Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
044 - Synthesia: Data Director - Why Data Teams Should Stop Trying to Be in Every Room
160 Plays16 days ago
Ed Mancey has a take that a lot of data leaders won't like — and he's got two years of results to back it up.
As the data team lead at Synthesia, Ed has built his function around a single idea: the data team's job is to be a multiplier for the business, not a bottleneck. That means owning platforms and tools, not seats in strategy meetings. It means drawing hard lines on responsibility. And it means trusting your business users to actually use what you've built.
In this episode, Ed breaks down how he's applied that philosophy at one of the UK's fastest-growing AI companies — from a two-year Omni implementation to enabling a sales manager to hit 150% of quota using tools his team built, without a single BI dashboard in sight.
We get into:
- Why being the C-suite's personal analyst is a career ceiling, not a launchpad
- What a truly self-service data function looks like in practice
- How Ed thinks about drawing the line between what the data team owns and what it doesn't
- The no-blame culture that makes delegation actually work
- What AI tools like Cursor and Claude are doing to the day-to-day reality of data work — and what that means for the teams building around them
If you lead a data team, or you're trying to figure out where to focus your energy to have more impact, this one's for you.
Recommended
Transcript
Introduction: Data Team's Role and Challenges
00:00:00
Speaker
If I say that the data team are responsible for building reports and dashboards for the commercial team and the commercial team is growing very fast and I'm slow or hiring or something goes wrong or whatever else, I become a point of failure for the commercial team, which is exactly what I don't want to happen.
00:00:15
Speaker
He was 150% to quota as a team last month and he says that this played a big role, right And that's what I mean by being operational, right? Ultimately, we provided the tools for him and his creativity to do better at his job than he would have done before.
Omni Sponsorship: AI Analytics and Business Impact
00:00:28
Speaker
Today's episode is brought to you by Omni. Most companies I speak to want AI analytics but are failing to put projects into production. That's where Omni is different. It's the semantic brain that grounds AI in the heart of your business logic, giving you governed answers, whilst also the depth to identify root causes. It's intelligence everywhere, from your spreadsheets to their chat feature, even within your product. Omni moves you beyond the dashboard. Don't just take my word for it. Trust teams like Perplexity and Synthesia that are already using Omni to deliver intelligence that people trust.
00:01:00
Speaker
Check out them in the show notes or visit omni.co. That's O-M-N-I dot co. Now, back to the show.
Ed Mansi on Data Teams' Operational Focus
00:01:08
Speaker
Hello everyone, welcome to another episode of the Stacked Data Podcast.
00:01:13
Speaker
This week I'm joined by Ed Mansi, the Director of Data at Synthesia. Today we're actually going to jump into what a data team actually is for and why Ed believes sometimes data teams focusing on the wrong things.
00:01:31
Speaker
We're going to unpack his view on how data should be operational and not just analytical, how he structured the team at Syncthesia to reflect that, and why he's deliberately drawn hard boundaries between where data what the data team does own and and doesn't own.
00:01:48
Speaker
We'll also touch on self-service and AI and tooling and include how they're using Omni to change the way that they look at data across the business and and commercial teams.
00:02:02
Speaker
Ed, it's great to have you on the show today. Thanks for joining me. How are you doing? Yeah, thanks for having me, Harry. it's I think about half of our team has come by you in some form or another, so it's great to to be on and and to join you.
00:02:16
Speaker
Excellent. Yeah, no, really looking forward. It's been a pleasure working with you um and helping build the team. And I think I'm really keen to sort of share more about your view on it. think it's definitely very forward looking and the type of organisation that you've been in, obviously at Synthesia and previously. So um yeah, for for the audience, Ed, I obviously know you well. Who who are you? What's your background and and and how did you get to where
Ed's Career Journey and Experience
00:02:41
Speaker
you are now?
00:02:41
Speaker
Yes, I'm Ed, and I'm the director of data here at Symphysia. Before that, I was um managing kind of go-to-market analytics at GoCardless. And then before that, I kind of cut my teeth at a variety of different roles at Groundwatch, where I was for sort of eight years. um My background actually there is I i joined in ah the RevOps team at the first doing sort of like order management type stuff. So the first stuff I was doing was...
00:03:06
Speaker
whenever a sales order came in checking it was a PO number and that the address was correct and all of that kind of stuff. But there's an opportunity there to do some kind of sales analysis on the side. um And I think over my time at Brandwatch, I must have worked functionally as a marketing analyst at times, sales analyst, finance analyst, strategy analyst, lots of different bits.
00:03:29
Speaker
um And so my background is definitely more on analytical side and the the the kind of commercial side, I guess, or the operational side, um but now it's Synthesia.
00:03:40
Speaker
I was the first data hire there when I joined two years ago, and I'm responsible for the data platform as a whole across product and commercial, and then all of the commercial and operational ways we use data.
00:03:54
Speaker
Excellent. And and who else is Synthesia for those that but that don't know?
