We have made writing almost free
There is something wonderfully circular about using AI to turn a fairly simple idea into a 150-page board pack and then using another AI to turn those 150 pages back into three pages for the director. It sounds ridiculous when stated that plainly, but it is a surprisingly believable description of where corporate information is heading.
Generative AI has dramatically reduced the cost of producing material. Analysis that once took a team several days can now be expanded in minutes. More scenarios can be modelled, more commentary added, more background included, more versions created and more appendices attached. Each individual addition can be defended as useful. The difficulty appears when all those useful additions eventually arrive in the same place.
Board packs were already struggling with this before AI. The AICD has long argued that good papers need to balance completeness against information overload, and that supporting material should not be allowed to bury the central issue directors are being asked to consider. More recent commentary has made the problem even clearer: board packs are getting larger without necessarily becoming more useful. (aicd.com.au) (aicd.com.au)
AI does not create that instinct. It simply removes one of the things that used to contain it. Producing another twenty pages once required enough effort that somebody had to believe those pages were worth creating. Now they are nearly free, and when the cost of production approaches zero the temptation is to include everything. We may as well add the analysis. We may as well include the history. We may as well give the board the whole thing and let them decide what matters.
Then we automate the reading
The obvious response is to put AI at the other end of the process. If directors are receiving more material than they can reasonably absorb, give them a tool that can summarise it, compare documents, extract the risks and identify the important passages. There is genuine value in that and the emerging generation of board platforms is moving rapidly in this direction.
But summarising a bad information architecture does not make the architecture good.
A summary is useful precisely because it removes things, and that is also its limitation. If an AI identifies the six most important issues in a large pack, a director can explore those six very efficiently. The seventh issue is more difficult because they may never know it was there. This is why the AICD's commentary on AI and board papers draws an important line between using AI to interrogate information and using it as a substitute for the director's own review and judgement. The technology can help directors find patterns, anomalies and questions. Responsibility for understanding the matter remains with the director. (aicd.com.au)
The more interesting solution is therefore not at the reading end at all. It is deciding what deserves to become board material before it gets there.
Not every useful thing is decision material
Organisations tend to treat the board pack as though all included information has the same status, when in reality it does not. Some material is essential to the decision. Without it, the board could not properly understand what management is recommending or the risks attached to it. Other material supports those conclusions and should absolutely be available if directors want to test them. And then there is background information that is useful, informative and potentially interesting, but not something management believes the board must absorb in order to make the decision.
Those distinctions sound subtle, but they matter. Once everything is simply attached to the pack, the difference between essential and available becomes blurred. A director faced with hundreds of pages has to work out not only what the documents say, but which parts the organisation itself thinks are material.
AI gives us an opportunity to make that structure better rather than simply making the pile easier to search. A board paper could become much more disciplined about containing the decision, the evidence needed to understand it and the risks the board is actually being asked to weigh. Supporting analysis could sit behind that paper, connected and searchable without pretending it carries equal significance. Wider background could remain available to a director who wants it without being presented as mandatory reading merely because somebody thought it might be useful.
That is close to the direction existing governance guidance already points. The AICD's guidance on board packs emphasises relevance, structure and the separation of secondary material from the central narrative. AI simply gives organisations much better tools for making that distinction operational. (aicd.com.au)
The best use of AI may be before the pack exists
Once information is structured properly, AI becomes considerably more interesting. Instead of asking it to tell a director what a 400-page pack says, management can use it to ask whether the 400 pages were necessary in the first place. It can challenge repetition, test whether recommendations are actually supported, identify assumptions that have not been evidenced and surface material that belongs in supporting information rather than in the decision paper itself.
Directors then get a different kind of tool. Rather than using AI simply to compress the material, they can use it to interrogate it. They can trace an assumption back to its source, compare the recommendation with what management said six months earlier, explore the downside analysis or ask where a risk has previously appeared. The technology becomes a way of moving through the evidence rather than avoiding it.
There is a broader lesson in this. AI has removed a lot of the friction that once forced organisations to be selective. That friction was inefficient, but some of the discipline it created was useful. The answer is not to make writing artificially difficult again. It is to recognise that when producing information becomes almost free, deciding what deserves attention becomes more valuable.
The purpose of a board pack was never to demonstrate how much information the organisation could assemble. It was to help the board understand what mattered. AI should make us better at that, not simply better at surviving the volume we created.
