In this week’s episode, I talk with Chris Parmer, co-founder of Plotly, about how the company is integrating AI into the next generation of data visualization and analytics tools. Chris walks me through the thinking behind Plotly Studio, their new AI-native environment where natural language prompts generate real, auditable code for charts, dashboards, and data apps. We discuss how this approach reduces bottlenecks for data teams, empowers non-technical users, and reshapes the role of the data visualization expert. We also dive into the limits of public dashboards, the rise of generative interfaces, and what a future of AI-driven exploratory analysis might look like. It’s a fascinating look at where data tools are heading and how analysts can stay ahead.
Resources
Check out Plotly at https://plotly.com/
Guest Bio
Chris Parmer is the Chief Product Officer and Co-Founder of Plotly, the premier Data App platform for Python. As the creator of Dash, Chris leads development efforts to make the framework the fastest way to build, deploy, and scale interactive analytic applications. As data science teams become a standard establishment within the enterprise, Chris works to ensure that even the most advanced analytic insights are accessible by everyone – whether or not they know how to code. His favorite part about working at Plotly is working with our passionate customers.
Different Ways to Listen to the Show!
New Ways to Support the Show!
With more than 300 guests and 12 seasons of episodes, the PolicyViz Podcast is one of the longest-running data visualization podcasts around. You can support the show by downloading and listening, following the work of my guests, and sharing the show with your networks. If you’re interested in financially supporting the show, you can sign up for my Patreon platform, make a one-time payment via PayPal, or shop one of the show’s sponsors.
Transcript
00:01.44
Jon
Hey, Chris. Wow. Good to see you. It’s been several years, I think, since we last talked. How are you?
00:07.10
Chris Parmer
Great. Good to be back. Thanks for having me.
00:09.69
Jon
Yeah. Yeah, it’s always good to have folks back on the show. It means that either they’re bored or I’m still around or a little bit of both. So um good to see you. So, um I mean, let’s get right into it. Like, where do things stand at Plotly? Like, where are you guys? how many people How many people are working there? Like, give us lowdown on where things stand.
00:30.28
Chris Parmer
Yeah, totally. So we ah we are over 100 people now. We’re distributed ah ah little bit around the world. We’ve got some folks in Europe now, but mostly us s and Canada.
00:43.41
Chris Parmer
um We still have a ah huge open source ah community behind our visualization technologies and our data application technologies. um and But most recently, we’ve gone full in on ai and we just released a new product called Plotly Studio, which is our um AI native product for creating data visualization, doing data it data analytics, creating dashboards and data applications. And um and it builds on a lot of the technology that we’ve developed over the last 12 years of the company.
01:19.66
Jon
um Okay, so i want I want to spend most of our time talking about the the new tools, the new capabilities, the AI stuff. but um for folks let’s Let’s just start at the beginning, though. For folks who are not familiar with Plotly, can you give them just like, you know, the quick little what it is and what they can do with it?
01:39.54
Chris Parmer
Totally. Yeah. So we are a data visualization, data analytics company. um Our roots are in code-based data analytics and data visualization. So primarily most of our customers use programming languages like Python to do their data analytics. They primarily work in more technical industries like finance, ah policy, bioinformatics, energy, and data science.
02:07.76
Jon
Mm-hmm.
02:08.17
Chris Parmer
And and Primarily, they use our technologies when traditional BI tools like Tableau or Power BI don’t cut it um because they need to do more advanced analytics or more custom data visualization, um or they’re primarily just doing their data analytics and code for a variety of other reasons.
02:28.15
Chris Parmer
um And now, as the company evolves, we’re kind of taking that same code-based um foundation and architecture, but bringing it into the AI era where now The interface isn’t necessarily just code, but natural language, which means that all these powerful and sophisticated capabilities that have been available for for over a decade that we’ve developed are now available to a much wider audience, which is exciting.
02:53.57
Chris Parmer
like
02:54.84
Jon
Okay, so ah AI, obviously that’s all anyone’s talking about these days. um I am intrigued about how you all are thinking about and incorporating AI into the into the tool set. So um so so is Plotly Studio right now, is that the kind of only place where AI is integrated or is that like the place where it’s most being used?
03:19.98
Chris Parmer
That’s the main place and it’s totally built sort of from this AI first viewpoint.
