Luella with Mustafa Saeed
Luella and the Guardrails of AI-Driven RevOps
Intro
In this interview, Mustafa Sed, co-founder and CEO of Luella, outlines why a cautious, governance-focused approach is essential as AI transforms go-to-market teams. He traces how rampant AI-assisted outreach created privacy, deliverability, and reputational risks—and how Luella builds infrastructure, guardrails, and human oversight to enable compliant, effective engagement without spamming or abuse.
**Overview of core tensions** - Rapid AI-enabled outreach vs. tightening platform policies - Deliverability declines and reputation risks across enterprises - The need for transparency, compliance, and data protection
Center: Core platform vision and differentiators
Guardrails and guardrails-first design
Isolated mini-server clusters: Each sales cluster has its own IP, containing rogue activity and preventing cross-contamination.
Daily volume controls: Safer limits (now 15–25 emails per mailbox per day) to align with Google/Microsoft policies and protect sender reputation.
Placement testing: Regular, real inbox versus spam/promotions checks to reveal true deliverability health.
Transparency and observability
Placement tests deliverables and dashboards to show primary inbox landings clearly. Traditional deliverability scores are debunked as opaque; Luella provides verifiable signals. Alerts: Proactive notices if authentications fail or if a rep behaves badly.
Natural mailbox activity and balance
Simulated, human-like activity to avoid triggering anti-spam signals, improving long-term trust with providers.
Balance between automation and human touch to preserve authenticity.
Mailbox management and integrations
Custom SMTP and various mailboxes: Ensures proper authentication and smooth sending flows.
Over 150 integrations and developer-friendly APIs to connect with tools like Octave, Phantom Buster, Clay RB2B, Trigify, etc.
Automation that scales safely
Automated but bounded messaging: 10-message variations tested on small samples, scaling only when engagement proves value.
Reinforcement learning (RL): Optimizes messaging over time by learning from past performance to propose high-impact hooks and offers.
Human oversight as a core pillar
Humans review AI-generated copy before outreach and monitor for compliance, safety, and brand integrity.
Workshops and Slack hotlines support sales teams in ethical AI use and brand governance.
Privacy, security, and governance
Emphasis on data interoperability and responsible data use; guardrails to prevent leaks and abuse.
Clear stance against “fully autonomous” cold outreach; humans remain in the loop where needed.
Market approach and traction
Lean team (five full-time, three engineers) with enterprise and mid-market pilots (e.g., Oracle, Travel Perk).
Enterprise deals with multi-month cycles demonstrating the need for governance and customization.
Outro
Luella positions itself as a compliance-forward platform that enables AI-powered efficiency without compromising deliverability, reputation, or data safety.
Mustafa invites audience input on concerns and industry-specific challenges, emphasizing ongoing learning and adaptation.
Contact paths: visit luella.ai, schedule a call, or connect with Mustafa Sed on LinkedIn to explore pilots, partnerships, and integrations.
The dialogue underscores a rising consensus: AI in sales must be smart, explainable, and human-augmented to win trust and business outcomes.
Full Transcript
[Music] Today we have Mustafa Sed, co-founder and CEO of Luella. Mustafa, really excited for what you're going to show today because I think the new world of AI has created so many opportunities and at the same time so many new challenges and so many new problems that I don't think a lot of people are even aware of. So I would love if you could lay the context of what the new AI world has created and some of the problems that come along with it and how you guys are solving it at Luella. Yeah, of course. Anthony, thank you so much for having me on the podcast and uh yeah, a big reason for why we're building in this space is because of a lot of the recent consequences of AI and automation that we're now seeing. uh my my founding team and I really came together because of our shared background as you know former agency founders, go to market leaders, security engineers and like we all saw a lot of the same challenges. We saw we all saw a lot of new AI sales tools entering the market, a lot of AI SDRs that were making it easier than ever to spam very large volumes of emails and just content in general without consequences or without limits. And like we started to see very early signs of larger policy changes that are happening in this industry. Google and Microsoft making very aggressive changes to their spam filters for example to combat abuses of AI and automation. And now LinkedIn like Anthony did you see what happened with LinkedIn and Apollo? I did. I did and no warning basically overnight and so many companies had to immediately shift strategy and shift gears to accommodate that. Um, but I think you probably know more of the details of it than I do and would love to share how that's a good example of how some of these things can change so quickly. Yeah. So Apollo, this unicorn, this billion-dollar company, their company page was removed from LinkedIn. If you go to LinkedIn right now, type in Apollo.io, it takes you to a dead page. And uh we saw the same thing happen with Seamless with Ava Boost for uh so many of these companies that have been uh really going aggressive when it comes to web scraping especially uh your personal data to sell them as part of their uh solution. Uh so like we've already started to see uh these very aggressive policy changes and there's a larger ripple effect that