96. AI Native Ops, How Brex Rebuilt Operations
In this episode we discuss: AI Native Ops, How Brex Rebuilt Operations. We are joined by Camilla Matias, COO at Brex.
Love The Operations Room? Please support us by rating and reviewing it here.
We chat about the following:
- What actually breaks first when a company scales: people, process, or structure—and how do you know?
- How do you build operational rigour without killing speed and culture in a fast-growing business?
- What does “good onboarding” really look like beyond just paperwork and checklists?
- When tools or systems fail—how do you know if it’s the software or your implementation?
- What’s the difference between building processes for today vs. building systems that scale for the next 12–24 months?
References
- https://www.linkedin.com/in/camilla-matias-06345a105/
Biography
Camilla Matias Morais is the Chief Operating Officer at Brex, where she leads global operations and drives the company’s execution and scale.
She joined Brex in 2018 as Head of Finance and has since held multiple leadership roles, including VP of Finance and SVP of Global Operations, before becoming COO in 2024. During this time, she has played a key role in building the systems and structures that support Brex’s rapid growth.
Before Brex, Camilla worked across finance, investment, and strategy roles at organisations including The Kraft Heinz Company and Victoria Capital Partners.
She holds a degree in Mechanical Engineering from the Instituto Tecnológico de Aeronáutica (ITA) and is known for bringing clarity, structure, and operational rigour to high-growth companies.
To learn more about Beth and Brandon or to find out about sponsorship opportunities click here.
Summary
00:00 – Intro, host banter, and setup of the episode tone.
02:30 – Introduction of Camilla + context around her role and operational scope.
05:30 – Early discussion on scaling challenges and what starts to break as companies grow.
07:20 – Deep dive into onboarding systems → moving from informal to structured 0–6 month processes.
09:30 – Discussion on tools (HiBob → Rippling) and implementation challenges.
12:30 – Balancing structure vs flexibility in operations.
16:00 – Performance reviews, documentation, and accountability systems.
20:00 – Operational maturity: what “good” looks like at different stages of company growth.
24:00 – People vs process tension.
29:00 – Reflections on mistakes, lessons learned, and what they’d do differently.
34:00 – Closing thoughts, key takeaways, and wrap-up
This podcast uses the following third-party services for analysis:
Podcorn - https://podcorn.com/privacy
Transcript
Hello and welcome to
Speaker:another episode of The Operations
Speaker:Room, a podcast for COOs.
Speaker:I am Brandon Mensinga joined by my
Speaker:lovely co-host Bethany Ayers.
Speaker:How are things going, Bethany?
Speaker:How do you like the new glasses?
Speaker:The new glasses are fabulous.
Speaker:I'm mostly convinced.
Speaker:I feel like there may be a bit too
Speaker:low.
Speaker:There's a bit too much eyebrow.
Speaker:This is one where we're gonna have
Speaker:to do a bit of a video or have like
Speaker:a picture for everybody to see what
Speaker:I'm talking about.
Speaker:But on the whole I like them and
Speaker:they are, I think my
Speaker:optometrist was calling
Speaker:them multifocal rather than
Speaker:varifocal, so maybe that's
Speaker:the new name. He also calls them
Speaker:spectacles so he might just be
Speaker:inventing his own.
Speaker:Basically, there's the top is
Speaker:looking far away.
Speaker:The middle is
Speaker:for me plus one, and
Speaker:then reading is plus one and a half.
Speaker:Okay, so I think I have two then.
Speaker:I have like one for my classic I
Speaker:Can't See Far Away, and then I've
Speaker:just got the pure reading one.
Speaker:So I was expecting because of the
Speaker:way everybody's talked about it that
Speaker:I was going to put these on and my
Speaker:whole world was going to be horrible
Speaker:and I was going to fall over
Speaker:for days, but it
Speaker:hasn't been bad at all.
Speaker:Like I don't notice the, I think
Speaker:maybe because it's only plus one and
Speaker:one and a half, I don't notice big
Speaker:changes sitting down,
Speaker:everything was fine when I tried
Speaker:them on, then she had me stand up
Speaker:and I was like, Whoa, somehow
Speaker:I'm walking on a bouncy castle.
Speaker:But very quickly I got used to
Speaker:the bouncy castle and now like
Speaker:the world is starting to be a bit
Speaker:flatter.
