If you’ve ever sat through a demo for a Business Intelligence (BI) tool, you know the story. A key metric on a dashboard suddenly dips. An "intrepid analyst" dives in, slicing and dicing the data with a few clicks. They join a few tables, write a few queries, and—voila!—they uncover a previously unimagined insight that saves the day.
As Benn Stancil, founder of the BI tool Mode, explained in a recent talk, "This sells! This story actually really works." It’s the narrative that built a multi-billion dollar industry. But Stancil, who has given this demo hundreds of times, reveals the conflict at the heart of the BI world: the story sells, but for most companies, it doesn't actually work.
If this narrative were true, why are so many teams unhappy with their BI tools? Why do our dashboards become "trashboards" that nobody uses? The answer, according to Stancil, is that the entire "intrepid analyst" fantasy is built on a lie. Not a malicious lie, but a foundational misunderstanding of the data most of us actually have. It’s the Big Data Lie.
The Myth of "Data is the New Oil"
To understand why the story fails, we have to go back to the early 2010s, when the "big data" hype train left the station at full speed. A series of iconic stories cemented a powerful idea in our collective consciousness: that data, like oil, is a raw material just waiting to be drilled into to produce immense value. We heard how Target’s data science team used purchasing data to predict a teenage girl’s pregnancy, suggesting data had prophetic powers. We saw it in Moneyball, where the Oakland Athletic's used decades of historical baseball data to build a winning team on a shoestring budget. We witnessed it when Nate Silver, hailed as a "witch," predicted the 2012 US election with stunning accuracy by analyzing millions of polling records. And we learned it from Facebook, which data-scienced its way to the magic "7 friends in 10 days" formula for growth.
These examples, and the "Data Scientist: Sexiest Job of the 21st Century" headlines that followed, created a powerful belief system. Stancil summarizes it perfectly: "A lot of people came to very much believe that data just contains value... And all it takes is us to have the right tools to get that insight out." This belief spawned a generation of tools promising to "unlock the power of your data." The problem? The premise was flawed for most of us from the start.
Target had $73 billion in revenue. The Moneyball team had nearly 12 million historical at-bats. Nate Silver had 650 million votes. Facebook had over a billion users.
As Stancil states bluntly, "This is big data." Most of us don't have that. We have a few thousand customers, not a billion users. Our charts don't look like the smooth, predictable curves from a dataset of millions. They look like this:
When faced with a chart like this, you don't "slice and dice." You don't "drill down." As Stancil hilariously puts it, "You squint at it and you're like, it's up-ish." This is the disconnect. We were promised a treasure map, but we got a squiggly line and a shrug.
So if the heroic "big data" playbook is a fantasy for most of us, what are we supposed to do? If our reality is more "up-ish" than insightful, it's clear we need a different approach—a new playbook designed not for finding treasure in petabytes, but for finding meaning in ambiguity.
A New Playbook for the "Small Data" Reality
If the old playbook is broken, what does the new one look like? Stancil proposes a new set of principles for finding value in the "up-ish" reality that most of us live in. This new playbook shifts the focus from heroic exploration to pragmatic interpretation.
Principle 1: Shift from Exploration to Interpretation
"The hard part is not creating this chart," Stancil argues. "The hard part is interpreting it."
Most BI tools are built for exploration. They give you endless options to filter, pivot, and visualize. But when your data is sparse, these features don't lead to clarity; they just create more confusing charts. The real bottleneck isn't getting the data; it's figuring out what, if anything, it means. Stancil's insight is that "interpretation of data is often a lot harder than exploration."
This is where speed and interactivity become critical. To interpret an ambiguous chart, you need to form and test hypotheses rapidly. Is this dip because of the holiday? Let me pull last year's data. Is it a specific user segment? Let me filter by plan type. Is it a bug? Let me look at error logs from the same period.
If each query takes minutes to run, you lose your train of thought. The friction of waiting kills the iterative cycle of questioning that is essential for interpretation. When you can test ideas as fast as you can think of them, you shrink the gap between question and answer, making the hard work of interpretation just a little bit easier.
Principle 2: Embrace Unscalable Work
This might sound like heresy to data professionals, but it’s Stancil’s most powerful point. He tells the story of a friend tasked with analyzing the sentiment of articles on a specific topic. She started building a complex AI model, only to realize there were just seven articles.
"Why am I building a tool to look at seven articles?" Stancil asks. "Go read them." It takes 20 minutes and yields a far richer understanding than any model could. This "unscalable" approach is incredibly effective for customer data. Instead of trying to find trends across thousands of users in a noisy dataset, go look at the raw activity of a single user.