03:21.63
Jon
yeah
03:25.65
Chris Parmer
So we’re not bolting it on. we really kind of went back to the foundations and thought, okay, if we were to build an AI first data visualization dashboard product, what would that look like?
03:28.18
Jon
Mm-hmm.
03:36.92
Chris Parmer
Um, and our approach is really about, you know, architecturally the AI is generating the code to do data analytics, to do data visualization.
03:49.35
Chris Parmer
And then our product runs that code and then displays the results to the user. And that’s an important distinction to know about because we aren’t asking the AI to do data analytics directly.
04:02.18
Chris Parmer
We’re asking the AI to generate code. And then that code will do the data analytics. And we will run it. And then we’ll display the results. So it’s important because a lot of the issues that you might see or think about with doing something very precise, like data analytics, with something very impre imprecise and prone to hallucinations with like AI um don’t really happen as much in this architecture because the AI is generating the code and that code is is precise.
04:30.36
Jon
Right. So I can ask, I can go into Plotly Studio and say, hey, I want to make a bar chart of the, you know, I’m loaded in the data. Here’s the, I want to make this bar chart, um ah you know, click return or, you know, even more detail, I assume than that, have a conversation with it, you know, build this bar chart for me.
04:46.86
Jon
It generates a bar chart and then it gives me the snippet of the code so that I can go in as the user sort of still control it.
04:52.47
Chris Parmer
That’s right. Yeah. So you have the full you know verifiability, auditability, your ability to edit the code as well.
04:53.39
Jon
Yeah.
05:00.48
Chris Parmer
ah you know it’s It’s not necessarily like sort of a black box that you might have if you are using AI for doing like image generation or something like that. There’s you know a lot that happens in the model that you might not have control over. it it’s ah It’s a different approach.
05:15.09
Jon
Right. so So if I bring, say, a big data set, and obviously I’m asking because I haven’t tried it yet, so this is perfect. I get like the the tutorial right here.
05:26.92
Jon
So if I bring in my big data set, Can I use Plotly Studio to essentially query or interrogate the data to determine what I eventually want to plot?
05:39.97
Chris Parmer
Yeah. Yeah. we We’re primarily focused on the output. So it works best if you kind of have an idea in your mind’s eye of the visualization that you want to create.
05:43.98
Jon
Mm-hmm.
05:49.98
Chris Parmer
You can ask very open-ended questions like, you know show me the top products over time you know by revenue. And you know in that statement, I’m not saying create a bar chart where the x-axis is this and the y-axis is this and aggregate it by that.
06:01.45
Jon
yeah
06:04.60
Chris Parmer
you know I’m just giving an open-ended statement or open-ended question.
06:04.86
Jon
Yeah.
06:09.11
Chris Parmer
And ah and we’ll we’ll make several attempts at generating code to hopefully answer that question. um But you know there there can still be a gap between the type of question that you can ask and the type of answer that’s even possible in the data.
06:23.90
Chris Parmer
But we try.
06:24.16
Jon
yeah
06:24.50
Chris Parmer
And I think that you know it’s an important thing in our product philosophy is We’re always going to be showing you the answer in a table or in a graph and something that is very rigorous without having the AI sort of hallucinate an answer, uh, you know, whether it’s that answer is in the data or not.
06:42.42
Jon
Right. And so if I want to create my bar chart in the style of the Financial Times or The Economist, I have like that sort of branded look is like I can do that in ChatGPT, but I don’t personally, I don’t really trust it right now.
06:51.26
Chris Parmer
Yeah. Yeah.
06:58.42
Chris Parmer
yeah
06:58.63
Jon
But like, could I do that in Plotly Studio and say, hey, i want to make here’s my bar chart, but I want to style it like The Economist.
07:04.44
Chris Parmer
Totally, yeah. you know and It’s going to vary on how much of like that brand style is in the model.
07:05.48
Jon
Yeah.
07:11.54
Jon
Right.
07:11.49
Chris Parmer
um So very common things, like The Economist, or we had a user recently be like, make my app look like the Matrix. you know It’s all dark, green, monospace fonts.
07:18.23
Jon
Oh yeah.
07:20.33
Chris Parmer
And it mean it does a pretty amazing job at it.
07:20.39
Jon
Yeah. Yeah. Mm-hmm.
07:23.71
Chris Parmer
But the cool thing about it, too, is you can also be as precise as you want. um If use these exact hex codes, it’s going to use those exact hex codes.