Speaker:I mean, you get used to it super
Speaker:fast. I mean I was told the same
Speaker:thing. I mean disoriented for a
Speaker:while because it's weird to like
Speaker:your eyes have to focus in the right
Speaker:spot. I think within, I don't know,
Speaker:two hours I was probably fine and I
Speaker:haven't looked back since, so it's
Speaker:not that bad.
Speaker:Yeah, and it's so nice to be able to
Speaker:pick up my phone and read and be
Speaker:able to look out and read, like it's
Speaker:just so nice.
Speaker:All right, so any special events
Speaker:this week?
Speaker:I went to a Local
Speaker:Globe event last night.
Speaker:I think it's called Open Court.
Speaker:I always find it difficult when
Speaker:people like have their main brand
Speaker:and then they have other brands and
Speaker:then you have to remember it, all
Speaker:the different rules.
Speaker:But I think I was called Open court.
Speaker:I think its happening once a month.
Speaker:It happened a bit last year and this
Speaker:was the first one for this year.
Speaker:And it's just bringing the community
Speaker:together and having different
Speaker:speakers.
Speaker:And I forget how good Local Globe
Speaker:are. I'm not just saying that
Speaker:because they're an investor.
Speaker:You know, their offices are
Speaker:really, we've been to them,
Speaker:like for a COO event.
Speaker:They're really open and generous
Speaker:about using their offices.
Speaker:They do a lot of community work.
Speaker:This event, the age range was
Speaker:probably 18
Speaker:to 70.
Speaker:And everybody who was just
Speaker:interested in AI,
Speaker:the speakers had the
Speaker:business leader, I can't remember,
Speaker:he has a weird title, but the
Speaker:and then the head of
Speaker:AI and ML for
Speaker:Spotify.
Speaker:It was interesting, but also I was
Speaker:like, wow, despite
Speaker:being in this space, but being on
Speaker:the business side, so little.
Speaker:He was such an impressive person.
Speaker:I looked him up afterwards just to
Speaker:see his name's Martin Gould.
Speaker:It's one of those people where
Speaker:he can make the complex easy
Speaker:to understand and compelling, and
Speaker:feels very much like a good
Speaker:professor, And it's also just.
Speaker:Clearly unbelievably smart
Speaker:when it comes to maths, but also
Speaker:just smart and
Speaker:well-rounded and personable
Speaker:and an amazing communicator, like
Speaker:just one of these people that God
Speaker:has touched more
Speaker:than the rest of us.
Speaker:Oh wow, sounds like a very
Speaker:impressive individual.
Speaker:So impressive.
Speaker:And so he did his, I think,
Speaker:undergrad and master's at Cambridge
Speaker:and then did his PhD at
Speaker:Oxford in maths and
Speaker:then also has a great love of music
Speaker:and so combines those two passions
Speaker:to create amazing personalization
Speaker:to make sure we all access the music
Speaker:we want to access.
Speaker:Sounds like an amazing bio.
Speaker:Yeah. And it was just a really, it
Speaker:was so nice to have all these
Speaker:different people come together.
Speaker:The questions were from
Speaker:what's your business strategy
Speaker:to the granola, like what's
Speaker:next, how are you going to keep
Speaker:growing to
Speaker:questions I can't even paraphrase
Speaker:on like the level of maths
Speaker:that people in the audience were
Speaker:asking the Spotify guy.
Speaker:It's great to see how
Speaker:vibrant the London startup community
Speaker:is and how diverse it
Speaker:is. It just got a real buzz from it.
Speaker:Yeah, for sure.
Speaker:Local Glob is one of the, if not the
Speaker:premier seed company, has to be
Speaker:close to the top three, I would say,
Speaker:in the UK.
Speaker:And it's been around for such a long
Speaker:time and they've invested in so many
Speaker:great companies as well.
Speaker:That space that we've been to
Speaker:before, it's a great space.
Speaker:And to your point, it's like
Speaker:relatively speaking open to the
Speaker:community, open to special events.
Speaker:If you're one of these seed-based
Speaker:companies, you can work out of there
Speaker:as much as you want type of thing.
Speaker:So it's fabulous environment.
