S4 | E26 | AI That Works: Practical Value, Enterprise Trust & What Comes Next – Cassie Harman @ Nitro

It's a product I use every day. but actually what's inter yeah, no, I do, and it's interesting because the more you dig into it and then actually looking at the website and looking at the brief before this, there's a huge amount more to the product that maybe some people give it credit to. So what I like to dig into is first and foremost, who are Nitro? and and ultimately what led you to join them what a couple of years ago, two and a half years ago?

Yeah.

Yeah, two and half years ago. So Nitro provides document productivity software that helps streamline our customers workflows, making them simple, smart and secure. So that's essentially who we are. We offer PDF editing, e-signature, smart redaction and automation tools, essentially. I think, though, in answer to your question, what really attracted me to Nitro was

just the size of the opportunity really and being part of a challenger brand.

eating with really big legacy incumbents and just seeing the opportunity to provide a really great user-friendly product that is pretty ubiquitous. We are in a lot of different industries across a lot of organizations. as you said, I've been building digital products in...

highly regulated industries for the last 20 years, like financial services, healthcare and energy. just those are the customers that often need nitrous tools. So that particularly appealed to me.

Mm.

Interesting. I obviously won't mention the company name by by name, I don't want to give them any promotion. But obviously you are competing with a lot of very well known companies in that kind of document management, redaction space, e signature. So when you came in as as chief product officer, what what was the first thing you looked at holistically as saying this is you know, this is where Nitro is and this is where you want to get to?

Yeah.

Well, I think in terms of, you know, I try to take a very customer-centered, user-centered perspective, looking at the opportunities and how our users were using our product. And I think that, you know, the thing with those kind of...

behemoths is that often they stop listening to their customers and they get further away from their users. So there's definitely an opportunity to really listen to customers closely about what they want. I think it's been particularly interesting as well. You know, we're at a time of great change with software in general.

If you go back just two and a half years when I joined Nitro, there was still, you could see that there was a huge opportunity for AI features and AI powering workflows. But still at the time, a lot of enterprises were very reluctant to adopt it because of security concerns and the like. And I think it was really important that we got the balance right between

meeting the needs of our customers then, today, but also making sure that we were thinking about the future, so balancing it with the future-facing things that we knew would solve their problems, that we knew there would be demand for further down the road.

Mm.

Mm.

Yeah, it's really interesting. I think the the company or the enterprise adoption of AI or tooling on the whole is is has ultimately changed. When you were coming in and obviously looking at the customer feedback and obviously looking at the product itself, what was some of that feedback you were getting back from those customers and what maybe were some of those barriers which ultimately you had to overcome as a go to market initiative or as a as a wider product org?

Yeah, well, I mean, think the first thing is that Nitro has always, we've had a sort of unofficial tagline of, you know, similar features, half the price, twice the love. You know, we've really made that kind of customer service piece front and center, and we wanted to continue that whilst also making sure that we were updating.

Okay, nice.

our product in the right way. think that, you know...

we were acquired several years ago by a private equity firm. So it had been a renewed interest in how can we really make this just the best possible product it could be. And so definitely diving in, listening to the user feedback around particular use cases that maybe were not fully addressed or things that could work better was definitely the number one focus.

Okay.

Yeah, and you're on you're in a really super hot spot, right? Because obviously there's so many tools, applications, different initiatives to kind of change workflows and obviously AI is front and centre of that, but ultimately it's the more practical everyday uses of AI that are delivering the most value for organizations. Why do you think that is? And do you think we're do you think organizations run the risk of almost getting too carried away with

trying to solve a problem they might not necessarily have yet, 'cause obviously the more practical applications is where obviously you sit in day to day, that's the not not so much the the less sexy side of it, but it's just not something you n naturally would look at.

Yeah, I mean, that is the thing. think genuine user problems and enterprises are rarely the sort of the sexy headline grabbing ones. And I think talking about the document productivity space, specifically the example that I always kind of pounce on, there's this feature where you can turn your PDF into a podcast, which is great sizzle for press release. that's

Mm.