07:31.70
Jon
isn’t it
07:33.26
Chris Parmer
It’s not going to hold and come up with a different one.
07:33.64
Jon
right
07:35.52
Chris Parmer
so and That’s kind of actually one of the amazing things about working with AI is that you can be as precise as you want. You can start with something open-ended, see what it shows you, the chart that it creates, look at it, reflect on it, and then become more and more and more precise over time until at the end, the sort of spec that you write to define your chart might be really, really detailed and control every single thing that you want.
07:52.57
Jon
then
07:59.50
Jon
Yeah.
08:01.06
Chris Parmer
But you don’t need to start there. And I think that’s a really thing that feels really productive about it. You can start with something very loosely defined and then iterate after.
08:09.40
Jon
Yeah, because it it is lowering the bar for folks who need to make a chart or a dashboard, but don’t care necessarily about the branded colors that they have to use or the font that they have to use. They just know they have to use it.
08:22.71
Chris Parmer
Exactly. Yeah, exactly. Mm-hmm.
08:23.96
Jon
um what um What happens with folks, and this is a question I keep getting asked about lots of different tools that are that are open source or cloud-based, but what happens with folks who have data that are secure, private, administrative, you know that they’re you know that can’t sort of go out in the world? like How have you tried to solve that challenge?
08:43.99
Chris Parmer
Yeah, what we see among our customers is that almost all of the large customers now are running their own model within their own cloud infrastructure. So ah you know for us, our product works best with with Claude, Anthropics Claude model.
09:01.29
Chris Parmer
And we run it ourselves in GCP. And in an AWS, we’ve got it running multiple places. um So we don’t even use Anthropics infrastructure directly. And so our customers will do the same thing. They’ll run the model within their own infrastructure so that it doesn’t leave the firewall.
09:17.97
Chris Parmer
um And we see customers with with different um in some cases using different models for different types of um analytics and data, depending on the privacy of it.
09:30.46
Chris Parmer
Like they’ll use one of the, one of, you know, open or Anthropics model for a certain level of data, but then other data that’s maybe more sensitive, they’ll use Lama or something like that, but they’re still all running it within their organization.
09:45.13
Chris Parmer
And, you know, there’s a lot of kind of interesting geopolitics around it Some customers that are only, you know,
09:49.37
Jon
Yeah.
09:50.54
Chris Parmer
only using non-Western ones, other ones that are only using Western ones, if it’s running within their own infrastructure, like the data isn’t leaking, but there’s some concern maybe about internal bias or something like that, that’s definitely way more on the like super high security, high regulated side of the industry.
09:54.51
Jon
Yeah.
09:59.16
Jon
Right.
10:07.90
Chris Parmer
But those are those are things that that some um folks are considering.
10:11.49
Jon
yeah
10:12.16
Chris Parmer
you know Our approach is just, well, we integrate, we’ve got an enterprise plan that if if that’s if that’s important to you, then that’s configurable at in our enterprise plan. Yeah. Mm-hmm.
10:21.86
Jon
Right.
10:22.54
Chris Parmer
yeah
10:22.63
Jon
So, so far, and I know it’s only been out for a few weeks, but clearly you’ve been testing it for, I’m sure a while. What are you in early sort of stages now? What are you seeing as the, as the use, like either interesting or, or just common use cases? Like, are people just kind of querying it to say, build me a bar chart, build me a line chart.
10:42.65
Chris Parmer
you
10:43.05
Jon
Let’s try seven other things. Like what are, what are the, what are you sort of seeing as a use cases these days?
10:47.16
Chris Parmer
Yeah, the big one is that in so many organizations, there’s a pretty technical data scientist or data analyst that is just totally the bottleneck for the team or the organization.
11:00.21
Chris Parmer
And we’re seeing this adoption now where other that that person’s stakeholders can work a little bit more on their own. um
11:09.97
Jon
Yeah.
11:10.66
Chris Parmer
just by working through natural language. So instead of sending the request ah you know by email, in you the stakeholders are already working in natural language. They have been forever, right?
11:21.35
Jon
yeah
11:21.69
Chris Parmer
They send emails asking for the chart that they want.
11:22.03
Jon
yeah
11:24.50
Chris Parmer
They can just do so directly in the tool. um
11:27.09
Jon
yeah
11:27.68
Chris Parmer
and that’s And that’s great. and and And we see a bunch of companies that have client-facing or customer-facing requirements, and their customers are always wanting the charts and the dashboards to be slightly different.
