Part 3 of 3, distilled from our May 2026 online event AI in Demand Planning. A 30-minute conversation between Nicolai Harsbo Pedersen (Client Solution Manager at Hakio, 8 years in fashion operations) and Glen Roth (Enterprise Sales Director). This post focuses on the third of the three things they covered: how AI shifts demand planning from reactive firefighting to proactive exception management, by connecting signals across systems that humans tend to read in isolation. 

Watch the full replay here.

Putting out fires before they start

Most planning teams know what their morning looks like. Open a few reports, scan a few dashboards, find the things that broke overnight and start working through the list. Glen framed the question for this section as bluntly as anyone could:

"Can AI help us put out fires before they even happen? Is it best to be more proactive rather than reactive?"

The answer, in Nicolai's view, is yes, but the mechanism is specific. It isn't about the AI predicting more accurately. It's about the AI watching more things at once.

Why humans miss multi-signal patterns

The honest part of this section is the diagnosis. Most planning teams already have all the data they'd need to catch a problem early. What they don't have is a pair of eyes that can watch all of it at once.

"The human brain or the human eye tends to look at one criteria in isolation and then look to the next one and then to the third one."

The pattern shows up differently depending on the size of the brand. In larger organisations, planners are split by category or function, so each person owns a slice of the signal. In smaller brands, the whole demand-to-supply exercise lives in one head, which sounds simpler but means the same person who's supposed to spot patterns is also the one running the day-to-day. Either way, nobody owns the cross-signal view.

What AI actually catches

This is where Nicolai gets specific about value.

"AI can watch all those metrics, all those signals simultaneously and then consolidate them into one signal. This flag, this alert, I think is extremely important. It's really where AI can create a lot of value."

The example he reached for was deliberately undramatic.

"A sell-through could look nice on the surface, but your stock is being consumed fast. Suddenly, it has been completely evaporated. You have a long lead time towards replenishment, so if this is decoupled between different people, you might react too late."

That's the pattern in miniature. Sell-through in one report. Stock cover in another. Lead time in a third. Each looks fine on its own. The combination is a stockout three months from now that nobody flagged because no individual signal triggered it.

The slide behind this part of the conversation showed Hakio's availability report. Not a forecast. Just a view of NOOS products with monthly availability projected forward. Months that look safe in green. Months trending toward yellow. Months in red. The point of the view isn't that it predicts the future better than a planner could. The point is that it shows every product, every month, on one screen. The planner can see the cliff before they're standing on it.

Lead time as a scenario, not a fact

The second concrete case Nicolai walked through is one of the cleanest examples in the whole event of where AI changes the math, not the spreadsheet.

"If we see a delay in what we're purchasing, and we all of a sudden need to take different decisions, what is the impact? What is the scenario of a PO being late? Lost sales. It could be lost sales, stockouts. If we see this quantification of potentially lost sales, what does that translate into? Lost revenue."

Then the trade-off:

"What does that lost revenue compare to when you're talking about maybe shifting modes of transport or speeding up production. If we have the chance to express something home because it's urgent, not that we should do it for the environment, but still, this will happen. This is the reality of the industry. If that cost is less than the lost sales, then you have a direct scenario plan and you can react on this. No calculations needed on your side."

The mechanic isn't sophisticated in concept. It's a comparison: cost of doing nothing versus cost of expediting. What's new is that the comparison can run automatically the moment the delay shows up, on every PO, without a planner sitting down to model it. No calculations needed on the planner's side, in Nicolai's words.

Overstock is just as detectable as stockout

The takeaway slide for this section made one point that's easy to miss:

Overstock is just as detectable as stockout, if you extrapolate early.

Most teams instinctively think of early-warning systems as defending against running out. The same logic works in reverse. Sell-through softer than the forecast assumed, stock cover building: the multi-signal pattern that catches a stockout early catches an overstock early too. Working capital and margin upside, both, on the same plumbing.

‍

Where this is heading: AI as an exception engine

It's worth slowing down on that. The phrase is doing two jobs at once. 

Exception implies the AI doesn't surface everything, only the things that need a human to decide. Engine implies it runs in the background, all the time, on every signal at once. The role of the planner shifts from finding problems to deciding what to do about them.

That maps cleanly to the morning-standup question this section opened with. Once the AI is doing the watching, the human work is the deciding.

‍

Three things to take with you

1. The early warning comes from combining signals, not watching them individually. Sell-through, stock cover and lead time each look fine alone. Combined, they spell out a stockout months in advance.
2. Overstock is just as detectable as stockout, if you extrapolate early. Same plumbing protects margin and working capital, not just availability.
3. Demand planning is moving toward AI as an exception engine. AI watches everything continuously and only surfaces what needs a human to decide. The planner's role becomes deciding, not finding.

This is the third and final post in the series. To hear the full 30-minute conversation in Nicolai and Glen's own words, including the bonus section on the DIY AI dilemma, rewatch the online event here.