Speaker:Interesting bits or a connection
Speaker:here is that I went to the
Speaker:60 minute mentor live
Speaker:podcast with James Mitra and
Speaker:he had two guests talking about
Speaker:talent acquisition and one of those
Speaker:individuals His name was Charles and
Speaker:he's now I think like the chief
Speaker:talent officer of lovable and
Speaker:He was plucked directly out of
Speaker:local globe for that role so
Speaker:apparently this Charles fellow
Speaker:similar crazy background like he
Speaker:was in one of Elon
Speaker:Musk's, the Neuralink company
Speaker:as the talent lead for
Speaker:Neuralync working directly with Elon
Speaker:Musk. And then he worked for some
Speaker:other company in a similar kind
Speaker:of a crazy awesome capacity.
Speaker:And now he's had lovable and
Speaker:apparently lovable kind of
Speaker:crisscrossed like all
Speaker:the VCs in particular
Speaker:for some reason looking for their
Speaker:talent lead and
Speaker:plucked out this as fellow to lead
Speaker:their talent.
Speaker:So the probation period,
Speaker:they tell candidates out of the
Speaker:gates, there is like a more than
Speaker:non-zero chance that you will not be
Speaker:here at the end of the three months.
Speaker:So the probation is deadly
Speaker:serious. And if you're an incoming
Speaker:hire, you need to recognize that
Speaker:you might be out the door in three
Speaker:months effectively.
Speaker:Interestingly as a side note.
Speaker:The regulations in the UK are
Speaker:changing next year, whereby this
Speaker:kind of two-year cutoff point that
Speaker:classically has been there where if
Speaker:you're an employee less than two
Speaker:years, you don't have a lot of
Speaker:rights and the company can terminate
Speaker:you, the company can terminates you
Speaker:very easily.
Speaker:Once your past two years it is much
Speaker:more difficult for companies to
Speaker:terminate. You have to go through
Speaker:performance improvement plans and
Speaker:special cycles to exit somebody from
Speaker:a company. So that two-year marker.
Speaker:Next april is becoming six months
Speaker:which is almost like probation away
Speaker:so the six month marker i think all
Speaker:companies in the u.k.
Speaker:I have to change their approach in
Speaker:terms of how they do that for six
Speaker:months very specifically to ensure
Speaker:that there's much more rigor around
Speaker:somebody's performance in a way that
Speaker:previously i don't think we really
Speaker:had to.
Speaker:So this is more than i think the seo
Speaker:is that out there this is coming and
Speaker:that your marker is not gonna be six
Speaker:months come next april.
Speaker:And I think it's the orientation of
Speaker:how you think about performance in
Speaker:that first six months to ensure that
Speaker:the person is
Speaker:high value needs to stay in the
Speaker:business. You're much more clear on
Speaker:that. What do you make of that?
Speaker:We've already done it.
Speaker:Oh, you've done it, you're always
Speaker:ahead of me.
Speaker:Well, I mean, we've done it
Speaker:internally.
Speaker:Our finance and ops and people
Speaker:person has
Speaker:built in the structure, set up the
Speaker:alerts, put in the
Speaker:performance reviews and
Speaker:documentation.
Speaker:So we now have a
Speaker:much more rigorous zero
Speaker:to six month process than we did
Speaker:previously. So when it's
Speaker:rolled out, we're ready.
Speaker:Oh man, you're going to have to give
Speaker:me some crib notes here on what to
Speaker:do.
Speaker:Yeah, and we're also, so
Speaker:we've done it all vibe
Speaker:coded for the moment,
Speaker:but what we're considering doing is
Speaker:we moved from High Bob
Speaker:to Rippling,
Speaker:but have not had a great experience
Speaker:with Rippaling, but I don't
Speaker:necessarily think that's a
Speaker:reflection of Rippiling, and it
Speaker:might be more of a reflection of how
Speaker:we rolled it out and what
Speaker:we are paying for.
Speaker:But rather than looking for new
Speaker:HRIS, we have decided that
Speaker:we're just going to Vibecode our own
Speaker:or code our own.
Speaker:And so we have different
Speaker:elements of it right now.
Speaker:And then the next step is to turn it
Speaker:into the app that we all use.
Speaker:So we have onboarding and
Speaker:offboarding has been done and
Speaker:their probation period and
Speaker:performance management frameworks in
Speaker:general have been done.
Speaker:And now the last piece is to put it
Speaker:together and build the holiday
Speaker:and sick leave tracking.
Speaker:And then we have our own HRIS.
Speaker:Yeah, that's awesome.
Speaker:So I'm going from rippling to
Speaker:creating your own situation.
Speaker:That sounds fabulous.