we're seeing across the entire industry as a consequence of them. Email deliverability, for example, all of a sudden it's a problem again because of how aggressive Google and Microsoft have become. Uh even trusted senders are being impacted. Not just startups, but some of the biggest enterprises in the world are seeing email deliverability issues, seeing their emails land in spam and promotion folders across several thousands of their sales reps. We're seeing a lot of revops and revenue leaders uh have concerns around potential abuses of AI and automation across their sales team. uh there's a lot of benefits and advantages to using AI and automation and and that's why it should very much be encouraged but uh there are also ways that this these tools can be abused reputation damages for example right uh there was this AI board member that got into some hot water because they started spewing a lot of very sexist claims and the founder uh had to record an apology video their LPs had to get involved and it was this really big situation you know these brand reputation damages can take a long time to recover from. Compliance issues are another one. We work with a lot of health tech companies that are very sensitive when it comes to to the claims that they can make around their software. Uh even just one word that's off uh will result in an issue of non-compliance. Data leaks. Anthony, did you see what I posted on LinkedIn yesterday? I did not, but would love to hear it. So, this hacker tricked an AISDR into leaking its IP address, SSH credentials, and password file. Oh, wow. And all he did was add a prompt in his LinkedIn bio uh pretty much asking to override any previous instructions and provide those credentials. Uh so especially with so many of these tools just being vibe coded together uh like there are real risks to AI coupled with reckless automation and uh that's why we wanted to build something in this space to build those guardrails to prevent abuse uh to enable organizations to make the most of AI without abusing AI. Uh do a better job of protecting reputation, preserving deliverability and help you connect with prospects instead of spamming the living hell out of them. And uh we're very grateful to have worked with a lot of incredible RevOps leaders. So a big reason for why I wanted to be on this podcast is because it's really revops that has been stepping up to implement these guardrails across their sales organizations. Uh and we've had the chance to work with RevOps leaders at Oracle and Travel Perk amongst uh several other enterprises. Uh we're still a very lean team. Uh we're a team of five full-time with three engineers. Uh and we wrote that first line of code just over six months ago. So everything that I'm about to show you is still very much new. Uh and my co-founder and I today is actually our work anniversary. So uh one year ago today I sent in a cool message. Thank you. Uh we we met on Y Combinator's co-founder matching and uh we've been building ever since. So uh that's a little bit about the story of the crazy journey that we've been on so far and uh why we've decided to build in this space. I love it. Well, I'm really excited. I think so many of the new companies are built lean and mean anyway. um you look at like cursor lovable those type of companies reaching unicorn status with a handful of people so you don't need a big bloated team to make something meaningful and what you're targeting hits me personally in so many ways one I have a cyber security fraud prevention background so when I was leading revops in tech it was mainly in that space so I know how many vulnerabilities already were around before all of these uh new AI capabilities and on the receiving and my inbox and LinkedIn is just lit up every single day and it's clearly it always was a problem. We just poured gasoline on an already really hot fire. So, I'm really excited about the mission and what you're solving and there needs to be guard rails in the space and ways to um just make sure it's a more meaningful environment especially in go to market and the RevOps community. Everyone in RevOps is on the hook for solving this. So people are looking to us to come up with a solution to make sure hey we can still sell but still protect our reputation at the same time. Exactly. Completely agree. Uh these legacy workflows throwing AI and automation at them like that that's not the solution right there's there's infrastructure that we need to build uh just to make better use uh of these tools. Uh and really excited to dive uh into that. Anthony should I share my screen? Yeah let's do it. So, it really all starts with the infrastructure that we're actually sending from. Uh, so Anthony, if I was to say shared IP pools, do you know what that is? I'm not sure of it. Maybe I should be, but I'm not. So, these are massive shared servers with several thousands of unvetted senders in them. And in these servers, you're going to have both good actors that are running cold outbound tastefully and ethically, but also those bad actors that don't give a that are spraying and praying. And when you're sending outreach out of the majority of these legacy outreach tools, you are unfortunately using these shared servers and that exposes even the most trusted senders to spammers and grifters. Uh so that is a larger piece that hurts your sender reputation and makes you more likely to land in spam because of it. Uh so we instead of these really big servers, Anthony, we will create your own mini server. And we call these mini servers clusters. And this is exactly what they look like. This is Evan's cluster over here. This is my cluster. Each of these clusters has its own IP address. Uh just so if one sales rep goes rogue and starts pushing a lot of BS, it