier to replicate. Spirit shows what happens when someone decides to strip-mine the evidence of them. And in a particularly elegant piece of economic self-sabotage, the prospect of extraction may make people rather less willing to leave anything worth mining.

Relationships are the moat

Coyle’s paper, The business of business is relationships, opens with a number most comms people will find faintly vindicating. By WIPO’s estimate, intangibles accounted for between 77% and 90% of the enterprise value of large companies in the top ten economies in 2024. Not factories, not stock. Stuff you can’t drop on your foot.

She proposes distinguishing between two kinds of intangible capital.

Knowledge capital is the codified, transferable stuff: software, data, brands, patents, designs. It can be owned, licensed, protected and sold. Its value often depends on being able to prevent other people from using it.

Organisational capital is harder to pin down. Management know-how, internal routines, the relationships that make things happen, and the vast body of knowledge about how the organisation actually works, as opposed to how someone in Transformation thinks it works after spending six months drawing boxes. It’s tacit, contextual and difficult to separate from the organisation that created it.

Her argument is that generative AI will put pressure on the value of codified knowledge while making organisational capital relatively more important.

When everyone has access to the same models, the same automation and the same ability to produce suspiciously familiar PowerPoint decks, competitive advantage has to come from somewhere else.

She models work as a chain of complementary tasks, where the weakest link determines what the whole system can achieve. Making one link spectacularly efficient doesn’t help much if the next one still involves persuading someone in Procurement to reply on Teams.

Who knows whom. Who can get something unstuck. Who understands why a particular process exists, even though nobody has documented the reason since 2014. Who you actually need to speak to in Finance.

These aren’t decorative features of organisational life. They’re part of how economic value gets created.

It’s the most rigorous economic case for internal communication and collaboration I’ve read in years. And it never once uses the phrase “internal communication”.

The afterlife of an airline

Coyle recognises that some organisational knowledge can be digitised and transferred, although much of its value depends on the context in which it was created. AI complicates that distinction because it can learn from digitally recorded activity, acquiring something that looks remarkably like tacit knowledge.

Spirit offers a rather grisly demonstration.

Google’s $10 million bid was selected at auction, although the sale still requires court approval. micro1’s higher late offer has not displaced Google as the selected buyer. And the approval hearing, originally scheduled for 14 October, has now been postponed until 6 November, following objections about employee privacy.

In a letter to Spirit and Google, 121 US lawmakers warned that removing names and other direct identifiers might not be enough to prevent individuals being identified from the records. Unions have also challenged the proposed transfer. Google says personal information will be excluded or de-identified by an independent third party before it receives the data.

There is a particularly revealing distinction in the proceedings. A court-appointed privacy ombudsman has recommended approval following additional safeguards for consumers. Employee privacy, however, fell outside the scope of the assessment.

The passengers have a privacy assessment. The people whose conversations helped run the airline are still fighting over what happens to theirs.

The hearing has been delayed, but the commercial proposition remains: records of an organisation’s working life have attracted substantial bids from companies interested in what AI might learn from them.

That doesn’t mean Spirit’s organisational capital was worth $12.5 million. It means a bidder believed it could extract that much value from the records. Those are different calculations.

An archive might tell you that everyone went to Jane in Finance whenever a procurement decision got complicated. It can’t necessarily tell you why. Was Jane the expert, the person with authority, the person everyone trusted, or simply the only human being who knew how to make SAP behave?

A model might learn to reproduce the pattern without understanding what made it work. Which, come to think of it, is a reasonable description of several corporate reorganisations I’ve witnessed.

This is the distinction that matters. The relationships were valuable because they enabled people to do things together. The records may be valuable because they allow a machine to simulate how those people behaved.

Relational value, recorded value and extractable value are not the same thing. But they overlap, and AI is making the boundaries between them commercially interesting.

The Enron email archive has been used by researchers for more than two decades. The idea that old workplace communications might be useful to outsiders isn’t new. What Spirit makes unusually visible is the market price attached to those communications, and the question of who gets to decide what happens to them.

Which is a problem for everyone who has spent the last decade persuading employees to put more of their working lives into the record.

The bit that goes first

In my Reworked piece I argued that when people suspect their words might become someone else’s asset, the first thing they stop writing down isn’t the decision. It’s the reasoning.

The trade-off. The reservation. The objection that got overruled. The explanation of why an apparently sensible proposal was rejected because somebody tried it three years ago and the results were catastrophic.

You end up with immaculate archives of decisions, stripped of every doubt, disagreement and inconvenient fact that might explain how anyone arrived at them. A corporate memory consisting entirely of the minutes, with none of the meeting.

Read through Coyle’s framework, this is more than a records-management problem. The reasoning is part of what allows organisational knowledge to survive beyond the people who originally developed it.

And when people stop recording it, the reasoning doesn’t necessarily disappear. It goes back into phone calls, coffees, private messages and the heads of people who happened to be in the room. Which is fine until one of them leaves, another gets promoted, and a third decides they’ve explained the same thing to six successive managers and can no longer be bothered.

What disappears is the organisation’s ability to inherit that knowledge.

Access becomes a matter of proximity again: who you know, which meetings you’re invited to, whether you happen to sit next to the person who remembers why the policy says something apparently insane.