Speaker:But just because we're not very
Speaker:complicated, we're 20 people, and we
Speaker:don't need to pay whatever it is
Speaker:for, we use half
Speaker:of it.
Speaker:And it doesn't work well, and it's a
Speaker:bit ugly, and it' annoying,
Speaker:so we might as well just do it
Speaker:ourselves.
Speaker:Yeah, that HRAS space
Speaker:is so weird because you're right, at
Speaker:the core center of it, there's
Speaker:really not that much that needs to
Speaker:actually be done for core HRAS,
Speaker:which is the reason why the high
Speaker:bobs and the ripplings have like
Speaker:12,000 add-on modules that you can
Speaker:buy because they realize the core is
Speaker:so miniature in size.
Speaker:So to your point, if there's any
Speaker:kind of like SaaS software that's
Speaker:pretty doable to get rid
Speaker:of versus like a HubSpot which is
Speaker:much more complicated, the core
Speaker:HRAs like front and center,
Speaker:I would say to kind of AI
Speaker:agent your way out of it.
Speaker:Yeah, and also like we've done it
Speaker:bit by bit because we've just
Speaker:automated different processes and
Speaker:it's like, oh, we've automated this,
Speaker:we automated that, we automate,
Speaker:okay, like, why don't we just
Speaker:link it all together and save
Speaker:ourselves some money.
Speaker:We've done the same for a partner
Speaker:portal because we were paying for a
Speaker:partner portal that wasn't
Speaker:expensive, it was like three grand a
Speaker:year, but it had a lot of
Speaker:features and we were using it as
Speaker:basically a form and a database.
Speaker:I was just like, no, there is a
Speaker:much better solution to this.
Speaker:That's such a SaaS software thing,
Speaker:isn't it? Every SaaS software, you
Speaker:kind of use the core essence or the
Speaker:functionality, but all the
Speaker:additional stuff, which there's
Speaker:always a ton of it, you know,
Speaker:another 90% of features you never
Speaker:touch. And then you're like, okay,
Speaker:do I really, you know again,
Speaker:same thing, like can I AI agent my
Speaker:way out of this just for the core,
Speaker:essence of what we're trying to do
Speaker:or because I don't need this other
Speaker:90%.
Speaker:Yeah, I mean, I think when you're a
Speaker:bigger company and you need it all,
Speaker:fair enough, but at our size.
Speaker:And it's all just knowing to pay for
Speaker:somebody to like, I mean, the
Speaker:partner portal was like super basic.
Speaker:Why are we paying for this?
Speaker:All right, so we've got a great
Speaker:topic today, which is AI Native Ops,
Speaker:how Brex Rebuild Operations.
Speaker:And we have a great guest for this,
Speaker:which was Camilla Matias.
Speaker:She is the COO at Brex.
Speaker:So the first thing I wanted to ask
Speaker:you was, Camilla had talked about,
Speaker:if you're going to follow a
Speaker:procedure, you don't need a human
Speaker:anymore.
Speaker:Whatever is kind of like an L1 type
Speaker:role just doesn't exist going
Speaker:forward.
Speaker:What's your take on Camilla's?
Speaker:Well, I mean, I think she was
Speaker:talking about her business in
Speaker:particular, but it ties
Speaker:in a lot to what we've all been
Speaker:talking about of what's the role of
Speaker:an individual contributor.
Speaker:There are no more individual
Speaker:contributors, everybody's running
Speaker:agents.
Speaker:I think the bigger question is
Speaker:how do people gain experience
Speaker:if there aren't any
Speaker:L1 jobs?
Speaker:And so I don't think that the entire
Speaker:company is going to age out and
Speaker:die or retire if we don't hire
Speaker:younger people. So I think we need
Speaker:to switch and start to think about.
Speaker:We need young people and
Speaker:increasingly the young people who we
Speaker:hire are going to be AI native in
Speaker:the way that the last generation
Speaker:were digital natives.
Speaker:And so what are the skills that
Speaker:they need and how do you pair them
Speaker:with the experts in the business?
Speaker:Because what they're gonna have to
Speaker:learn is they'll come
Speaker:with the ability to do a lot more
Speaker:with AI than we
Speaker:have, but we have the business
Speaker:experience.
Speaker:And so how do teach
Speaker:them the business experienced?
Speaker:With their AIs, maybe a buddy system
Speaker:might work, but I think we need to
Speaker:hire in more young people and if we
Speaker:don't, we're going to be in trouble.