With a tool like MotherDuck, you don't need a complex pipeline to do this. You can query your raw event data directly to "read the story" of an individual user's journey. For example, let's say you want to understand what your most active user from the last week was actually doing.
Copy code
-- Stancil's point: Sometimes the best insight comes from looking at one customer.
-- With MotherDuck, you don't need a complex pipeline to do this.
-- Just query your raw data directly.
-- First, find our most active user this week
WITH user_activity AS (
SELECT
user_id,
COUNT(event_id) AS event_count
FROM events
WHERE event_timestamp >= NOW() - INTERVAL '7 day'
GROUP BY 1
ORDER BY 2 DESC
LIMIT 1
)
-- Now, let's pull their entire event stream to "read the story" of their session
SELECT
e.event_timestamp,
e.event_type,
e.properties
FROM events e
JOIN user_activity ua ON e.user_id = ua.user_id
ORDER BY e.event_timestamp;
The result of this query isn't a high-level chart; it's a narrative. You can see every click, every page view, every action this power user took, step-by-step. This is the "unscalable" insight Stancil talks about. It doesn't tell you what all users are doing, but it gives you a deep, qualitative understanding of what an engaged user's journey looks like. That's often far more valuable than another "up-ish" chart.
Principle 3: Honestly Assess Your Data's Scale
Stancil's final piece of advice is to be honest about the scale of your data. Don't use a "big data" sledgehammer for a "small data" nail. Big data problems are real, and they require big data tools like Snowflake or BigQuery. If you're managing petabytes of data for a global enterprise, those are the right tools for the job.
But MotherDuck and DuckDB were built for the other 99% of us. We excel in the vast space below that massive threshold, where most companies operate and where the challenges are different. It's not about wrangling petabytes; it's about getting fast, reliable insights from datasets that fit on your laptop or a modest server. It's about using the right tool for the job.
Conclusion: Embrace Your "Up-ish" Data
The "big data" dream set an unrealistic expectation for many of us. We were told our data was a gold mine, and we felt like failures when we couldn't find the gold. Benn Stancil's message is a liberating one: your data isn't the problem. The "small data" reality isn't a failure; it just requires a different, more pragmatic approach.
Stop chasing the "intrepid analyst" fantasy and start embracing the messy, ambiguous, "up-ish" world you actually live in. Shift your focus from exploration to interpretation, embrace the unscalable work of looking at individual examples, and choose tools built for the scale of data you actually have. The best way to begin is to try a more direct, interpretation-focused approach yourself. You can start with a local DuckDB instance or sign up for a free MotherDuck account and see what stories your "small data" can tell.
Transcript
0:00[Music]
0:15uh so I'm going to talk about this uh how we thought we had big data and we built everything planning for big data and it turns out we don't actually have big data and while that's nice and fun and seems chill it's actually a big problem uh and I'm asking everybody here to please help because I don't know what
0:28to do um okay so uh by the way my background I am one of the founders of a company called mode it was a bi tool acquired by thoughtspot uh thoughtspot was also a bi tool at thought spot I went from being the CTO of mode to the field CTO at thoughtspot I don't really know what that is but I was like hanging
0:43out in the field I guess um thought spot kind of thought I was like me uh and so then I went to being exploring something new uh or an angel inv angel investor or adviser or an independent contractor or whatever uh I also like yell at the internet uh this is kind of the general Vibe of the the
1:03blog um more recently I have taken a job uh for the Harris walls campaign uh if you ever a chance to go to Wilmington Delaware don't um but that is where I live now I have to make a a very important caveat here uh nothing I say has anything to do with any of this uh I am not kamla Harris she did not endorse
1:21this message uh we all exist in the context of where we came from this talk does not it fell out of a coconut tree the VIN diagram of me and this talk and Kamala looks like this uh nothing I say here has anything to do with any of this campaign so please uh this is just me talking as
1:38the founder and CTO of mode though really it's me talking as a sales rep of mode um because what I want to talk about is like bi demos so I've given a lot of demos uh of bi products of mode this is what mode looks like of thoughts spot this is what thoughtspot looks like and we give these
1:54sort of demos there are always two things that like I did in these demos one is you say you have bad Wi-Fi uh like a trick for demos is always be like Wi-Fi sucks and then if something goes wrong oh it's that bad Wi-Fi um the second thing is they all kind of followed the same script and so this is
2:09actually a real script that I found on an old Google Drive thing of a mode demo uh and I'll walk through some of the like lines from this actual deck um that was how we gave these demos and so it' always be a story like this it'd be like all right let's suppose we have some hypothetical company say it's like
2:22similar to LinkedIn or whatever there are product dashboards for these sorts of stuff we'd show some product dashboard uh and then we'd show this little like blip and like what's this this is like a number that looks bad what's going on and so you tell a little bit of a story about like oh our sales team noticed that something's wrong we
2:36want to look into that maybe it's something to do with like uploading pictures or whatever madeup feature you'd write some queries You' do some work and you say okay like I'm pulling a bunch of event data that includes these various actions about users and stuff like that um and then you would show like some more data and then you'd say