Yeah.

that's not really that useful for most people's day-to-day jobs. And if you kind of compare that...

to something like the ability to select hundreds of forms at once, press a button, and then it doesn't matter if, you know, some of the data and the forms is handwritten or some of the fields blank, you know, whatever, just to extract all that data from all of those forms to a spreadsheet automatically in a matter of seconds, you know, that's just save someone at an insurance firm or, you know, a company running clinical trials or someone.

processing mortgage applications, hours and hours of manual effort and eliminated, you know, human error as well, the possibility of making typos or, you know, transcription errors. Is it exciting? Yeah, it depends on your definition of exciting, I suppose, but I think those are the...

Hmm.

the real time saving valuable applications of that kind of technology. And I think, you know, to give you a sense of the scale of it, we worked out that our customers in the last 12 months spent 29 and a half million minutes manually redacting documents. That's around like 56 years of manual. So I think the size of the automation opportunity is huge. And, know, and that, as I say, that's really what attracted me to Nitro.

Wow.

Well, that's a wild stat. And it actually is probably a nice segue into. I'm really fortunate to interview great founders and leaders from startups and tech companies like Nitro. And one question I always ask is how do you measure the ROI or the effectiveness of the new AI tool? And actually, the one thing that always comes out is it's a very hard, but mainly it's down to time. It's down to you know time saving. so obviously moving from that.

impressive demo, you know, which a lot of these companies c roll out, you know, their go to market folk and then moving into the actual adoption and day to day use of the product. Why do so many companies go really in early stage on a new tool, get very excited about it, but then drop off rates and user rates tend to fall fall off a cliff. And and what what does Nitro do to ensure that doesn't happen?

Yeah, I think it's a real problem today, actually, and, you know, partly because it's easier than ever to find a proof of concept that looks absolutely great in a demo. Reductionizing it is far more difficult and you've got several problems to solve. You've got scaling and performance, you've got security and data privacy. And then also with AI, you've got consistency. So

Mm-hmm.

Yeah.

Hmm.

You can ask Claude to detect and redact the EII in a document. It will find an open source tool to do that.

But the results aren't going to be perfect, probably. And you actually get slightly different results each time. And if you kind of contrast that, you can still use Claude as the interface. If you've got an MCP integration with a proper enterprise-grade set of APIs, you're going to get an excellent, consistent result each time. you know, if you're looking for adoption, you want software that's going to give you consistency, that's performant, that's in a secure framework.

Yeah.

where you're not putting your data at risk. And obviously getting it wrong for enterprises has got major compliance consequences. We actually ran a survey on AI use in enterprises and security, but an organizational level was actually the top barrier to AI adoption. So it was cited by, think, 54 % of managers and 49 % of execs. But I think the adoption piece

Mm-hmm.

You have to make sure that you're actually solving a problem for your users. I think if you're not doing that, then they're not going to use it.

Yeah. You're completely right. Yeah, there's so many shiny tools and I've been guilty of it when we I I run a recruitment company and the next thing that comes out, we must sign up on the free trial and you use it for a bit, and then you go, Okay, we don't really need that. And then the the renewal comes and it's slightly more expensive, and then you actually look at Claude a bit closely, look at your current setup, and yeah, i i I think everyone's guilty of that. And but you obviously are handling in most cases sensitive documents and you touched on something really interesting about kind of

security, privacy, governance, which we're seeing one of the a biggest opportunities for companies right now, but also one of the biggest barriers for companies adopting AI. How do you approach that from a guardrail standpoint, from a governance standpoint, or whether it's directly in the product or the advice you're giving to your customers?