Digital working was supposed to make that knowledge more widely available. Instead, we risk rebuilding the old hierarchy with better software.

And there’s a nasty feedback loop here. The more organisations try to extract value from the records of work, the more reasons employees have to keep the interesting parts of work off the record. The archive becomes cleaner, the organisation becomes less capable of learning from itself, and everyone congratulates themselves on their information governance.

Of course, this isn’t an argument for recording everything. Quite the opposite. A permanent, searchable record of every workplace conversation would be an extraordinary resource, provided your ambition is to create the world’s most comprehensive database of things people wish they hadn’t said.

People need room to think aloud, change their minds, disagree and occasionally say something stupid without wondering whether it will become training material for a future product.

The challenge is to preserve the knowledge organisations genuinely need without treating every interaction as something to be harvested.

That requires judgement about what gets recorded, why, for how long and for whose benefit. Not merely another exhortation to work out loud.

Measuring the wrong thing

Coyle ends with a measurement agenda, because economists, like comms professionals, have an unfortunate tendency to want numbers for things that resist being counted.

Her proposed indicators of organisational capital include organisational structures, management practices, employee ratings and investment in technology. They’re a starting point for research, not a definitive framework, and she acknowledges the difficulty of measuring something so contextual.

The trouble with proxies is that they’re wonderfully easy to count, which makes it tempting to forget what you were trying to measure in the first place.

An organogram can tell you how many layers separate the CEO from the warehouse floor. It can’t tell you whether anyone on the warehouse floor trusts a word the CEO says.

We have more sophisticated instruments in the digital workplace: collaboration patterns, organisational network analysis, employee listening. But these have their own limitations. A Teams graph can show you who talks to whom. It can’t tell you whether they’re collaborating brilliantly or engaged in a six-month argument about who owns the spreadsheet.

Interaction isn’t necessarily collaboration. Activity isn’t necessarily value. And a record of communication isn’t necessarily a record of what mattered.

There’s another complication. Measurement changes behaviour, particularly when the people being measured know that their conversations might be analysed, retained or repurposed.

If people become more guarded about what they write, the resulting data may still look comprehensive. There will be messages, meetings, documents and beautifully coloured network diagrams. But the record will increasingly reflect what people are prepared to make visible, rather than how the organisation actually works.

The instrument isn’t necessarily broken. The behaviour it’s observing has changed.

That matters because we are increasingly tempted to treat digital traces as a complete representation of organisational life. AI makes it possible to analyse them at a scale that would have been unimaginable a few years ago. It doesn’t make them a neutral or complete account of reality.

And the more commercial value we attach to those traces, the less neutral the process of producing them becomes.

Who owns the residue?

There’s another question running through Coyle’s paper that Spirit brings into focus: who benefits from organisational capital?

Among the relationships she expects to become more important are those between employers and employees, including the negotiation over how productivity gains from AI are shared.

It’s a pleasingly civilised arrangement, assuming the employees still have jobs, somebody has invited them to the negotiating table, and the table hasn’t already been sold to Google.

At Spirit, the people who produced the archive are no longer working for the airline. The proposed transaction is being negotiated through bankruptcy proceedings, with the potential proceeds forming part of the estate. The people whose conversations make up the archive don’t individually control its sale, and there’s no reason to assume they will personally benefit from the value their records might create for an AI company.

Organisational capital is collectively produced, but the rights to its digital remains are not necessarily collectively held.

This isn’t entirely new. Employers have long controlled substantial amounts of the information produced in the course of work. What has changed is the range of things that information might be used for, and therefore what someone else might be prepared to pay for it.

Spirit makes that imbalance unusually visible. And the privacy objections demonstrate that ownership of an archive does not automatically settle the question of what may lawfully or legitimately be done with everything inside it.

For anyone running internal comms or a digital workplace, this creates an awkward problem.

We’ve spent years trying to persuade people to share what they know, explain their decisions and make work visible. We did so because organisations lose an enormous amount of knowledge when everything happens in private conversations and the heads of people who eventually leave.

We weren’t wrong about that. But we built the systems on an assumption that now looks increasingly quaint: that the record of working life would remain primarily useful to the people doing the work.

So the answer cannot simply be better transparency about retention policies. Telling employees that their conversations might one day be sold to train AI models is admirably candid. It is not necessarily going to encourage them to keep talking.

We need to make deliberate choices about which knowledge deserves a lasting record, how long informal communications should survive, what secondary uses are acceptable, and what meaningful limits apply to the exploitation of workplace data.

Trust isn’t something you can retrofit with an engagement campaign after you’ve designed a system that gives people every reason to keep their mouths shut.

Just because storage is cheap and someone in IT thinks the archive might be useful to an AI model in 2031 doesn’t mean every half-formed thought deserves to be preserved for posterity.

We also need to recognise that employees have legitimate interests in what happens to the records of their working lives, even when those records sit on corporate systems.

Coyle makes a compelling economic case that relationships will become more valuable as AI commoditises knowledge. Spirit shows that the digital remains of those relationships may have a quite separate commercial value. And the people creating both are beginning to notice.

The challenge now isn’t simply to persuade employees to write more things down. It’s to give them good reasons to believe that doing so won’t eventually be used against them, or turned into somebody else’s product.

Because an organisation can own the archive. It can’t compel the candour that made it worth having.