2:50oh look we found something interesting people aren't uploading pictures like here some sort of insight this is our least used feature and you'd kind of keep digging into it you'd iterate through this more and more like we got to dig deeper um you'd show some more like dashboards and drill downs and stuff like that you'd ask like does this
3:04apply to our biggest customers you'd build some sort of model to assess it you'd show the model and you'd be like look here's some interesting stuff that I found uh this is a good place to start solving this problem and then you'd say like okay great now let's go build a report and dashboard on top of this
3:16thing and and here's your dashboard and like this is the Arc of every bi demo that you ever give where basically there is a problem you kind of find this problem you're like what's going wrong some like Intrepid analyst comes along shout out to Jonah Hill from Moneyball and like finds the solution to this problem in the data by exploring and
3:33like spelunking their way through all of this hard stuff and they turn things around it's like that is the Arc of all of these demos and so there was a company that was a mode competitor uh called Periscope they like existed from 2012 to 2020 or so uh and they had this bit on their website that always kind of
3:49resonated with me that told this story pretty well so this was like on their about page and it said pariscope data was built by a small team of hackers working out of a San Warehouse in San Francisco blah blah blah uh and that moment when blip in the data makes you say wait a minute and I think like for a
4:03lot of data people this kind of feels right like this is the sort of thing that we're chasing and so it's like all right we have data there is some sort of blip uh that gives us a clue of what to look at we like slice it and we dice it and then we drill into it and then like
4:15voila there's insight and like this is very much a story that we all like to tell um and so if you do this enough if you like give enough of these demos you will notice two things uh and so I've done a bunch of these demos if you do them when you're like old then there are two things that come out of this one is
4:32it works that like this sells this story actually really works mode had a bunch of customers that were sold in exactly this way Periscope had according to their website a thousand companies that did the same thing that were probably sold in this way bigger bi tools so the powerbi which is much bigger than both mode and Periscope says roughly the same
4:49thing of like quickly find meaningful insights within your data Tableau also a very big bi company similar story of like discover and share insights that can change the your business in the world so like this story sells this is like the key story behind most of the ways that bi tools sell is this kind of you're going to find stuff in your data
5:06there's like Insight in there and if only you had the right tool to go drill into it you'd find all this stuff the other thing you come to learn though if you do this bunch is it doesn't really work that like the story works and the story sells but you end up with a lot of this sort of stuff uh so this is like an
5:22old meme from I guess a few years ago but like trash boards they're dashboards that nobody uses Lindsay was just talking about this there's a bunch of these things out there that're like you build these tools and nobody actually quite uses them this way that the story sounds nice people buy the story but like what actually exists isn't the
5:37story like that's not the thing that actually happens and so somebody wrote this like substack uh Britney Davis called the revolving door of bi it has a line in it that says despite the high upfront investment costs companies continue to swap out their bi tools every two to three years I have no idea if this is true but like it sort of
5:53rings true to me uh and so I put it up here but like it feels like the way this works and it's one of the reasons why there's kind of constant L new bi tools coming up is because you can always kind of sell someone who had a failed version of that that narrative they had that narrative they bought it because they
6:07were going to find all this stuff and then it didn't actually work out and so they go looking for a new bi tool because their existing one didn't live up to the to the sales pitch and so this story again of like the intrepret analyst that finds all of these things kind of only exists in movies uh to you
6:21know like Spock and Margin Call um like finding this stuff like this isn't how it actually happens this isn't the way that bi tools are typically used which is if you think about it it's like a little bit weird that we've been trying this forever uh like we've been trying to do this forever bi tools have been around
6:39for a really long time they've been selling this story for a really long time it still doesn't really work like we've been trying to do this for 20 30 years and we still don't actually have a thing that sells this story the way we think it does and also people still buy it uh which is kind of weird too and so
6:54all this is like I don't know that's strange uh why is this happening and why haven't we figured this out I think there's kind of an answer to it uh to all of these qu well the other questions uh and it comes actually from from Minnesota not anything to do with Tim Walls actually uh but there's a way to
7:10answer this that starts with Target um so Target's based in in Minnesota so if anybody remembers like this was 12 years ago back in 2012 there was this very popular article in the New York Times that was this um this is about Target it's about how Target's data science team like figured out a bunch of stuff analyzing all of their big data to help