Yeah, well, I think that you've got to make sure that you're not using customer data to train models. That's of number one. And I think making sure that you're compliant with all the regulations that are relevant for the industries that your customers are in. I also think it's really important not just to sort of think about InfoSec, but also ensuring that security best practices are

integrated into the software development lifecycle process, because I think, you know, you can tell the difference if somebody hasn't been thinking about security or data privacy until they've sort of started building the product or feature that shows, and you have to sort of retrofit it. That's never as good as when it's just a part of

Hmm. Yeah.

the software development process right from the start. I think it then becomes much more seamless and baked into everything you're doing. Yeah, so probably that.

Yeah, no, that that that's fair. And I think it's also the education piece internally, isn't it? Because obviously you're you can adopt this, but if you've not got someone championing this internally, then you know, again it's there's still this nervousness and the reliance on you as the vendor, which ultimately shouldn't be. 'cause obviously in internally they need to make sure they've got the right data setups and data environments, which yeah, I completely agree with. And I I'm I've obviously I've just come back from San Francisco, which are really interesting speaking to people over there around this kind of

100.

big push for companies dropping AI off the kind of name of their brand and business because obviously there's this everyone's using AI or it's AI powered. In your experience in product and maybe you're advising customers or just reflecting yourself, how do you tell the difference between what genuinely is an AI innovative product that's gonna change you know change the game here versus a wrapper versus something that's not as attractive?

Yeah, I mean, I think you can, you can tell fairly quickly. The question is whether the products, you know, doing something new, doing something you couldn't do before. But the second question more importantly is whether that thing matters to the customer. I think that there's a lot of companies, you know, as you say, that are kind of hoiking their prices up supposedly because of all this extra value they're providing with, with their AI tools, which often, by the way, is just like chat GPT with a glow up.

Yeah.

Yeah.

But customers aren't adopting them or finding them useful. And sometimes that's just because, you know, they're very early on and they haven't, they're not sort of particularly accurate or they haven't got as good as they can be. But that can also be because they're not really focused on a problem that the customer has. And with those price rises, I think in that scenario, over time, customers are going to vote with their feet. I think, that that pressure is particularly fierce on the big

Mm.

publicly traded SaaS companies, they haven't had to be at the cutting edge of innovation necessarily in the past. And then all of a sudden they're being pressured to show evidence, not just that they have AI powered features, but they're successfully monetizing them. And that's, that's really tough, but, but yeah, ultimately I'm always going to bring it back to the same question. You know, what customer problem is this actually solving? That's going to tell you immediately whether it's real innovation or whether it's just AI washing.

Yeah, and that's certainly coming out in the wash more now for some of those companies maybe had huge valuations and what's what's now happening, especially token costs and infra costs, and everyone's starting to see hang on a minute, what actually are we paying for here? so I think it's a really interesting point. But you and Nitro have actually taken a purpose led approach around AI disappearing into the product though, rather than kind of saying we're an AI platform, an AI product. Why is that and what what's been the benefit of that messaging?

Yeah, well the first thing I would just say is that there are...

compliance reasons to make sure that your user knows what's AI powered and potentially also cost implications and transparency is very important from a trust perspective. in that sense, it shouldn't be completely invisible to the user, but certainly in terms of AI becoming just a frictionless part of the user's workflow. Yes, absolutely. know, our CTO at Nitro likes to say that AI should

be an implementation detail. And I think that that's really nice way of looking at it. We be making it all about us and our cool tech, you know, we should be making a bit about you and how we can help you get your job done quickly and easily. And ultimately, whether it's AI or

Yeah, fair.

Hmm.

a hamster wheel and some elastic bands that are powering a feature. In theory, it shouldn't really matter to a customer as long as it works well and it's secure. I think ultimately, you you want to reduce friction for the user to ensure a better user experience and also to drive adoption. And so those principles have definitely been front and center as we've been making those product decisions. But I think there's also

a benefit to the organization overall in that approach, because we know there's a lot of shadow AI use in most enterprises, even where there are policies against it. More than half of the managers we surveyed said that sensitive documents are being uploaded to public AI tools within their organization. creates risk and compliance issues. the fact of the matter is, tend to optimize the

Mm.

convenience, know, that's, that's, and the approved tools are accessible from, where you're already working from the applications you're already in, you're more likely to use them and you've got less reason, at least if they work well, that's a separate, that's a separate issue to take the extra steps and the extra friction that, you know, opening up your own personal chat GPT app and copying and pasting content into it with, with need.