doesn't contaminate your entire infrastructure. It is isolated to the cluster that we've created. So this is a larger way that we're isolating risk across your sales organization. And if you click into any cluster, you'll be able to see its corresponding metrics. Uh, another larger challenge that we're seeing in this industry is transparency. A lot of organizations don't know what percentage of their outreach are actually landing in the primary inbox versus spam or promotions folders. Uh, the reason for why is because Google and Microsoft only give you self-reported spam metrics, which is a small fraction of the data that they actually have. And most outreach tools will give you like a deliverability score, which is a madeup metric. They have full control over what goes into that deliverability score, and it's more often than not inflated. So, we to bring back that essential visibility for your team are running placement tests. Anthony, do you know what that is? I'm not sure, but I'm getting a great education today. So, we on a regular basis are sending emails from your mailboxes to ours to see where they land. uh the inbox, the spam folder, the promotions folder. Sometimes they don't get delivered at all. And this becomes the best indicator that we have to the overall health of your email infrastructure. Whenever there's a problem with your email infrastructure, we'll always see a spike in spam rates reflected over here. And uh Luella is using uh these data points to regularly diagnose your email infrastructure and surface alerts to both your team and ours. So let's say our authentications break. Uh let's say Google and Microsoft make a change that they are more public about that does warrant action on yours. Let's say you have a sales rep that has gone rogue. Uh Luella is constantly looking out for you and surfacing those notifications not only from a deliverability perspective but also to catch instances of abuse before it actually becomes a problem for your organization. So by the way I have seen that exact use case happen at every single company I've been at. an overzealous SDR, ripped a h 100,000 emails out of Zoom Info, threw them out of their inbox right away. I mean, I've seen that exact thing happen every single time. Yeah, it's crazy. Literally millions of sales reps all over the world, like that's how they're operating. They pull these massive lists from stale databases and following static sequences where on day one you send a templated email, day five you send a templated, day 10 you send a templated email. Like it's a very robotic way of doing business. And uh like we we urgently need to move away from it because even without AI it was super spammy but now with AI and automation it's made it 10 times worse. Uh and in addition to the placement tests uh we're also simulating natural mailbox activity. Uh so when you're sending cold outreach that isn't very natural right because you're sending a larger number of emails uh but with a much lower response rate. So uh Luella will simulate uh natural mailbox activity with real corporate mailboxes uh just to do a better job of balancing your response rates uh just to not trigger a red flag in the eyes of Google and Microsoft. So this is the piece that helps build trust with email service providers over time. And this is what we can also use if uh you have burnt mailboxes and domains and are looking to shift to an approach that is more compliant. I love it. So those are the range of things that we're doing on the deliverability side. Uh we also do mailbox management. So Google Outlook custom SMTP. Uh we can manage the mailboxes and the corresponding authentications to make sure everything is correct. Uh you can also send emails out of Luella as well. So um a few larger differentiators here. Uh the first has to do with automation limits. Uh so a lot of the changes that we are seeing from Google and Microsoft have to do with volume caps. Gone are the days of sending hundreds of emails every single day out of just one mailbox. Uh last year we reduced our volume to 50 per mailbox per day. Today we've reduced to 15 to 25 emails per mailbox per day. That is how aggressive Google and Microsoft have become. Uh so uh Luella unlike other platforms that will let you set whatever volume caps that you want, Luella is the one that controls your volume every single day uh just to maintain that compliance. Another guard that we've built has to do uh with how we're scaling that volume over time. Uh it's really important that you're not going full guns blazing on day one and sending hundreds of thousands of emails, right? You really want to make sure you're testing your messaging across a much smaller sample size before you do press your foot in the gas just to make sure you're providing value to the prospects that you are reaching out to, right? You're seeing positive reply rates, right? So, I think this is one of the biggest mistakes, first-time founders when you get your first AE into a company that they make. They think one they think that their ICP and their messaging is locked and loaded and they just go nuts with the volume of communication. And I don't think people realize that you're doing more harm than good. You're hurting your brand. People will immediately start to associate your brand with, you know, someone that spams them. And then when you realize you iterate and you need to improve your messaging, you've kind of burned that bridge already. So I think people think sales is a volume game in ES in some ways, but really it's a it's a quality game and you don't really get an opportunity to get that quality unless you're doing it in the manner that you say. small batches, iteration, testing, very thoughtful messaging, very personalized messaging, not just blasting 