Synthesia's AI Video Platform: Benefits and Applications
00:03:58
Speaker
Yes, so Synphysia is um a sort of an AI video platform. um I think the the way to describe it really is that we all sort of know that video is more engaging than text. um But video generally is like far more expensive historically to create and also to maintain. um synvisia gives the ability to have 90 of the engagement benefits of the video i mean it doesn't look exactly like something should shot in the studio but it's not too far off but with 10 of the cost of making video um examples of the kind of places where we have success would be sort of like learning development teams or other teams where they know they get a big benefit from using video but sort of can't afford to be going in a lot or something like
00:04:41
Speaker
I don't know, an explainer for a water bill is ah the kind of topic that actually the text needs to change very often. um So it's quite hard to have that be an ongoing thing that you can have on your YouTube channel. But with Synthesia, videos become living docs. So in the same way that you can update a Google Doc and it says something new, you just type in a new script and you get something new back. So if you've got something...
00:05:05
Speaker
Like, I don't know, explaining a water bill or something is very good for that. And and so we're a we're tech company based in London providing AI video. Amazing. Yeah, and I think um one of the coolest sort of AI companies coming out of the the UK for sure, and if not the the world. So check them out.
Purpose of Data Teams: Enhancing Job Performance
00:05:24
Speaker
um Okay, so Ed, taking a step back, I suppose, what what's your view on what a data team is actually for? Why why do you the data' actually exist in in your opinion?
00:05:37
Speaker
I think data teams exist to sort of help people in the company do sort of their jobs and ideally any job where sort of data might or might not be useful. so A
00:05:52
Speaker
ah a manager for an SDR team needs to understand in detail the performance of each of their reports to give them more support. Or it might be a member of customer success team needs to um you know tailor the outreach to a specific client based off their recent usage. Or it might be um how do we sort of like programmatically make sure that the sort of from the right lead goes to the right person. So and at its core, data itself is like, I think, quite a tangible thing, right? Every sort of like row or record in a database, it represents something in the real world.
00:06:28
Speaker
um And so the data team is taking those representations and making it kind of structured and clear and whatever else, but then making it as easy as possible for people to then use that on an ongoing basis for their role.
00:06:43
Speaker
Excellent. And how do you measure success?
Measuring Success: Data Accessibility and Business Alignment
00:06:46
Speaker
I think that's the thing that most teams are looking at. What are the signals that you should be tracking and maybe what the signals that you shouldn't be tracking for the success?
00:06:58
Speaker
um for For us, I think that the main thing I think about is that there is a lack of sort of like a combination of people in the company are doing their job using data and they're happy right ultimately particularly for our commercial teams if our commercial teams hit target we're doing well and if they don't we're not right and that we should hold ourselves to that and be like fully aligned with that um and so you want success stories and whatever else without happening and and we have a lot of those um
00:07:31
Speaker
And then it's so it's a base event, providing a platform to that and how successful it is. So do people have access to the data they need, right? Like it is the data modeled in a way where they'd expect it? Are all the business systems they need lined up?
00:07:46
Speaker
Is it kind of seamless and easy for them to do, right? So it's almost like the user satisfaction. And we can see that in things like um we have very high utilization of
Focus on Operational Execution Over Decision Support
00:07:56
Speaker
Omni as an example, right? Like um I think that we've grown that's steadily and significantly and we have almost no users who don't use us on a weekly basis which is a really good example that we're doing something well and baking something in into a workflow rather than a kind of a one-off piece um and where it is frustrating and we had some mis issues recently around uh our customer success team being frustrated with amount of product usage data they had that's an example of where you know you're failing right because ultimately there's someone saying i would like this piece of data to do my job better and i i don't have it available to me
00:08:32
Speaker
You said something really interesting now, obviously, about embedding into workflows. And I think that's something that you you've spoken about a lot, about data being primarily operational and systems focused rather than just this sort of ad hoc analysis type of work. Can you you help sort of explain your your thoughts and beliefs around that, that but then points?
00:08:57
Speaker
yeah i i think that like as a team you're looking to drive kind of business impacts and to be aligned with a successful company and um in general i don't actually think that comes from decision making right i like it's obviously important for companies to like make the right decisions and have a big bets but um From my experience, where companies really succeed or fail is their and their ability to actually operate and execute well.
00:09:26
Speaker
um And if you can play a bigger role in that, I think that you're going to like have more leverage and be be more successful. And so the question is not sort of like, are the people you're working with making the right decisions? It's like, are they performing well in their jobs?
00:09:40
Speaker
um And if everybody's performing well in their jobs, the company's going to do well. So you should hold yourself accountable to what you're accountable for and let others hold them themselves accountable for what they're accountable
Optimizing Data Pipelines at Synthesia
00:09:52
Speaker
for. But make sure that whenever they sort of need your support, you're kind of giving it in that way.
00:09:57
Speaker
Interesting. site I think I'm keen to unpack a bit later in a bit more detail, particularly i think with your your career um ah path as as well.
00:10:08
Speaker
but it would be good to understand when you joined Synthesia, you said you first data hire. What was the landscape of data? Was there tool How was it perceived within the business? And then yeah what what have you set out to build from there?
00:10:24
Speaker
Yeah, when when I joined, there was something which had been built um sort of like jointly by the head of RevOps and the head of product at the time. So there's a DBT pipeline on top of Snowflake, Fyvetran for ingestion.
00:10:38
Speaker
There was no BI tooling at all, actually. um And it sort of, it worked perfectly. well enough in that sort of for the right power user it got the right output but it definitely wasn't structured in the way a data person would structure it um And I think as an example, you know, the dbt pipelines were kind of slow and they failed a lot and they they they weren't reliable or um they use Metabase as a BI tool And, you know, in every single a SQL query in Metabase, there would be sort of like a CTE that calculates segment, as an example, right? And that's obviously something that actually you want to standardize in your ETL, right? So that everybody's seeing segment in the same way. So it was a bit, it was a bit all over the place.