11:40.04
Chris Parmer
um And they’re instead asking those customers or clients to use Poly Studio instead and ask their own questions. And instead, now the role of some of our customers and even the more data scientists and data analysts is more around the data preparation side of things.
11:56.53
Chris Parmer
ah and getting all of that set up so that then everybody else can be self-service in the analytics downstream of that, which is really exciting. I mean, there’s tons of bottlenecks there.
12:06.41
Jon
Yeah.
12:08.34
Chris Parmer
It’s it’s really great for people to be much more empowered to work on their own.
12:12.62
Jon
Yeah. But what do you see when that person who is not the data person is doing these queries using the AI? Um, but they’re still, they’re probably not then looking at the code that’s generated.
12:26.02
Jon
Is there still, have you seen like they can build, they can build now the chart that they want, but are they still going back to that data person to say like, Hey, I’ve done this. Can you check?
12:36.08
Jon
Or is it like, there’s enough faith at this point that they’re just like, I just kind of need this bar chart for internal work. It’s, know, yeah.
12:41.97
Chris Parmer
Totally, yeah. There is now a lot of faith in the code itself. And I think what happens more is
12:48.88
Jon
Hmm.
12:52.36
Chris Parmer
there’s more interrogation around the data. In a way that there always has been, though, I don’t think the dynamic is really different. and And in a way, it’s it’s almost helped because the stakeholders that are looking at these these charts often are the domain experts of the data anyway.
13:11.91
Chris Parmer
So it’s easy for them to spot check things that are obviously wrong, or even to create the follow-up visualizations or tables that they can use to spot check their own results.
13:14.76
Jon
Oh.
13:23.07
Chris Parmer
So you know you create a bar chart that’s an aggregation of a bunch of other data. And the way that you verify that isn’t necessarily just by looking at the code to see if the Panda’s syntax is correct.
13:33.36
Jon
Yeah.
13:33.46
Chris Parmer
But the way you want to do it is to then like drill in and look at each one of those bars and take a sample of the data and kind of do all this manual spot checking work, which is something that often the stakeholders are pretty good at because they’re the experts in the data anyway.
13:48.01
Chris Parmer
um And so we see a lot more kind of more importance around building tooling around that spot checking, auditing the data kind of thing.
13:48.39
Jon
and
13:58.06
Chris Parmer
And I think the more people that are looking at the data, the better that is.
14:02.14
Jon
yeah Yeah, for sure.
14:03.05
Chris Parmer
yeah
14:04.34
Jon
um You mentioned the the dashboarding world. I’m curious um where Plotly sort of sits. I noticed on the main page of the website, there’s a big focus on dashboarding.
14:15.97
Jon
I’m curious where Plotly is with dashboarding and also how it how it intersects with the with the new AI tools.
14:23.39
Chris Parmer
who Yeah, so what we look at, you know if you’re just working in a chat based interface today, like using chat GPT, and you ask a question, you’ll get an answer that’s an answer at a point in time, right?
14:35.73
Jon
Mm-hmm.
14:35.98
Chris Parmer
So if you upload a data set and you ask, you know, and for a chart or something like that, chart’s not going to change, that answer isn’t going to change over time. And so there’s still this importance in my view that you are creating these artifacts that have eight that can have a long lifespan and that will actually update when the data changes.
14:54.94
Chris Parmer
So if you’re working in chat today, you might get an answer, but then you know you don’t want to have that same 50 back and forth you know chat tomorrow to get that same answer and do that every single day.
15:05.71
Jon
Yeah.
15:08.03
Chris Parmer
right And so that’s our view is that
15:08.32
Jon
Yeah.
15:11.47
Chris Parmer
if If as a side product of doing data analysis with an AI, you get this long-lived dashboard that updates in real time with your data that you can always go back to without needing to go back to AI and have the same conversation over and over, that’s really good.
15:27.46
Jon
Mm-hmm.
15:27.48
Chris Parmer
So that’s our view. We view as more of this tool to be creating these dashboards, and then dashboards have a long life. Consumers of the dashboards often well we’ll want to look at the same thing in a same format every every day. And the consistency of that is really nice.
15:44.46
Chris Parmer
Of course, a lot of the downsides of dashboards still exist of like, you might want to view it in a slightly different way. And I think now AI enables those end viewers to customize their own view in a much easier way.