Course. It's a creature of habit, aren't we? Yeah.

Yeah, I say yeah.

Yeah, it's quite a terrifying situation I think we find generally as a as a as a global community and user of AI because I've had the same feedback from a customer you're working with. They've gone all in on AI, they've got their internal models, though, but actually they've been finding out the team members are basically trying to bypass that and just uploading, creating content through ChatGBT, but then it's causing issues. So d where does the onus lie then? Because that's a really interesting point. Is that on

Internal education is that ultimate or is that putting guardrails in place where effectively you're blocking the usage to open source models? What what's and more of those frontier models? What how does it work in practice? Because I think from a s security and privacy standpoint, if I thought my data was just being uploaded willy-nilly, private health data up to Chat GPT, I'd be livid.

Yeah, yeah, terrifying, isn't it? I think that it has to be, you know, balance of all of those things, to be honest. So I think that, you know, it makes sense that there are policies and, you know, that there's enforcement of those policies. But ultimately, I don't think that's going to be the thing that makes the difference. think, you know, organizations have to, you know, do the responsible thing there. But I think that also

Mm.

it has to be about utility and usability. So where people are most inclined to use shadow AI.

kind of go around existing processes is whether doesn't exist, you know, a good alternative because ultimately people are focused on getting their job done. That's what they want to do. And so they're going to do that however they can. So it's sort of in, you know, leadership's best interests to make sure that they're sourcing tools that help people get their job done, you know, quickly and effectively, I think, and making sure that those things

again, as long as they're secure and they're, know, they boxes there, that those things are getting approved for organizational usage.

Mm.

Yeah, it's really fair. I think it's the yeah, the education piece as well. I I'm fortunate I speak to a few clients and people who actually part of their end of year reviews around AI usage and actually they're actually monitoring usage at granular level, which I I'm not sure how I feel about that, but that's these are big kind of SaaS companies who are now saying well we're gonna actually monitor your how much you're using our models and how much you're using internal. How do you feel about that? Is that something which you think is that a risk? Is that an opportunity?

I have quite strong feelings about this Alex, because I think, you know, ultimately again, I'm a product manager through and through, so I care about outcome rather than output. And what scares me a little bit is that if you're focusing on...

you know, on how much you're using AI rather than the quality of the outcome that you're producing, that is a sort of dangerous shift backwards. think, you know, just generally there's a lot of, a lot of conversations at the moment that focused on speed. That's absolutely great. And definitely the use of AI can, you know, speed development up massively. But at the same time, you don't, you don't

just want to put mediocre products out into the market and you're not going to get the right outcomes if you do. It's not all about speed. The other thing that you hear a lot about is bottlenecks. Like bottlenecks are always spoken about.

No.

They're everywhere.

So moving. But the thing is, yes, that is true. And I think that there definitely needs to be a conversation about you can't just expect to layer AI on top of your existing processes necessarily some things may have to change, or otherwise you it will sort of end up with with real bottlenecks. But all time what

Mm.

sometimes gets called bottlenecks are actually quality gates. They're actually the kind of human decision points where judgment is required, where prioritization is required that are actually really important, but make sure that you get the right outcomes. So I think, you know, we need to kind of have that conversation, take a step back and have that kind of broader conversation about what we're trying to achieve.

Mm.

Yeah, and I think it ties into that probably the broader conversation that's happening front and centre right now with those frontier labs around the rate at which we're adopting AI, pushing these new models out, and it's happening so quickly, that I think people's now expectation is we've just got to use it, we've got to get on it, we've got to go all in on this. Well actually, you know, until you have those guardrails in place, the training, the the expectation on your point, the outcomes being measured, then I think you're just gonna have a lot of slot being delivered.