00:11:23
Speaker
um And the lack of a BI tool was particularly sort of concerning for me knowing the growth it was about to go through. And it served a very small number of very technical users very well.
00:11:35
Speaker
um But it didn't serve the broader business at all who had very little access to be able to self-serve on data. right Everything had to go through um a RevOps person or kind of a product person because you know there was no going to self-serve tooling. So for me, that's kind of the first thing to do, it seemed obvious was to try and rebuild.
00:11:58
Speaker
And
Building a Robust and Efficient Data Team
00:11:59
Speaker
we started from scratch the dbt pipelines in a way which would scale, which would be easy to maintain ah and where, um, You know, I just binned off all the old code and kind of started again from scratch. And then the other thing is knowing my strengths and weaknesses, it was to bring in a data engine in. um Because in a 12-month period, we'd seen something like a 10x growth in the snowflake spend, which happens when you don't have optimized pipelines, you've got a fast-growing company, whatever else. It's like, well, we need to get that under control. And again, understanding that the...
00:12:32
Speaker
risks to data generally being something that can be used a lot is actually the platform falling over, right? So the first thing to do is to bring someone in to solve that that problem.
00:12:43
Speaker
And now, two years later, how has it started to sort of all piece together? You started with yourself, stripping back all of the code, bringing in a data engineer. what what What's the makeup sort of years later what have been some of the biggest challenges along long the way?
00:13:01
Speaker
Yeah, so the makeup now is that we've got a quite a mature team. um So ah on that sort of data engineering and platform side, there's myself and a data engineer, and we're bringing another data engineer in.
00:13:15
Speaker
We've got on the kind of a bit separate to us on the product side, we've got a product analytics team who kind of serve a product, individual product needs very well. um And then we also have on on the kind of commercial side, someone who does all of an analytics engineer who's does all this sort of the business logic, um a data scientist who um does some really cool, interesting stuff actually about, you know, building models for kind of commercial benefit.
00:13:42
Speaker
um And then we recently brought in someone who's sort of fully focused on the front end. So the title AI Knowledge and Analytics Manager or Knowledge and Enablement Manager, or something like that. um But really it's,
00:13:57
Speaker
ah we're moving to a new world where people are going to want interact with these tools differently. We think it's right to have someone full time on that self-serve experience. um So we're still quite a small team and I like that. I like it that our kind of way of working where there's no kind of task management. We can sort of trust each other and and work pretty fast.
00:14:19
Speaker
um And things are fairly stable for our old stack, which is Fivetran, DBT, Omni for our BI tool. But we only this week launched Omni and Claude.
Scalable Data Platforms and Self-service
00:14:32
Speaker
And I think there's a kind of a new stack and a way of working, which we'll see.
00:14:37
Speaker
And we've also brought in someone on the finance side for a data engineer. So that's a team structure. Nice, excellent. And yeah, I mean, you mentioned obviously this AI knowledge analytics manager, a lady called Gina that we we've brought on board and and we're actually going to be recording and an episode to dive deep on her role as this sort of AI knowledge manager and so owning the self-serve.
00:15:00
Speaker
a lot deeper. right I think one thing that's clear is that you sort of focused in on building um the platform that is enabling self-service scale rather than the the embedded type approach, which imagine there is is probably more familiar than companies like GoCardless.
00:15:19
Speaker
Has that been the right call? What made you make that decision to to build more of a a platform that is a bigger enabler rather than this sort embedded um approach we see so often over the years?
00:15:33
Speaker
Yeah, it was definitely the right decision for us. um And I think that for sort of two reasons. One is um you need to draw a line between what you're responsible for and what you're not responsible for. And once you decide you're responsible for that thing, you need to live up to that and be responsible for it, right? And that means that you need to resource against it. And more importantly, if you're not able to deliver that, that becomes a blocker to the business's growth. So If I say that the data team are responsible for building reports and dashboards for the commercial team and the commercial team is growing very fast and I'm slow or hiring or something goes wrong or whatever else, I become a point of failure for the commercial team, which is exactly what I don't want to happen.
00:16:16
Speaker
So the more that I can say, what what am I going to actually be able to scale and deliver and draw that line and really focus on that, the more that I can make sure that I'm a multiplier for the rest of the business. rather than a potential point of failure. and and And you kind of scale the business in a weird way. And suddenly, when you have a data team and...
00:16:36
Speaker
I think we've all worked in them and you now have like 20 analysts or something. It all kind of changes and it's very it's very difficult, right? And hard to run in the same way and it becomes a point of contention.
00:16:47
Speaker
And on the other side, um i think you really get worse outputs where say, um I don't know, we have a quite big SDR team and one of the SDR managers was able to build a dashboard looking at call to connect and how that was changing. And and in their investigation, they found that,
00:17:05
Speaker
um After we call a number five or six times, it's very, very unlikely it's going to be connected. it Sounds obvious, but they were able to do that and understand that the reason why the call to connect was dropping is because we weren't providing enough new phone numbers to the team and they were re-ringing the same numbers too many times.
00:17:23
Speaker
um that isn't Now I explain it to you, it's kind of an obvious thing to look at, but I just don't think it's anything that any data analyst I've ever worked at worked with would look at. Actually, when you when you take a step back, it requires quite a deep operational knowledge of that job.