15:54.28
Jon
Mm-hmm. Right. um So I don’t ah personally, so I’ll admit, I don’t have much experience using Plotly and dashboards. And I’m curious, ah this is really two part question.
16:06.90
Jon
So the first part of the question is, are people using the dashboarding capabilities in Plotly sort of, you know, historically for public or internal public facing or internal facing sort of dashboards?
16:19.54
Jon
And secondly, in your experience, Do people actually use the public facing dashboards? I have more and more hesitation that when I, you know, create sort of this like data exploration tool and put it on the web that nobody really uses it.
16:33.22
Jon
Like internal is, I think, a different world.
16:34.02
Chris Parmer
totally yeah yeah yeah totally yeah i totally agree i think where we’ve seen the most the most value the most uptick is when people are building these dashboards that are really used in operational settings
16:35.68
Jon
You and I are working on a project together. We need to explore the data together in real time. That’s a different, I think, ask. But public facing, I just feel like we just put a bunch of stuff out there and nobody looks at it.
16:47.69
Jon
So that’s kind of two-parter there.
16:59.88
Chris Parmer
right and and we’ve we’ve kind of come up with different terminology like we call them data apps sometimes as instead of dashboards, to signal this more like this is something that’s used in your day-to-day operations, right? It’s really tweaked for the viewers that are using it that need to reference this data every day, whether they’re monitoring like the electric grid and they’re out in the field looking at the data.
17:20.73
Chris Parmer
it These dashboards are used kind of as reference materials or they’re used the monitor trends over time and things like that. Or they have you know input and output so you can actually upload your own data or update the data behind it because you’re an end user within this organization that’s modifying.
17:38.48
Chris Parmer
That’s definitely what we see. And I totally agree with you on the public side.
17:40.23
Jon
Yeah.
17:41.79
Chris Parmer
It’s very tempting to put up this sort of generic chart builder and say, oh, yeah, well, anybody can just look at whatever they want.
17:49.69
Jon
Yeah.
17:50.76
Chris Parmer
It’s actually often not true.
17:53.81
Jon
Yeah, that’s right. Yeah.
17:55.67
Chris Parmer
and And I think about this a lot now with AI because yeah we’ve built our fair number of chart editors over time. And they need to become so sophisticated to be able to answer the broad range of analytical questions that you might have.
18:12.33
Jon
Mm-hmm.
18:12.57
Chris Parmer
And so it is now so much easier just to create the bespoke visualization that you want immediately without going through this you know generic chart editing interface.
18:23.89
Chris Parmer
A classic or an example I’ve been exploring a lot on the public side is I look at ah San Francisco’s 311 data, which is like the complaints line. So if you have if there’s somebody blocking your driveway, you call this number. And the data is public, and it’s awesome.
18:37.96
Chris Parmer
And we have a new mayor in the city who’s been great. And I’m kind of curious, like, OK, can we look at the data to see before and after? Has you know response time to city got better or not?
18:48.81
Jon
Mm-hmm. Mm-hmm.
18:50.55
Chris Parmer
Has it gotten better in certain areas, neighborhoods, or not? And there’s, ah of course, an Explore This Data interface on the website. It’s really tricky to to come up with a definitive answer to these types of questions.
19:01.80
Jon
and
19:03.27
Chris Parmer
You’re doing complex comparisons. You’re doing computations to look at at the diffs of different, you know the time range between open and closed date. You’re comparing certain things against each other.
19:15.13
Chris Parmer
like There’s a lot of analytics that happens behind the end simple visualization. And I think that’s the piece where, speaking out loud, that’s the piece that we’ve kind of missed when we create a generic chart editor is that there’s just so much analytics required before you get to the chart.
19:20.71
Jon
Yeah.
19:33.08
Chris Parmer
And it’s not like big data analytics. It’s not like data pipeline stuff.
19:35.49
Jon
No, right.
19:38.70
Chris Parmer
it’s It’s sort of data prep. like we haven’t I think as an industry, we haven’t had a great terminology around it. It’s this lightweight data analytics, data prep stuff. um And when we’ve built just a chart editor in the past, it hasn’t always included that. And I think that’s that’s what we what I love about Studio today is that it’s generating the code and all of that code that you might need to do to prep and shape and analyze the data before you just put it directly into a bar chart is all part of the core experience, right?
