Welcome to Temu BI

Where dashboards are cheap and insights are impossible to find

AI is about to do something extremely dangerous to your analytics program: Make the cost of building dashboards insanely cheap. Last week I built a complete pipeline, warehouse and financial reporting app in under three hours on a personal Claude license. Eighteen months ago that’s two weeks of BI team labor and almost $5,000 all-in, just for the dashboard. Now your boss can build something great looking with a spreadsheet, a prompt and a prayer. So are we in for a new golden age of self-service? Ask anyone who bought a twelve dollar pair of shorts on Temu how that worked out by the third wear. When the cost of something collapses, quality doesn’t rise to meet it. Welcome to Temu BI.

---

Thank you to MotherDuck for sponsoring Super Data Summer on the blog. MotherDuck Dives are the easiest way to build and deploy analytics apps with AI. Check them out here.

---

Why Dashboards Were Expensive

Building a dashboard was really hard. Requirements gathering, data modeling, SQL writing, design, validation, maintenance - it all added up to a ton of human work that could only be done following months of learning. And the tools were expensive - I remember paying $30,000 for a single Cognos administrator license in 2010. The human and financial friction in the system limited volume, and limited volume forces prioritization. In every organization the demand for dashboards far outpaced the supply chain for building them. The downside was obvious - people can’t get the data they need. The upside is only becoming clear now; friction is a quality control mechanism.

Hyper-Temufication

That’s all out the window. Claude and ChatGPT build ‘good enough’ report outputs in minutes from spreadsheets. They aren’t going to win an Iron Viz contest, but they don’t have to. The demand for perfectly formatted, beautiful dashboards rarely came from the business - rather, it came from the practitioners themselves, who took pride in making something elegant. There’s no time for pride in the age of AI.

Like your first $12 pair of good enough shorts, the initial burst of freedom from AI - authored BI is exhilarating for end users, who are used to waiting somewhere from hours to infinity for their data requests to be answered. But soon the problems become apparent. AI can explore data, model it, generate SQL, visualize it and even explain it. But it can’t tell you it’s accurate, or meaningful, or wise. AI builds without discernment.

And it builds fast. Where we used to have a single human authored dashboard, there will soon be dozens or hundreds made by AI. Humanity was like a levee holding back the data flood, and the levee just broke. There’s no patching it back up; the time to build a raft is now.

The Failure Modes of Temu BI

I’ve been dire up to this point, but let me be clear - this change is probably good in the long run, when we’ve ironed out all the kinks. Waiting so long to get data questions answered that they become irrelevant is not a sustainable way to operate; getting the right answers quickly is good; getting a flood of unmanageable AI data slop is bad.

Temu BI will fail in four big ways. Be on the lookout.

Dashboard Sprawl

The systems to create are exploding right now; curation is far behind. This is the classic Excel problem on steroids. When everyone can build impressive outputs via the upload button, but comparatively few understand and can validate what was built, the volume of content explodes and the quality craters. Current incentive structures within organizations have not adjusted to this bounty - walking into a meeting with a good looking visualization carries value so everyone will do it.

False Confidence

Okay so everyone walks into the meeting with their own numbers. That’s not new. But now they also bring an AI generated sycophantic narrative about how their numbers are right (and they are also really good looking and smarter than their rivals too!)

Metric Drift

The classic ‘we have three different definitions of revenue and customer’ is not easy to solve - oftentimes it’s totally valid that different parts of the organization view the same metric through a personalized lens. Data governance programs and data catalogs exist for this reason.

What AI changes is the speed of metric mutation. Even when you’ve established a rock solid, agreed upon definition of a metric it only takes one person letting Claude inject a hallucinated calculation to begin a cascade of bad answers rippling through the organization. The problem is old, AI intensifies it.

Analytics Theater

Everyone wants to be data driven, or at least to claim they are. But a lack of data was often cover for going with your gut - and many decision makers preferred that anyway, data be damned. Now that building analytics is easy, the temptation to generate data assets that justify your intuitions will be impossible to resist. Analytics Theater is when you torture the data, build the fancy charts and put on a big show of being “data driven” when the decision was made before the dashboard was opened.

Resistance Is Futile

Holy crap that all sounds horrible! We have to lock Copilot down and cancel the Claude subscriptions immediately to save the firm from a data disaster!

Of course this is impossible. Be honest, who controls the budget for data tools? Even if it falls under IT on the accounting statement, does the BI team really have veto power over what tools users can access? In my experience, basically never. You can’t shovel the flood back behind the dam.

But you shouldn’t want to. We have a real, longstanding and unsolvable problem in business intelligence and analytics - people want more of our product than we can possibly supply. BI adoption is stuck around 25% and has been for decades, and no amount of drag-and-drop innovation was going to solve that.

For the first time, AI lets our users truly feel like they can do it themselves without us. The dangers are clear, but here’s the opportunity: Embracing AI and enabling end users to build fast, accurately and safely makes you more valuable, not less!

Look at those four failure modes of Temu BI above - those are HUGE problems. Solving them is worth way more than building yet another dashboard nobody reads.

If BI Is Temufied, Become Amazon

The cost of the analytics product - the dashboard - is plummeting to near zero. It’s a terrible business for a human being to be in. But managing the environment that delivers dashboards, curating them, managing context, building AI apps - that’s an opportunity.

Start by learning how to dev with AI now. A $20 a month Claude subscription and a free MotherDuck account are all you need to build analytics apps today. Get the basics down, learn where it succeeds and where it fails. Getting AI to build a ‘good enough’ app is easy; getting it to build the perfect app is hard. Learn perfect then teach others. Start creating the infrastructure to scale AI authored analytics - you literally just need a shared drive and markdown files to begin, nothing else.

Above all don’t wait. The $12 shorts business is commodified, you can’t make money there. But the management and logistics of distributing $12 shorts on demand at scale is incredible. Let AI be the shorts factory outside Hanoi; set your sights on becoming Jeff Bezos.