Empowering Users in Dashboard Creation
00:17:41
Speaker
um And so I also think that you get better better set of analytical answers and responses when you take a step back and you pass it on to the team and you say, look, your job is to use this operationally. Right now, I will provide you with a tooling. I'll make it as easy as possible for you to use. I'll provide you with the training and enablement. I'll support you if you have any questions on how to do anything or not. but ultimately it's not my job to build reports and dashboards for you that's your job right because you're going to use it so it needs to be as tightly bound to the way you work as possible um and so that's why we've really focused on on on being that multiplier on making it as easy for the end user as possible and giving them as much freedom as possible
00:18:27
Speaker
and also saying and and this is probably the hardest thing like if i'm not responsible then i'm not responsible what that means is that like if you go and you build a measure and it's looking at the business in the wrong way that's your problem right and your problem to fix and that everybody is comfortable with that and that's how we work and that's fine and and having showing that restraint is important right because once you start involving yourself again you become responsible and then you're the blocker on the business right and that like the We've grown incredibly quickly while I've been here.
00:18:59
Speaker
um And I don't think we would have been able to support that kind of growth if we also decided we needed to be embedded in all of the teams. It's a really, really interesting point and something, yeah, as AI is becoming more and more prolific and smarter than um then we can imagine, I think the it's all about the framing of the problem and asking the the right question um as we move into this AI analytics area. I think they get what you're saying, that the the business stakeholders, the people that are in their teams and dealing in their day-to-day
00:19:34
Speaker
challenges they're the ones that can hopefully ask the better questions maybe then than than the ah the analysts is that sort of the the idea yeah definitely and i think that um like you have to come at it with a
Respecting Stakeholders' Expertise
00:19:48
Speaker
ah kind of a humbleness and respect of other people's world right and that like you know I remember I've had times in my career where someone more senior who's not worked in data has said, oh, you know like can we change our tool stack in this way or get this or whatever else, right? And I felt um a bit insulted by someone telling me how to do their job, but also I felt they've just in general not made the right choice. They've not had the right information in front of them, right? and They don't know. And that there's something the same with like, if you bring in someone who's ultimately a data person first and you ask them to build a sales dashboard,
00:20:23
Speaker
they're not a salesperson or a sales manager. And so they they just don't know or quite know quite the same way. And they there there are subtle things they would or wouldn't look at, but I think those are important.
00:20:34
Speaker
um And so I think you have to take the craft of other people's experience seriously. And you have to accept that they ought to be running their teams and they're responsible for the ultimate outcome. And so they should therefore be given the the kind of the freedom to to work as they want.
00:20:54
Speaker
One thing I think that if I was another data leader listening to this that I might question is Synthesia is arguably one of the most exciting companies globally, hires very analytical tech focused people across the business I would imagine.
00:21:13
Speaker
What you're talking about is putting quite a lot of onus on obviously the the other teams. How have you found that enabling them and their uptake on made building dashboards being a bit more technical? um yeah has Has there been any blockers there? Because i know there's definitely organisations where they yeah the the stakeholders wouldn't wouldn't have the the know-how to to maybe do that.
00:21:35
Speaker
Yeah, I think it's like identifying the the bits of the business which you need to work with closely. So um there are definitely people who wouldn't be able to build dashboards and reports. And I love Omni. I think it's a great tool. I don't think there's any BI tool that's easy for the long set of users to use.
00:21:52
Speaker
um But then I think it's identifying and making the case for those teams to then be responsible for resourcing against it. So we have a very strong RevOps team here.
00:22:05
Speaker
um I think all of the Redbox team are comfortable writing SQL and working with data show and whatever else. And they build out a lot of it, but they are the business partners of the commercial team. Most of them have worked in commercial roles before and they sit in on all commercial stuff in a way that we don't. Right. So like from my perspective, if...
00:22:27
Speaker
if i have someone who's joining all the forecast calls who works on sales training and enablement who used to be a salesperson who is actually the the sales ops person for me that's not like they can build a dashboard fine right the point is that like it's up to the sales team to decide do they want a sales ops person doing that or do they want the sales managers doing that or whatever but that's their choice to make right if i'm in a function if i'm in a place where No one in that department has that skillset. So if it's the case so I'm, I don't know, i'm ah I'm a data leader and I'm working with a commercial team and none of the commercial leaders feel comfortable working with data. And also there's no operational function working with data there.
00:23:15
Speaker
then I think my job is to try and make the case that they need to hire a an operational person who has a skill set rather than that person be a data person. and At the start, there might be dotted lines and whatever else, but they kind of have to really live on that side of the house, right not be sort of seconded over there or like a spoke. They really are a member of that team first and foremost.
00:23:40
Speaker
interesting how how how did that um pan out at synthesia is like is this something that that that you pitched and sort of is this sort of the vision you had from day one or or is this something that sort of grow um sort of embryonically or yeah and how would you position that again if you uh you said obviously you know you started that out of a team to make ah a higher in that space i'm thinking about how would other leaders Maybe they're listening and thinking what you're saying sounds really relevant, but how how' do you actually get there?
00:24:11
Speaker
mean, I'm i'm lucky that Tony, who's the sort of the senior director of RevOps here, he's very technically minded and and also wants to work that way and also wants all the members of his team to be able to work that way.