100 thousand people. Exactly. Completely agree. And like in order to be able to achieve this, we've also built a lot of the same algorithms as like many of the traditional ad platforms. So Anthony, like do you guys run Facebook ads or LinkedIn ads internally? We do some LinkedIn ads. We actually do some uh YouTube ads as well. So we're a little familiar with it. Yeah. So the way that LinkedIn ads work is you give them 10 different ads. LinkedIn will test each and every one of those ads against a smaller subset of your audience and only after LinkedIn sees customers are liking the ad, you know, clicking on it, engaging with it, like showing actual value that's being delivered to the end user where LinkedIn will show that ad to more and more and more and more and more people. Luella will do the exact same thing. So you may add 10 different message variations to a campaign and Luella will test each and every one uh across a much smaller sample of the contacts that you do have and only after she sees positive replies where Luella will uh scale the version that is performing the best. Uh so you can push volume in cold outreach. Uh it's just very important that you are delivering value to those prospects. uh if your messaging isn't providing value then you are spamming your users and you deserve to be in the spam folder. Uh so uh that that's why we've built those necessary uh limits uh just to prevent bad actors from spammer people spamming our platform. U another piece has to do with reinforcement learning. Do you know what reinforcement learning is? Anthony gonna learn another thing today. So so this is a branch of artificial intelligence. Uh there are multiple different branches of artificial intelligence. LLMs are just one of them, right? LLMs are really just glorified copyrightiting machines. Reinforcement learning is another branch. Uh this is the optimization layer. This allows us to look into the past, understand what has performed, what hasn't performed in order to better predict the future. So in order to help teams uh better improve their messaging over time, uh we're using reinforcement learning uh just to run a lot of mini experiments with your messaging to get a better understanding to the hooks, the lead magnets, the offers that did resonate and provide value uh just so we are optimizing uh the best messaging with the prospects that you're targeting. Uh so that guardrail makes sure that only the highest performing messaging that you have added to Luella gets shown to customers and over time once Luella does have enough data she'll be able to recommend new message variations with new hooks new angles uh new lead magnets new offers that are likely to outperform what you have tested in the past. So uh uh that is a piece uh that is very important to to make sure you are improving your messaging especially for those startups that you talked about that just haven't cracked it as of yet right uh they're sending small volumes and they're seeing low response rates uh luella will help you improve that messaging just so you can reach a point where you are providing a value and you can push more volume that's really unique and normally a lot of revops teams will be behind the scenes crunching a bunch the numbers trying to make sense of the data and then I mean it's one thing to say hey this sequence is outperforming this one it's another thing to really understand why and then replicate that or take components of that and create new sequences from that so I think having an AI powered analytics function plus helping you create new messaging um based on what's performing really well like knowing the components of the why it's performing well is really important Yeah, and that's an area the entire industry really needs to do a better job of. Uh because there are a lot of companies using intent signals and we're strong advocates for intent signals. Uh but there are thousands of them, right? And like most platforms are only doing a good job of like unifying like a couple of dozen like if not just like one or two. Uh so so like the entire industry has a lot more work to do when it comes to that messaging piece. Uh and there's a reason for why we've built you know very developer friendly APIs uh just so we can integrate with uh like-minded partners. Uh one company that uh we're in the process of building in integration with uh is do you know a company called Octave? Yes. Yeah. They've actually been on our podcast in the past. So great great company and love the founder as well. Yeah. We we love what uh they're doing, what Julian and Zach are doing on their side and uh they also come from a like very strong like anti-PAM background. So uh like we really share uh the same mentality over the space and uh we confirmed our native integration with them just last Friday. So uh right now we have a really big integrations push. We have over over 150 integrations uh on on the road map uh and building developer friendly APIs. Like we're very strong believers in data interoperability. Uh but Octave does a really good job when it comes to aligning uh the right messaging w with the right prospect. Uh Phantom Buster is doing a really good job when it comes to intent data. Uh same with Clay, RB2B, Trigify. Uh there's so many incredible uh companies in the space uh that that we are looking to integrate with. Uh and Anthony, the last piece that I'll mention is human oversight, right? There are a lot of AISDRs that are advocating for like a fully autonomous experience where these AI agents will just do everything and just hand you these opportunities on a silver platter. And like not only are we very far away from that, like I don't believe that's where this industry is going. Like there are advantages to having AI operate