20:05.11
Jon
Mm-hmm. Yeah.
20:06.73
Chris Parmer
don’t know.
20:06.78
Jon
So is the, I mean, I just, this has been bouncing around my head now for the last, i don’t know, 10 months, basically that there’s a lot of public dashboards out there.
20:07.15
Chris Parmer
I’m curious what you think. You clearly have a… you
20:18.69
Jon
a lot of them people make for fun or to show that they have technical chops. All that’s good. There are a lot that you go, you know, in my sector, the nonprofit social policy, public policy sector, you see a lot of dashboards, you know, here’s the job data and you can go explore for,
20:34.31
Jon
your state, your county or whatever. And I just, I feel like if your target audience is the sort of like, I’m going to say like regular person, right?
20:47.96
Chris Parmer
Yeah.
20:48.31
Jon
Not another analyst, not another, you know, a social scientist, not another data visit enthusiast. if you’re if you’re If you’re trying to reach that, you know, that dad coming home from work,
21:02.13
Jon
I don’t think they want to explore dashboards. I think they just want to know the answer like right away. And so I kind of feel like what often happens is like the dashboard part is at the top.
21:13.36
Jon
And then maybe there’s a description at the bottom where I’m starting to feel like it should be flipped. You should tell a story and then say, okay, so here’s the main, here’s the bottom line.
21:20.13
Chris Parmer
Mm-hmm. Mm-hmm.
21:24.36
Jon
Here’s the headline. Here’s like three examples. If you really want to go find your zip code or your area, like, you know go in and and dive in and i know that that explanation is uh sort of context uh that that experience is going to be context dependent right if you if that is important to you to know xyz then you’re going to spend more time but i just think we just pour all this stuff out and if this is expectation that people are going to dive and explore the data i just don’t think they’re doing it
21:55.68
Chris Parmer
yeah Yeah, I agree. And I think it it comes down to sort of the role of data viz, right?
22:00.97
Jon
Yeah.
22:01.25
Chris Parmer
In cases, it is like ah it’s a lookup table, as you mentioned. You are very interested in in the certain you know the house price average house price in your zip code.
22:12.78
Chris Parmer
And great, like a data exploration tool will give you that.
22:12.88
Jon
Yeah.
22:16.28
Jon
Mm-hmm.
22:16.57
Chris Parmer
Other times, you’re interested in the broad range of like what’s happening to housing prices and then in in the city. but you’re not even just interested in that. You’re like, why is that right and
22:24.83
Jon
Yeah, right.
22:26.28
Chris Parmer
included in the data itself.
22:27.79
Jon
No. Because that’s just the data, but I want the understanding. I want the story around it, which is, you know, and this is, I mean, all this is purely just a feeling and anecdotal, but I also kind of feel like when I look at major, the major newspapers, I read Washington Post, the New York Times, the Guardian, um you know, l LA Times, bunch of others, the number of those sorts of interact or exploratory interactive dashboards seems to be far and few between these days.
22:56.16
Chris Parmer
Yeah.
22:56.69
Jon
where it’s more focusing on the storytelling than on the big data pieces or the big interactive exploratory pieces. Because my guess would be they have discovered that you don’t get a lot of bang for your buck for that.
23:09.93
Jon
And people go to the Washington Post to learn the story, the understand the context or the content rather than like, here’s this 700 clicks and filters to get to this one number.
23:23.75
Chris Parmer
Totally. Yeah, and I think the amount of like sort of design effort that goes into creating something that’s highly consumable is also super unappreciated.
23:34.58
Jon
Yeah.
23:35.77
Chris Parmer
And so, and the way you present your data for a particular answer is, is often so bespoke if you want to do a really good job at it.
23:46.03
Chris Parmer
And very generic.
23:46.20
Jon
Yeah, I mean, i i’ve heard I heard a number the other day that the Johns Hopkins COVID ah tracker costs $13 million dollars to build, which one could argue, whoa, that’s a lot of money.
24:00.31
Jon
But Also, like how many millions of people went to that dashboard every day.
24:06.37
Chris Parmer
right
24:06.60
Jon
But that, of course, is also a a hopefully unique situation where people want to know for their own health what’s happening around them. They don’t really.
24:17.59
Jon
you know, what’s happening in the other in another part of the world doesn’t necessarily impact them directly. They want to know if they can send their kids to school, if they can go to their grocery store safely, right?