00:24:25
Speaker
um And then I think the other thing is I was very clear about this from day one, right? That like, this is what I think I am and I'm not responsible for. And in ah In a weird way, people quite like it when you're quite direct about what you're not going to do, right? Particularly if you say alongside it, and you know, I don't want this resource, I don't want this, don't want whatever else, right? like it I'm not saying um just, look, I'm not responsible for this, I'm not going to do it. I'm saying, you know, if you want me to be responsible for dashboards and reports and whatever else, ah you need to hire me a team of six analysts or whatever. But I don't think that's a good use of my time or your time or whatever else and the right way to do it.
00:25:09
Speaker
I want to run a small team and have that live outside, right? So I think i think you've got to... give up some head count and give it some resource to do it, right? Because you are ultimately saying you're responsible for less.
00:25:22
Speaker
um And then make the case that those people fit better on the other side. And um in general, I find that people are quite receptive to it because people want to own their own number, right?
00:25:34
Speaker
Like if you're a sales leader, kind of nothing is more frustrating than you having a view on the business and someone else telling you that actually their view is the view on the business and is correct and not yours, right? That must be incredibly infuriating.
00:25:48
Speaker
So I think people actually are very happy for the people who have the view on the business and who support them to be very close to them. I think that's an excellent way of framing it and yeah, some really good tangible, the body that actually don't.
00:26:03
Speaker
You've we've mentioned a few times about your stack and your Omni's come up um quite a bit. um You've been a customer for two years or so now as well as one of the first things you got in nu How has Omni changed how you guys and the business interacts with data and particularly as we move into this sort of AI analytics space um um and the the self-service that that
Omni's Role in Scaling and Metric Management
00:26:29
Speaker
that can unlock? It'd be good to help understand what transformation that's been.
00:26:34
Speaker
Yeah, at first, Omni was what enabled us to really scale out super fast. So um what I love about working at Synphysia is the kind of the freedom and the pace. And I was used to ah a slow environment. And when I joined, I put a roadmap together for BI tooling. um And I imagined I'd spend...
00:26:52
Speaker
A couple of months, you know, getting all of the requirements or whatever else, and then a couple of months doing vendor evaluation and then probably three to six months of onboarding. So um I joined in late April two years ago, so 2024.
00:27:07
Speaker
And so I imagined that we'd be going live with something in January, February 2025. we actually ran the ball pack out of omni in july 24th so like you know we just went straight in super deep um and that's because part of saying to the rest of the business is you ah you're responsible is it's saying that the the mart's layer or your kind of your your sort of like your warehouse obviously belongs to the data team and that's clean but then metric definition that in general actually lives with the business right like i don't really care when we're when we look at our um time to close here we take it and we compare the sales qualified date to the close date and we only look at opportunities above a certain minimum threshold um and we also exclude certain deal types and whatever else right and like
00:28:04
Speaker
i I couldn't care less about that, right? Like I don't care whether or not the right threshold is zero or 1,000 or 5,000 or 10,000. I don't care if we blend new business and upsell or not, whatever else, it's not my problem, right? um That ultimately is however the sales team want to do it so that they can understand their business to run it better, right? Obviously, um finance might have a different opinion, but that is for finance and sales to negotiate and sort out, right? like It's not my role to play parent there and choose one that's right or whatever else. And so Omni, where it's different to other tools, is it gives the end business user the ability to edit in the semantic layer.
00:28:43
Speaker
And that meant that particularly with a kind of a data literate rev ops and finance teams we have, that we could say, look, here are the end assets. you go build what you need, build a semantic layer, and then in time, I'll sort of incorporate it from the edges into the center.
00:29:00
Speaker
And so we're able to in that way mobilize the entire business into building out Omni rather than it being a data team problem. And that people are know they're responsible for their own parts of the semantic layer. um And it lets you move very fast, right?
00:29:14
Speaker
It also means that you end up with semantic layer that's very tightly bound to what is actually being used, right? Like people, it's not, you know, the way it used to work with Looker is that we would go and say, oh, what are all the things that we think we need to look at for contacts or opportunities, whatever else, and build that out. And then it wouldn't be quite right. And there's a slow process of feedback and whatever else. And it's just this big ongoing mess. And then...
00:29:39
Speaker
somebody wants to add a new i don't know they want to add a measure for um mql from tier one countries and then someone who's got to do it whereas here it just kind of happens at the edge it happens very fast um but as you say with ai tooling things are changing so ah the the challenge with that and um it definitely fit us at the time. And this was also before sort of AI tooling, like coding and stuff really took off, at least in in our team, is you had the challenge is you end up with a very messy semantic layer, right? And in a way, I don't care about that because I'm the...
00:30:16
Speaker
The benefit to a clean semantic layer is just that it's clean, right? Like the most important thing is that it's used, right? And the people who are in the tool and we have very little in our business that happens in spreadsheets. Almost nothing happens in spreadsheets, which is pretty rare.
00:30:29
Speaker
um But with an AI tool, that AI needs better context and it needs essentially also access to everything and everything's not to be in a shared model. So if you've not used Omni, if someone goes and creates their own measure, they don't actually have to create a PR and promote it. They might just keep it in their workbook and that's kind of fine for them.