100%.

Yeah. And I I really interested. I know look, job displacement's one thing which we kinda get away from. I I definitely believe and I'm in recruitment and we've seen this first hand here, there's a big opportunity here. Jobs are being upskilled, up leveled, that kind of entry level band is being almost sort of changed. but obviously your product, you know, document controllers, kind of data entries ultimately could do you know, could displace them all together. But

What's your view on kind of the next three to five years in terms of if we look to that kind of position that Nitro are in and how do you see that ecosystem looking and and and where do you think AI will quietly eliminate folk?

I think it's such an interesting and difficult question. I think that, you know, there's certainly been a plethora of very bold statements on the topic of, you know, what work AI is going to eliminate or what jobs. But I think that if you look at the

predictions that the Sam Altman's of the world are making. They're experts on AI. They're not necessarily experts on human behavior and organizational change. So far, I'm not sure that they've been that great at predicting the future. you know, we've seen the companies that have truly tried to replace people on a large scale with AI in any job where there's ambiguity and

nuance and where good judgment is critical, it hasn't actually worked very well. And, you know, I'm stunned every day by what AI is capable of. But then at the same time, you see it can't correctly tell you how many legs a photo of a chicken has, you know, and we're back to capability reliability gap. It's great until it's not. And I think, you know, sometimes as well, you see people equating a model benchmark performance on things like

Yeah, exactly.

passing the bar exam with AI's ability to replace a lawyer. And, you know, clearly there's quite a gap between the two in terms of what a person actually has to do every day. So that's where I think maybe, you know, AI has been a little overhyped. So I guess my prediction is I look at the next kind of three to five years, I think AI would take a huge amount of repetitive manual tasks away. I think it would be used to, you know, synthesize data and

Mm.

pattern match and the summarize and create and test code and solve problems. But I still think it'll need human oversight in that kind three to five year timeline. you know, we say that Nitro empowers you to work on what matters. And I think a large part of that for me is decision making. So I don't think human decision making is going to get eliminated anytime soon.

Mm-hmm.

Yeah, I'm glad you said that because I agree. And actually, I think let's be honest, there's certain parts of our everyday job that we just think, God, I wish I didn't have to do this every day. And actually Nitro's front and center of that, the redaction of documents, the scanning of PDFs. You think, remember sp years ago going to photocopier and just photocopying all of the CVs. This was years back. And thinking we had to scan that, we had to black that out. And actually you literally can use Nitro's product and it does it for you.

which I'm all for. So what's next for you guys? What what can people expect to see when they obviously have the podcast, they can they can see obviously what you guys are doing, but what new products and features have you got coming out?

Yeah, so we've just launched our Automate product, which is basically a set of really powerful APIs that can be integrated in with, you know, our

customers existing ecosystem. Also got a very, I'm very excited about this new app that we've got, which is sort of using those capabilities on the backend, but it's something that is for the end user. within Nitro, you can create your own workflows essentially. So if there's something that you do, you know, every day or every week or, know, whatever, you can create a workflow where that can be done automatically.

Nice.

for you and you can choose where you want to stop in that process and review at each step if you want to look at the output and what it's done and check that it's okay. Again, that human oversight and judgment piece. But I think that that has real potential to make our users' lives even easier.

Sure, we'll do. I think it's a great product. As I say, when Patrick presented this to I was like, yeah, I know those guys. And then obviously it's great, you know, but it's taking a closer look, it's the what you're doing, the problems you're solving, the time you're saving for your companies and users is is massive. So thanks so much for coming on this afternoon, Cassie. It's been really interesting, and I'm yeah, looking forward to getting this one released. Awesome. Take care, have a one. Bye-bye.

Thanks so much for having me, Alex. Thanks. Bye bye.