24:27.11
Chris Parmer
Uh-huh.
24:27.32
Jon
So think that’s kind of a unique case, whereas generally we’re just putting this stuff out there. Which does lead me to a question about if you see the future of dashboarding… okay I’ll put it this way.
24:42.27
Jon
You’ve described Plotly Studio as ah query the tool to generate a chart for you. um Do you see a future, either in Plotly or just generally speaking, where the end user queries the, I guess, ah an interactive piece, so we’ll call it dashboard, where they query the dashboard using natural language. Instead of having to click all the buttons, they just ask the question.
25:08.73
Chris Parmer
Yeah, i think i think that will happen, but I don’t think it will be on the dashboard itself. I think it will just interacting with the data itself.
25:16.32
Jon
okay
25:19.68
Chris Parmer
And instead of the dashboard being this sort of fixed UI, it’ll be a generative UI. It’ll be a generative dashboard where you will be asking questions and getting a unique chart that maybe hasn’t been viewed before or originally created by the original author before.
25:31.27
Jon
Thank you.
25:42.09
Chris Parmer
um And so as an author ah or an expert in the data, your primary job is to prepare the data in a way and maybe put in some of your own context of important things about the data so that then the consumers and the viewers um can have their own exploratory experience where they can ask the questions directly so in the case of my you know sf public data one i don’t want to go to ah dashboard and ask for a question because the the charts that are presented to me might not might not have the necessary controls or be presented in a way that can answer my question
26:25.62
Chris Parmer
I do want to go in and say, hey, how has the response time for 311 cases changed in October of this year versus October of last year? And I would love it if the answer showed me sort of proof in the form of a data, right?
26:43.04
Chris Parmer
Like in form of graphs. I don’t want to just see, it oh it dropped 5%.
26:45.10
Jon
Right.
26:47.46
Chris Parmer
I’m not going to trust that 5% number because that, you know, that could, is that a median?
26:50.64
Jon
Yeah.
26:52.98
Chris Parmer
Is that a mean? Is that the 90th percentile? How is it skewed?
26:55.20
Jon
Yeah.
26:56.21
Chris Parmer
Right. So instead The AI is then showing you a set of charts to say, hey, this is before and this is after. Come to your own conclusion. But it’s generating charts that the original author might not have known about because in a data set like that, there’s hundreds or thousands of permutations and types of questions that you can ask.
27:05.05
Jon
Right. Yeah.
27:13.21
Jon
right
27:14.49
Chris Parmer
so And I think that’s the really unique thing about this new era of data visualization is that you can start to ask any question that you want and see this almost generative data visualization.
27:26.21
Chris Parmer
and And it’s hard as an author to know the types of questions that your users will want to ask unless you work in a really tight organizational setting.
27:34.18
Jon
yeah Right. but but that’s the But the way you frame it is really interesting because it in lots of ways changes what the job is or the task is of the data viz developer becomes less about the data viz, but more about facilitating or analyzing the data and then facilitating the a set of questions that a user might want to ask.
27:46.51
Chris Parmer
Mm-hmm.
27:57.85
Chris Parmer
That’s right. Yeah. And I think there’s a lot to you know to for that person to create the initial stories, to see the viewer, to know the types of things that they can ask about. And I think that’s that’s a really tricky thing about AI.
28:12.06
Chris Parmer
Today, I’ve heard it called like the jagged frontier, where AI is remarkably great at certain things and then terrible at other things.
28:18.26
Jon
Yeah.
28:20.44
Chris Parmer
And unless you’re an expert and have a lot of experience using it, you really know what is what, right? And I think for user are coming into a dashboard, they might not know the types of questions that can be asked about the data.
28:26.49
Jon
Right.
28:32.95
Jon
Mm-hmm.
28:33.59
Chris Parmer
Seeing a set of stories that were already created by the data viz expert, um that show how to structure your your questions and your analytical queries, but also show you the realm of possibilities of what you can interrogate is a really important part.
28:48.10
Chris Parmer
But then you you leave it up to the end user to kind of choose their own adventure, without needing to create a bazillion charts for each end user themselves.
28:57.63
Jon
Yeah, and also putting it into a a finite space, I would guess, reduces those hallucinations, right? Like you’re not going to get, you know, if I if i went into ChatGPT or Cloud right now and asked for 311 data, it could pull from the New York City database rather than the San Francisco database.