00:30:48
Speaker
But the AI doesn't know if that exists, right? So we've now had to go through a process of rewriting the entire semantic layer for AI. um But we found that with cursor we can create a new topic in 10 minutes and actually with this entire existing base of reports that are heavily used it's very easy to work out what we should be building so we were able to rewrite pretty much our entire semantic layer in a few weeks.
00:31:14
Speaker
which is which is nice that's crazy yeah and and it sounds like these think one of the other things that you've you've brought up to me as well is this um bias to to speed over um you know 100 accuracy and on and guard rails and and not being like you referred to earlier not not being ah a blocker how how is that How has AI been ah an enabler of that and what challenges has it has it caused? Because I think some of data teams obviously are very cautious about you letting their stakeholders self-serve and getting the the wrong numbers. what What's been your view on that?
00:31:50
Speaker
Yes, I mean, it's let us go very fast. um I think that um people are always going to find the wrong numbers and so actually it's like what happens when they find the wrong numbers both like you can have a culture of sort of blame and anger and whatever else or you can kind of lift it out and be that it's normal and like one thing we really try and do is have a very open sharing culture and one where like if someone's got the wrong numbers that's fine they've got the wrong numbers right right we'll get them the right numbers and they're sort of try and take there's some anxiety and some stress and some tension and some conflict that can be there and the earlier the wrong number surfaces and the less people use it or whatever else that isn't there right like obviously if
00:32:38
Speaker
a sales leader spends two weeks preparing a deck and a finance leader spends two weeks preparing a deck and they have very different conclusions and they turn up to a meeting and they've one, you know, one of them's wrong. That is an issue, right? But that's an organizational issue about these two people not speaking to each other and an issue like you know that and isn't actually a wrong number issue in my opinion, right? A lot of other things have gone wrong. That's not something that we face by the way, but like it's, you know, the kind of story you hear.
00:33:06
Speaker
um but So I think it's partly like an organization thinking about how you react, but then something that we're trying is in Claude, you can kind of speak to Omni using our connector.
00:33:19
Speaker
And what happens when you get something wrong and you ask it to be different or whatever else, we've got a Claude skill that prompts you to change the AI context and then raises a PR with a data team, right? So had one yesterday, which um it completely...
00:33:36
Speaker
I know blew my mind right so we use the Salesforce case objects a lot and different teams use it very differently so there's sort of managed onboarding cases which one team is using the kind of a deal desk you're using it for deal reviews and they're using it in their own way and then we've got a proposals team that uses a proposals team there are people who deal with things like ah RFPs um And in general, you use the close date for when the case was closed or competed completed, right? That makes sense.
00:34:06
Speaker
um So that's kind of how we built it out. And then ah we launched Omni and Claude and one of our proposal managers said, hey, how many um how many ah RFPs did we complete in March? And he got the wrong number because for his team, for whatever reason, and You know, it's not the decision I would make, but it's not my decision to make. They used the created date, not the closed date.
00:34:27
Speaker
So he got the wrong number, right? And the skill auto fired off and it said, you know, well, what should it be? It should be the created date. And it raised a PR and a linear ticket for us and said, you know, We've changed the AI context for, we're like, oh, well, we we need multiple topics because the data is different. So we'll have a topic for the proposals team. It's separate to the topic that the DLS team use. And in the proposal team's topic, in the AI context will say, when someone asks for the number completed, use the created date.
00:34:58
Speaker
um And so that's the other thing with the wrong numbers. It's like, can we surface out fast? And then can we have the user in ah in a way that doesn't cause them any pain, tell us what needs to change? And then that way at the edge, hopefully it just gets cleaner and cleaner, right?
Agile Approach to Data Management
00:35:14
Speaker
That's a fascinating story and I think another sort of link to the systems and the process that you've built in to the team and the technology to be able to have that that that agility as well, to be able to get that flagged automatic PR sent and to make them changes, you know, ultimately sounds like in, yeah what, minutes?
00:35:34
Speaker
Yeah, it took few minutes, yeah. Amazing. Really interesting ed um I suppose one of of the other areas which I was keen to touch on with you and it's sort of what we've been alluding to but i just is career development. um You've had quite an interesting career and have got this view I think which is really relevant about where data is going. um If you were someone in their career today, say so starting out as a yeah as an analyst, as ah as an engineer, what where are the what the areas that you'd most advise someone to to really push and develop and to to help them excel in in their career?
Responsibility and Impact in Data Roles
00:36:16
Speaker
i think you've you've got to, ultimately, if you want to grow your career, it's about taking responsibility for things. And so you've got to be looking at at things you can do in parts of the business and outcomes that you're responsible for and growing those outcomes and then being able to demonstrate that those outcomes are kind of going well, right?
00:36:40
Speaker
And so I think that's quite different to... um A lot of people want to sort of like be in the room with senior people or sort of like...
00:36:55
Speaker
And that's fun and interesting, exciting. And I definitely learned a lot from doing that. So I think that there's definitely a big upside that and your understanding of a business and how it works and whatever else. But there' there's a kind of a natural limit to trying to be senior by being around senior people and sort of like being in the conversation.
00:37:13
Speaker
Like actually, if you can turn around and say, I'm responsible for this and this is what happened, then I think you will sort of go far. People want to give you more responsibility, right?
00:37:25
Speaker
And in that way also just taking the initiative, right? and And like trying to play on the front foot. And that means understanding what the kind of practical applications of a data set or a data kind of stack might look like.
