When businesses talk about using AI to improve operational efficiency, reporting is an obvious target.
A team spends three hours every Monday collecting numbers, comparing last week with the week before, updating a spreadsheet and writing a summary. Introduce AI, reduce that to 30 minutes, and the ROI is easy to calculate.
But after building AI analytics systems, I've come to think that's solving the less interesting part of the problem.
The bigger inefficiency isn't how long the report takes to produce.
It's that somebody has to remember to produce it in the first place.
Most reporting workflows start with a human
Traditional business intelligence is surprisingly dependent on human initiation.
Someone opens a dashboard because Monday is reporting day. Someone notices conversion is down and decides to investigate. Someone remembers that a particular product category performed unusually well last quarter and checks whether it's happening again.
Once the investigation begins, modern analytics tools are very good at helping.
But what happens when nobody starts it?
This is particularly problematic in ecommerce, where there can be thousands of products and enormous numbers of customer interactions.
Overall revenue might be stable while something important is changing underneath it. Customers could still be viewing a popular product but adding it to their carts less frequently. Returning customers might suddenly be converting differently. A new product could be gaining traction much faster than expected.
The information is there.
Nobody happened to look.
We discovered this by automating our own reporting
At Stormly, we've been building AI-based analytics that can investigate behavioral, product and ecommerce data.
One use case is straightforward: instead of manually preparing a weekly analysis, the system can run it automatically and send the findings by email every Monday.
Initially, I thought about this as a conventional productivity improvement.
If an analyst spends several hours creating a report and AI can perform much of that work automatically, we've saved those hours.
But once we started using this type of workflow ourselves, I realized the bigger advantage was different.
We had removed the requirement that somebody remember to look.
The analysis could happen whether someone had put “check analytics” on their calendar or not.
That's a much more interesting form of automation.
The next step is deciding what deserves investigation
AI makes it possible to go further than scheduled reporting.
Suppose sales decline.
A traditional automated report can tell you that sales are down.
An AI system can start investigating why.
It might determine whether traffic or conversion changed, identify which products contributed most to the decline, compare customer segments and look at behavioral changes. If an existing analysis doesn't answer the question, it can generate SQL to investigate something more specific.
External context matters too.
If sales of a particular category have fallen internally, but market interest in that category is also declining, that's very different from seeing your sales fall while the wider market is growing.
The objective isn't to generate a longer report.
It's to decide:
Is there something here that a person needs to know about?
Operational efficiency is also about attention
This has changed how I think about efficiency.
We normally measure automation by counting minutes.
A task took 60 minutes. Now it takes 10. We saved 50 minutes.
That's useful, but it ignores another scarce resource inside organizations: attention.
Managers can't continuously watch every metric, customer segment, product and operational process. Analysts can't investigate every possible combination of data.
That means businesses are constantly making an implicit decision about what not to examine.
AI can change that equation because machines can perform the repetitive looking while people reserve their attention for judgment.
Instead of asking an ecommerce manager to monitor 5,000 products, let the system monitor 5,000 products and bring the five unusual situations to the manager.
The human hasn't been removed from the decision.
Their attention has been moved to the part where it creates more value.
Don't automate a bad workflow
There's an important lesson here for companies introducing AI.
If your current process is:
Collect data → build report → find interesting change → investigate → decide what to do
the obvious automation project is making the first two steps faster.
I'd challenge that.
Ask whether the workflow could instead become:
Continuously analyze → identify what deserves investigation → investigate → bring evidence to a human → decide what to do
That's not simply a faster version of the old process. It's a different process.
Three questions can help identify opportunities:
- What does someone have to remember to check regularly?
- Which problems are discovered only because someone happened to look at the right metric?
- Where could AI bring evidence to a decision-maker without making the consequential decision itself?
Those are often better automation opportunities than the most time-consuming task on a process map.
We've spent years trying to make dashboards easier to build and reports faster to produce.
AI gives us an opportunity to rethink the workflow more fundamentally.
Don't just automate producing the report. Automate deciding when there's something worth reporting.