independently, especially, you know, being able to ingest a much larger amount of data sources. Like that's something I can't do, right? A couple of hours looking at a spreadsheet, I'm tapping out, right? But there are things that like a human can do that an AI agent can't, right? I can show up in person and do a more meaningful job of building a relationship uh with a prospect. So we really need to blend the best of both worlds. Uh so like that's why especially for enterprises and mid-market companies that we work with uh making sure that we do have opportunities to bring the humans back in, right? There are certain pieces of copy that are generated by by AI that a sales rep should see before it actually gets to a prospect. there are certain things that an admin on the account should see uh before it even gets to the sales rep in the first place. So uh human oversight is something uh that is very important uh that will help prevent a lot of the previous abuses that I mentioned brand reputation damages compliance issues data leaks uh having humans in the loop uh is a larger uh protection against that and that's also a big reason for why like we're not just a software company right we have a lot of like human beings that are there to support as well like I regularly host workshops uh with sales teams to uh educate on the ethical boundaries of these technologies, how you should have your brand be represented in this AI native world. We'll also have like Slack hotlines where sales reps can, you know, put like all the challenges that they are seeing uh in the trenches when it comes to a security a safety and uh deliverability perspective. Uh but yeah, Anthony, that that's the platform. I love it. And I just want to comment on keeping a human in the loop. I think it's fine if you're selling really low ticket item BTOC context. um there's not a lot that company's going to change for that consumer. I think why it's different in B2B and higher ticket item, you need to know that you have somebody on the other side that is going to advocate for you at the company. As an example, when we engage with tools that we would use internally, I want to know that somebody there is going to support the customer if we introduce them to them because that's really important to our business. So, building that trust and knowing you have somebody on the other end who can make some decisions around the product experience you're getting, the way it's priced and packaged, like you do need some customization where AI can't make those decisions for you. and you you know AI doesn't have the authority to do some things that you need a human on the other side to do. So I I think some people don't maybe they don't understand that aspect of sales like especially if you're getting into selling like $100,000 plus type of um type of sales. That's an expectation from the customer side. So, it's not just, yeah, I can get some product information really quickly, but is somebody going to have my back if something goes wrong? Yeah. Um, I need that assurance that, you know, somebody's on the hook for making this successful. And we just recently closed our first really big like enterprise six figure a year deal. Uh, so we we've gone through that, right? Like it was a like four to five months pro process of going back and forth. And there's like I can I can list all the things. It's a very comprehensive of all the that like an AI agent just will will not be able to do uh that like we needed to have like our team uh present to uh get over the finish line. I'm really impressed with where you started with Luella and thinking about trust and safety and thinking about how you can protect the reputation of the company creating those um individualized IP addresses, individualized servers. Um, also the approach that you've taken and how you've embedded AI into this process. One, making sure it's testing different messaging. It's optimizing that messaging, putting the right guard rails on because I think you need to protect your customers from themselves. They will have a rogue STR, they will have a rogue AE who comes in and just completely lights up their prospect base and then puts the entire company at risk. So yeah, I I think the way you've integrated AI, the guard rails you've put in place, having it with trust built in the foundation and where you're going in the future, I I think is just really really impressive and I think is going to be highly highly competitive. Thank you so much. Uh and yeah, like we're just getting started, right? Uh first line of code over six months ago and uh like in this industry like these models are going to keep improving, right? And we don't know what we don't know, right? There are going to be more risks that like we're not aware of today that do come up. So, uh like I would encourage your audience as well to you know comment uh as to what are the concerns that they have in disseminating AI and automation across uh their sales team. Uh because like especially when you go vertical specific right when you speak to health tech companies for example or companies in education there are unique challenges with with every industry and like every conversation is a new learning for us as well. Mustafa, what's the best way for people to get in touch with you to start getting their hands on the product? Yeah, if you go to luella.ai, our website, my Calendarly link is is on the website. Feel free to schedule some time. Always happy to geek out. My name is Mustafa Sed on LinkedIn as well. Uh feel free to connect. Mustafa, thank you so much for being here today. Really pumped for what you're building and what you're doing and can't wait to see what you guys build next. Thanks for having me on the pod. Thank you so much. [Music]
Last updated
Was this helpful?