7:28them make better decisions and this was from from February 16th 2012 the story like the Crux of the story was they figured out like Target figured out that a teenage girl was pregnant before her dad did they started sending like flyers and stuff to the house to be like here are various pregnancy products and diapers and those
7:45sorts of things and so the dad got all upset because he was like why are you sending these things my daughter she's 17 and then like six months later sending a letter being like I'm sorry I got mad it turns out she was pregnant how did you know that and I didn't and so this sort of story like made this
7:59idea really popular that data had this kind of prophetic powers that like if we had enough data we could predict the future and and the target story actually turned out to be like sort of Hal true but that didn't really matter at some point we developed this idea that this is like what data was capable of and
8:14there were a bunch of things at this time that were basically telling the same story so Moneyball which is about like using data to to make sort of better sports teams this movie came out in 2011 the actual sort of practice was was a good bit before that but it became very popular around the same time um
8:30Nate silver sort of peaked uh around this time there was a website called is Nate Silvera wit.com uh for the 2012 election this was what it looked like in 2012 on November 7th 2012 Nate silver probably a witch because he predicted all the all the states now is Nate Sil a witch is like a not safe for work site
8:46it's weird don't go to it um I don't know what happened there uh this is all from like the internet archive because the other one's not good um so this was in 2012 in 2013 there were talks like this this is chth like Facebook exec SL Spa scam artist podcaster um in 2013 they Facebook popularize this story about like we
9:09figured out that if you have seven friends in 10 days then Facebook's going be super popular and like they data science their way to growth um there was this article that Lindsay referenced this is from 2012 about how data scientists are the sexiest jobs of the 21st century um and we had sort of phrases like this of like data is the
9:24new oil uh also from from 2012 and so all this kind of kind of built this idea and it's like very clear what the idea here is which is like okay you've got oil this kind of raw material if you drill into it then like you produce value data is the same thing you've got like data and you drill into
9:40it and there's Insight um and so we all kind of came to believe this like very deeply that this is just a true thing that data contains value that like this is not a controversial statement at all I think most people here are probably like yes absolutely that's that's of course what I believe um and you may
9:55have like various caveats to it where you're like oh but clean data or updated or governed whatever whatever tool you're trying to sell that makes that thing true um but fundamentally like a lot of people came to very much believe that data just contains value there is value in the data that you have and all it takes is us to like have the right
10:11tools to get that Insight out of it and you see this all over the place so like I went to a conference about a year ago for whatever reason I was like looking at the various slogans of all the companies and the booths um and they all kind of had this same bit to them and this is like not to say anything about
10:25these companies but like they all have the same exact idea of like unlock the power of your data unlock Enterprise data for modern analytics unlock the value of customer data maximize the value of your data master your data Master data hearing helping you make the most of your data bring data to life set your data free like all this stuff is
10:44basically the same message of like data contains value and we are the thing that helps get you at helps you get at it and like we take this as just like a fundamental truth we take this as like this is not a thing that we could even possibly question we might question how we get at it but like that statement is
10:59true true and so the thing I have is like okay like I don't know maybe not like that I have some questions um shout out Martin Luther um one of the reasons is okay if you go back to those examples like Target Target in 2012 made $73 billion uh they make like a 100 I had no idea where they
11:17were this big um Facebook in 2012 had 1.06 billion users the A's when they were doing all this Moneyball stuff were able to analyze almost 12 million historical bats from like retro sheet which is where all the the historic data comes from when Nate silver was called a witch uh he could look at like the last I don't know five or six elections and
11:35there were 650 million votes to analyze and like this is Big Data these are like a lot of numbers and when you have data this big you can make nice charts that look like this like when you have this many samples of things there is a lot of stuff in that like you can do predictions like this um this is an
11:51example of of like how they did some of the Moneyball stuff this is over a hundred years of baseball data and so you get these nice curves and charts that like seem like they contain a lot of insight but we're all here at like small data SF uh we do not have most of us aren't making 73 billion dollars we
12:08were like Periscope we have a thousand customers a thousand ain't that big um Facebook had a billion users mode in some blog post I found said we had 300,000 users this was in like a marketing blog post so this is the absolute biggest number we could have possibly said um I don't know what the real number is but like it definitely
12:23wasn't bigger than this uh when you have these numbers like your charts don't look like this your charts look like this uh like I don't what is this I don't know like your charts look like this this is from some powerbi thing I found or this this is from an actual mode board deck like what do you