Speaker:Yeah, yeah, so I think you're
Speaker:exactly right. It is absolutely
Speaker:fascinating because we talked to
Speaker:True Search to do a tire
Speaker:for a VP of product.
Speaker:What True had told us was that look,
Speaker:if you're looking for a VPA product
Speaker:right now, you're gonna get one of
Speaker:two people. One is gonna be the kind
Speaker:of like older, broad-based
Speaker:experience that's done your run
Speaker:before going from 20 million ARR
Speaker:to 100 million ARRR, but they're not
Speaker:gonna be AI native first, which is
Speaker:what you're look for.
Speaker:And if you want to find that person
Speaker:that's a VP product, they're going
Speaker:to be much younger, much less
Speaker:experienced, much more of an
Speaker:ambitious step up role to become
Speaker:like a senior leader in your
Speaker:business. But they're gonna come
Speaker:with that skill set that you're
Speaker:looking for out of the box in terms
Speaker:of being AI native first.
Speaker:And this is not an L1 role that I'm
Speaker:talking about, obviously, but but I
Speaker:think your comment applies to this
Speaker:as well, which is, I think we're
Speaker:going to need AI native,
Speaker:first young people coming in that
Speaker:are like thinking purely through
Speaker:as opposed to us where we're trying
Speaker:to like backport our thinking to
Speaker:figure this stuff out basically and
Speaker:I think some combination of the two
Speaker:somehow this is what the outcome
Speaker:is going to look like.
Speaker:Yeah, I think for Camilla's
Speaker:business it's easier because her L1
Speaker:or customer service, I'm thinking
Speaker:I'd love to listen to how
Speaker:law firms are going to address
Speaker:this because presumably we're still
Speaker:going to need lawyers in the future.
Speaker:How do you train them?
Speaker:So when she was transforming Brex,
Speaker:and she did a fabulous job of this,
Speaker:and you'll hear this a little bit
Speaker:later on in the interview that we do
Speaker:with her, she was saying that the
Speaker:real unlock was intensity
Speaker:and operating like a seed-based
Speaker:company. So she was say Brex is
Speaker:quite a large organization, but she
Speaker:had to think from like a C point of
Speaker:view in the way of like saying
Speaker:to herself and her team, you know,
Speaker:I'm the CEO of the company, I have
Speaker:responsibility for this entire
Speaker:organization, but you know what?
Speaker:I'm not gonna do that.
Speaker:What I'm gonna do is like cancel all
Speaker:my meetings.
Speaker:Pull together a small team of
Speaker:individuals, including myself, and
Speaker:we're going to go forward with this
Speaker:mantra of AI, the AI native
Speaker:and build it out from the ground
Speaker:floor.
Speaker:So effectively in a weird way, she
Speaker:was almost like seconding herself
Speaker:as a CEO to lead the charge at the
Speaker:outset. And the outcome that she's
Speaker:delivered that you'll hear on the
Speaker:interview has been pretty
Speaker:transformational for Brex and I
Speaker:think all of us thinking through
Speaker:CEO's and CEO's in this case having
Speaker:the courage to do something like
Speaker:that sounds pretty radical and she
Speaker:had the backing of a CEO so that
Speaker:makes a difference i think but in
Speaker:any event what do you make of what
Speaker:she did.
Speaker:And her board.
Speaker:I just think that her
Speaker:interview is one of the
Speaker:most important and most interesting
Speaker:and valuable that we've ever done.
Speaker:It's almost like, should we talk
Speaker:about it or should we just tell
Speaker:everybody, listen to it?
Speaker:If you want to understand how
Speaker:to cut your cost to serve by 50%,
Speaker:if you want understand how truly
Speaker:get efficiency from
Speaker:AI, spend the
Speaker:I don't know how long it's going to
Speaker:be, 45 minutes?
Speaker:Listening to Camilla.
Speaker:There you go. So I think we've set
Speaker:Camilla up. Let's park it here and
Speaker:let's get on to the conversation.
Speaker:I remember like the beginning of
Speaker:like last year, Pedro comes
Speaker:to me, and we were having like
Speaker:our, Pedro is our CEO and
Speaker:founder, we were having like our
Speaker:feedback session, he was asking,
Speaker:Amila, what would you do if you're
Speaker:starting Brex today from scratch
Speaker:with like all the technology that we
Speaker:have in place?