00:37:39
Speaker
And then proactively saying, right, well, this is where we're going to grow our capabilities. this month or this quarter or whatever else and doing it and measuring yourself by that like how have the capabilities of what you've done changed right how how can people in the business act now and where they couldn't act before and if you can act answer those questions well then you're adding value to the business and and you'll ultimately be rewarded i think that can be more and I think tracking that and changing capabilities and the outputs you talked about is ultimately what to then be able to bring up in either your performance review or further interviews. I think that's something that we're definitely seeing missing in a lot of people and in the industries lack in general is the ability to talk about the impact that they've had an organization as a data professional.
00:38:31
Speaker
um Flipping that there, e that was obviously for someone that's sort of been there you know in the weeds of starting their career and developing their career early on. What if you're a head of data a director or VP that's trying to increase their sort of team's leverage in the business and impact that they're having? what Where would you actually be putting your energy and and what should they maybe stop focusing your on in in your opinion?
00:38:56
Speaker
I think it's it's starting with um sort of what are people doing in the business, right? And that like,
00:39:07
Speaker
as you sort of like, if you're a head of data or VP or whatever else, or you should have a good working knowledge of like how a business functions and operates and what is kind of happening.
00:39:20
Speaker
um And then from now, it's ah it's about identifying opportunities for for people to use data, for for the data itself to play a big role, not even necessarily your team,
00:39:33
Speaker
right, but as a whole and then i think building that platform, right, and sometimes this is making the case to someone else that they should do something differently, right, or it might be championing a culture where people are sharing the examples of what they're doing well or whatever else, um but understanding, you know, like you you have an opinion about how data could be used, so I think it's taking that and making that case, um and then in terms of your own work,
00:40:00
Speaker
that the reliability of the platform and the scalability of it, right? Like that that idea that like you are not the blocker is something which is kind of table stakes, but I think definitely slows a lot of teams down um if they're kind of focusing too much on the sort of interesting projects.
00:40:20
Speaker
Excellent. um So before we wrap up, Ed's, I think it would be great to just touch upon Synthesia and where you guys are are going.
Future Goals: Unstructured Data and AI Scalability
00:40:31
Speaker
what What does the roadmap for for data look like over the next stuff and the act was years yeah years to come?
00:40:40
Speaker
Yeah, it's it's hard. i think that... um AI tooling is really going to change the data world. And I haven't quite worked out in my own mind, and I don't think anyone else has quite how this works, but I think there's sort of two big changes that are kind of hat up happening on their way to happening. So one is a massive increase in unstructured data and sort of documents and stuff. And that like, um when,
00:41:15
Speaker
I don't know. We have... um One of our CSMs built something which um reads through old customer value stories in the slide decks and the Notion, and then it goes to Omni and it gets some usage data, and then it goes through emails to get sort of like other bits of information. It puts all together to put like a draft value story in place for a customer and send it out. It's really cool and really impressive.
00:41:40
Speaker
um I think how we... programatically structure those google slides which have all the value stories in and tag them and whatever else for like fast retro like retrieval is like a really interesting data problem like almost like search as a data problem i guess i'm not exactly sure what the right answer is but i think that's one big focus for us um because i think it will unlock a lot of value like you know every time know we have a ah monthly performance review for our go-to-market teams
00:42:15
Speaker
we should be, when we're building that, like actively querying all of the previous ones, right? And understanding it in a way that we we kind of are right now, but do a lot better. um And similarly, like i think that the that the other big thing, I think, is the skills files or equivalent for AI tools. So it's almost like we now have two semantic layers. So we have one semantic layer, which is how Omni speaks to Snowflake. But there's a second one, which is how Claude speaks to Omni.
00:42:49
Speaker
And we've got, as I mentioned, you know, we've got a skill file that automatically suggests changes to the code base when Omni's struggling. um We've got one skill file that kind of teaches Omni how to, you know, try a second time if it finds the wrong answer or whatever else, right? Like if somebody...
00:43:06
Speaker
um asks to find out out about, I don't know, be example like, um I don't know if they're a customer of ours or not, but ah you have some sort of European kind companies which have got kind of accents in their names. So if you search for Nestle, well, the Nestle has actually got a little kind of hyphen above the right? So a search for Nestle might pull up nothing,
00:43:30
Speaker
like it like it will be in the skill file that it knows to kind of ask that again when it gets it wrong and find it and whatever else and that there's probably something similar with how interacts with notion or whatever else and that at the moment i find that the process around building maintaining scaling all of that is kind of a bit broken as well right like all of those skill files should live in a repo they should be standards for how you make them they should be you know useful um These are important parts of the stack. There should be kind of PR reviews in order to kind of change whatever else. right so um
00:44:04
Speaker
I think for us, the big bit is how we handle all of that. And then finally, this isn't for us, but I think we'll need to support is we've let people go crazy and clawed and they're doing really cool stuff.
00:44:18
Speaker
And we've got something where, um actually, I suppose it does affect us in a way now, I think about it, where, don't know, one of our SDR managers he's got something that runs for three or four hours and it ends up producing amazing one-to-one notes and notion with all of his team. With not just sort of how they're doing in the quarter, but like which of their peers are like particularly great in different areas where they might partner together, which accounts based off the accounts that have been successful recently, should they be working? Whatever. So it's having really good um effects. Like he is He was 150% to quota as a team last month, and he says that this played a big role, right? And that's what I mean by being operational, right? Ultimately, we provided the tools for him and his creativity to do better at his job than he would have done before, right? And like, I can't say that that particular sales qualified opportunity is due to my team, but I can say that some percentage of them must be, right? If my team weren't there, weren't working the same way, it wouldn't have happened.