12:40do with this you don't slice into this like you don't dice into it you don't drill into it like you squin at it and you're like it's up is I don't know and like this
12:54is this is like what the problem with bi is is the thing we are not doing is like taking a bunch of data and we have all this insight and what bi will tell you is if only you buy the right Tool uh then we will go from the left to the right but the problem is most of us have
13:10like data that's like this it's not that big and what do you find in that I don't n like it's uppish I don't know um and
13:18so that's a problem and so like to paraphrase Peter teal who had like the whole flying cars thing we were Pro promised these previously unimagined insights this is an actual line from snowflakes website previously unimagined insights uh and instead we got these kind of directional Vibes where you like look at the chart and you're like it's uppish I don't know um so this is a
13:37problem so this is like the the Legacy a little bit of big data was we were promised all this stuff and it turns out the raw materials we have just isn't that valuable so uh what do we do about that I like I don't know I this thing is telling me I have two minutes left uh which is great because I don't know how
13:52to solve this and I wanted to run out of time before I had to get to the solutions um I'm also like a sales guy who's trying to sell you so I don't know uh and I'm like more of an angry critic who doesn't really have ideas that's why we're critics um but I do have like three I guess is points one is that I
14:08think a lot of bi tools could focus a lot more on this left side of things where we give sort of prescriptions of what to do and not to create your own the Temptation for a lot of bi tools is like help you explore help you figure out stuff on your own like it turns out we're not that good at that part um the
14:24thing we need help with is like what in the world do we do with it the the hard part is not creating this chart the hard part is interpreting it and so like bi tools could lean a lot more into helping me interpret stuff so like interpretation of data is often a lot harder than exploration and what most
14:39tools focus on is exploration and that's how we end up with charts like that that are like I don't know what is this the second thing I would say is founder mode uh no um not that at all uh do things that don't scale so I have a friend who was once asked to like recently to figure out the the en
14:59sentiment of articles about a particular topic it was like all right we need to understand the sentiment of how people are like discussing this topic on various news sites to understand whatever she wrote like a bunch of stuff to do like various fancy AI sentiment analysis things and all that kind of stuff uh when she did that she started
15:17like build this thing and then she realized there were seven articles about that topic um there were a lot more than seven articles about Taylor Smith the New York Times this is a bad example for that but uh whatever thing she was looking at there were seven articles and it was like why am I building a tool to
15:30look at seven articles go read them that will take me 20 minutes this actually is true for a lot of things that like data folks do that one of the very early things we did at mode was build a bunch of charts like this this is an individual customer this is like how One customer was using mode
15:44every day and rather than trying to do a bunch of fancy analysis and figuring out these Trends and stuff like that when you have reasonably small data just like looking at every example isn't often that hard and you learn a lot more by looking at stuff like this than you do trying to like cram this into some sort
16:00of smart model or something that's like a trend where can actually make Trends off of the size of the data that you have so like focusing on individual examples actually like it doesn't sound great it doesn't sound like science um but in practice it actually works pretty well when you have data of the scale that most of us have the third thing I
16:17would say uh and the thing I would close with is if you're like I don't want to do any of this small data sucks I want to work with big data great uh come work for Comm Harris um she's spending a whole bunch of money on ads like $500 million like that is very big data uh so there
16:31is Big Data out there uh also if you want to like hire some data people a whole bunch of us are about to be unemployed in exactly six weeks um and and they are all like people who work very hard for not SF salaries um so
16:46uh that is the third option is don't do small data SF do like big data Wilmington I don't know uh so that's where I will end uh thank you everybody thank you
17:01[Music]
Related Videos
2026-01-21
The MCP Sessions - Vol 2: Supply Chain Analytics
Jacob and Alex from MotherDuck query data using the MotherDuck MCP. Watch as they analyze 180,000 rows of shipment data through conversational AI, uncovering late delivery patterns, profitability insights, and operational trends with no SQL required!
Stream
AI, ML and LLMs
MotherDuck Features
SQL
BI & Visualization
Tutorial
2026-01-13
The MCP Sessions Vol. 1: Sports Analytics
Watch us dive into NFL playoff odds and PGA Tour stats using using MotherDuck's MCP server with Claude. See how to analyze data, build visualizations, and iterate on insights in real-time using natural language queries and DuckDB.
AI, ML and LLMs
SQL
MotherDuck Features
Tutorial
BI & Visualization
Ecosystem
2025-11-19
LLMs Meet Data Warehouses: Reliable AI Agents for Business Analytics
LLMs excel at natural language understanding but struggle with factual accuracy when aggregating business data. Ryan Boyd explores the architectural patterns needed to make LLMs work effectively alongside analytics databases.
AI, ML and LLMs
MotherDuck Features
SQL
Talk
Python
BI & Visualization