29:06.13
Chris Parmer
Yes.
29:15.87
Jon
But if my Plotly, Plot Studio dashboard fine, populated with San Francisco data, then the hallucinations should be minimized.
29:27.07
Chris Parmer
Exactly, yeah. And there’s a tremendous ah amount of work to sort of create the set up the environment so that an end user is ready to go and ask those questions.
29:28.16
Jon
Yeah.
29:35.77
Jon
Mm-hmm.
29:38.07
Chris Parmer
And that’s so that’s a lot of stuff we’re thinking about within our product is these different roles where one person is just setting up the data, just setting up the environment, putting in additional context and about the data sets so that the end users can come in without needing to do all that work themselves and have and and have a productive experience.
29:59.97
Chris Parmer
So, yeah, you know I think there’s these different roles. And I think the other approach that we take is that we want users to be able to view their answer as a chart or as a table.
30:12.73
Chris Parmer
and not just have the AI hallucinate a story.
30:13.18
Jon
Yeah.
30:15.82
Chris Parmer
If you go into ChatGPT today and you say, tell you know have 311 cases improved since the mayor has come in, they might look up news articles or they might hallucinate you know something else about it and say like, yes, of course they have.
30:26.78
Jon
Yeah.
30:32.24
Chris Parmer
Or you know if you structure your question in a slightly different way, be like, I love the mayor. Isn’t it amazing? They might agree with you, agreeable.
30:37.73
Jon
yeah
30:40.60
Chris Parmer
But we take this approach as very data driven. We’re not asking the AI to tell you you, know, create an insight for you.
30:43.57
Jon
Yeah.
30:49.07
Chris Parmer
We’re asking the AI to generate code that will use the numbers for you and present the numbers to you and you can come up with conclusions.
30:51.37
Jon
Right. Right. Right. Yeah. So on the roles piece, do you, I mean, I think a lot of people in the data viz world are understandably worried about how AI will impact, you know, is, is data viz not going to be, you know, our data viz creator is not going to be a thing in the future, but do you think Um, the, the role of the data visualization person, the specialist, whatever it is, is gonna be focusing more on, I guess, facilitating those questions and doing the analytics rather than focusing on how do I make this like really great chart and, and just facilitate the data exploration.
31:33.81
Chris Parmer
I think eventually it could happen that way. I think today, the way I view the AI tools is more like productivity tools. And if you are a great data viz practitioner today, use AI tools to make you to help you work faster.
31:42.61
Jon
Mm-hmm. yeah
31:51.93
Chris Parmer
um And in all of those data viz skills that you have today, will serve you really we’ll in knowing how to craft the right data visualization. That is still important. And then I see that among our users. I see that internally of certain folks being more productive and effective at using Plotly Studio than others, even though we’re all just writing natural language. But the way we structure our visualization, you know the mechanics of whether we do that in code or and in English has changed a lot now.
32:23.02
Chris Parmer
But you still think in a really systematic way you still think in terms of database you still have the data viz in your mind’s eye that you want to create. and that’s And that’s essential. That’s an essential skill I don’t think is going away.
32:36.56
Jon
Right. um Okay. Well, to to wrap us up, ah folks have heard this conversation. They’re either terrified or they’re super excited. let’s Let’s go with the super excited folks.
32:48.51
Jon
So um where should they go? um What do they need to get started with with Plotly or Plotly Studio or any other tools that we’ve been sort of talking about?
32:59.27
Chris Parmer
Yeah, just go to plotly.com, P-L-O-T-L-Y.com. You can download the product for free there directly from the homepage. We’ve got a great community forum as well. There’s a lot of good discussion about this and show and tell people sharing what they’ve created. and And after you sign up, you’ll you’ll get an email from us. And if you want to talk further, you can just reply to that email. And a lot of those good still go directly to me.
33:24.85
Jon
Terrific. Love Getting right to the guy. All right, Chris.
33:28.68
Chris Parmer
That’s right.
33:30.23
Jon
Thanks a lot for coming on the show. This was really interesting. I’m excited to go in and start to play around with it and see what these new tools could do. Yeah.
33:37.20
Chris Parmer
Thank you.
33:37.66
Jon
So thanks again for coming on the show. Appreciate it.
33:39.19
Chris Parmer
Great to be here.
I’m curious — have you tested this approach in real-world situations or larger workloads?