00:45:14
Speaker
Um... But if that runs for a few hours every single time, I'm almost certain that like it's either like hammering snowflakes somewhere or it's hammering Claude somewhere or some other AI tool. And I think the other thing is, like how do we let people go crazy and sort of do all of this, but then at some point say like in a stable state, this is this is how something is scalable, right? And apply data engineering theories and practices to how we use AI tools to set up like our token spend or whatever else is is under control. So those are some of the things we're thinking about.
00:45:53
Speaker
um Hopefully with AI tooling, we can do that with a pretty lean team still. Sounds like there's a lots still lot to do ahead and to you and just at the the start of the building, which is is super exciting.
00:46:09
Speaker
Ed, I think the final thing to touch upon on Synthesia, you're ah we're hiring at the moment. I'm sure you're going to continue to hire. Who's the type of person that really fits in and thrives at Synthesia, at least under your data team?
00:46:29
Speaker
I think you've got to really be able to embrace the autonomy and the sort of yep the chaos almost, right? Like we, the way we work is to give people a lot of freedom and to sort of like have them just go at it. And some people love that and some people don't. And um if you need to be tracking your tickets or whatever else, and you need a lot of structure and you you need to know for certainty what you're working on in the month, it's not the right place for you.
00:46:59
Speaker
um And then also you need to be able to let go, right? Like our business users are going to go and build a report with a wrong data in, right? and And that's fine. That happens, right?
00:47:11
Speaker
And if that makes you too anxious that you feel the need to go and correct it and whatever else, is not going to work. So you need to be able to respect the boundaries we're setting up as a team and and really trust and business users and have ah have a mindset that says, you know, like they're responsible for what they do and I really want to see them succeed. But if they fail, they fail. And that's that's on them, right? Rather than like a a feeling that you sort of know better and that you you want to involve yourself.
00:47:42
Speaker
Amazing. I think that's a really good summary of but yeah the people that that you've got in the team and and why they're doing so so well. So, Ed, before I let you go, it's been a pleasure to speak with you and and get your views. um but What would you say is probably one of your most controversial views on the the data industry that maybe most don't agree with? And that might be what it maybe where you you think it's going or challenges that it has has currently.
00:48:10
Speaker
um I mean, we've touched on it, but I think my most controversial view is I think that a lot of data people, particularly semi-junior data people, to be honest, I think they don't have enough respect for their stakeholders.
00:48:25
Speaker
And I think they kind of think because you know, if you're doing well in data, you've probably been pretty smart and done pretty well and whatever else. I think that they they look at a ah sales leader and they see someone who they think isn't isn't as smart, as good at them.
00:48:42
Speaker
And they think because they understand something on the dashboard, they think they kind of get it and get it better. And um I think a lot of data people imagine themselves as mini-CEOs or something and I think that it really it really holds them back.
00:48:55
Speaker
So um I think that working well with stakeholders means really treating them with respect and they're what they're good at with respect. um So that would be my my controversial opinion. It's far too often I see people who don't do that.
00:49:09
Speaker
I think that's a excellent one. And so we've touched upon a lot and in the podcast over the the three seasons that we've been running now is the ability to partner well with a stakeholder is is a true differentiator. And think what you've mentioned about having that respect and empathy for what they do is is and ultimately why it's what built people build a trust on a relationship and a partnership on. so Controversial, but I think very, very true. So um yeah, ands it's been a real pleasure. Thanks so much for for joining me. It's been a long time coming. um So yeah, thank you again.
00:49:45
Speaker
Cheers. Thanks, Henry. Always great to chat. Brilliant. And that's it for this week, folks. We will be joined by Gina, also from Synthesia, at a later date, to really take a deep dive more into Omni AI and Claude and all of the stuff that Ed's mentioned as well and on a much deeper level. So keep your eyes peeled for that one. But for now, thank you and goodbye.
00:50:06
Speaker
Hi everyone, just a quick one from me. If you've enjoyed today's episode, I'd be so grateful if you could hit that follow button or leave us a rating. Even better, pass the show on to a friend who might also get some insight from it.
00:50:19
Speaker
It really helps us grow the community and continue to share amazing conversations. I also wanted to take a minute to talk to you about Cognify. For those of you that don't know, Cognify is the leading recruitment partner for modern data teams.
00:50:33
Speaker
We help some of the world's best organizations scale data and drive real value from the hires that they make. If you're thinking about building a team or making a hire and you're struggling with talent or just want some insights on the market, then I'd love to jump on a call with you and tell you a bit more.
00:50:51
Speaker
Equally, if you're looking for a job and want to find your next dream role, then reach out to myself or any other Cognify team. We'd be happy to see if there's anything on our books that we can help you with and give you general advice on the industry.
00:51:04
Speaker
Finally, big thank you to Omni, this season's sponsor. If you'd like to learn more about the AI analytics that Omni can deliver you, then check out the link in the show notes or come speak to me. i can happily point you in the right direction.
00:51:17
Speaker
Again, thanks for listening and look forward to seeing you a few weeks time.