Speaker:And I remember, like the theme was
Speaker:like, oh, we're not adopting AI
Speaker:enough, or quick enough.
Speaker:And I look at him, he's like, Oh,
Speaker:I would build everything.
Speaker:Differently from, I mean, if we had
Speaker:to build now, if like everything
Speaker:that is available, the concepts
Speaker:would be different.
Speaker:It's just hard to change.
Speaker:And he was like, okay, so let's
Speaker:rebuild from scratch.
Speaker:And that's the thoughts that I had.
Speaker:Like, okay, what if I would
Speaker:actually rejoin from scratch?
Speaker:What if I was hire my first
Speaker:operator, my first function in
Speaker:ops with all the technology?
Speaker:I remember in the past, everything
Speaker:was about, Okay, how do I document
Speaker:things? How do I make sure that I
Speaker:have the right metrics in place, the
Speaker:right SLAs?
Speaker:How do make sure I have the SOPs
Speaker:that are codified that many people
Speaker:can do that, and they're not just
Speaker:making the decision based on
Speaker:whatever they have in their heads.
Speaker:When I started thinking about this,
Speaker:I thought, what if I can codify
Speaker:all of that with AI,
Speaker:and then AI is just
Speaker:part of the team, because it doesn't
Speaker:need to be a technology itself, but
Speaker:it should be part of a team.
Speaker:And the managers, the leaders, they
Speaker:are not like leading
Speaker:humans and training your pupil, but
Speaker:they are also training the
Speaker:technology in their favor.
Speaker:So it's not that the business
Speaker:outcome of the role has changed.
Speaker:The leaders, everyone in the team is
Speaker:still have like the same business
Speaker:goals.
Speaker:It's just how you achieve the roles
Speaker:becomes different.
Speaker:Could you explain what you mean
Speaker:by codifying
Speaker:everything from scratch?
Speaker:So did you just sit and talk to AI
Speaker:and explain to it everything you
Speaker:wanted to do?
Speaker:And have a big database of it?
Speaker:Or what do you actually
Speaker:mean by that.
Speaker:I think if you fully start from
Speaker:scratch, maybe that's what you do,
Speaker:because you start like just talking
Speaker:with AI like about about the problem
Speaker:that you have to solve.
Speaker:In our case, we had a bunch
Speaker:of like all the SOPs, all
Speaker:the process all the decisions that
Speaker:we have made in place.
Speaker:So it's about okay, now that we like
Speaker:all this knowledge, how we
Speaker:codify it, I think one of the first
Speaker:things that we did, it's not only
Speaker:about codified acknowledgement of
Speaker:what was created, but it was
Speaker:like truly rethinking about
Speaker:the roles that we needed in place.
Speaker:So if you think about very entry
Speaker:levels of operations, it doesn't
Speaker:matter if it's a support function
Speaker:or like an ops or like a risk
Speaker:function, but you normally
Speaker:are expected to follow
Speaker:in detail a procedure, like,
Speaker:okay, you have to like follow those
Speaker:rules.
Speaker:I think for me, the first thing is
Speaker:if you're going to follow a
Speaker:procedure, you don't need a human
Speaker:anymore. That should be fully done
Speaker:by AI.
Speaker:So whatever is like this L1
Speaker:type of like role level, this just
Speaker:doesn't exist.
Speaker:And as long as we are all moving
Speaker:towards like this new environment,
Speaker:this is not as necessary,
Speaker:that's going to be the beginning.
Speaker:The second is when you grow
Speaker:in your career, normally you're
Speaker:training more people, normally
Speaker:you're becoming a leader or subject
Speaker:matter expert.
Speaker:And you're passing their content,
Speaker:like, what do you do?
Speaker:What is all this institutional
Speaker:knowledge to the team?
Speaker:You can still do that.
Speaker:But instead of passing to the Team,
Speaker:you're not only going to be joining
Speaker:at other
Speaker:different groups of people,
Speaker:you're actually going to join that
Speaker:with AI.
Speaker:So the expectation of
Speaker:this other type of role is,
Speaker:OK, you need to be able to put your
Speaker:SOP in a prompt.
Speaker:You need to able to.
Speaker:Not only look at the metrics,
Speaker:SLAs of the teammates of a group,
Speaker:you actually need to be able to
Speaker:do an evolve and see if
Speaker:the agents are performing.
Speaker:So you still have the same type
Speaker:of fundamentals of
Speaker:an operator, but you're applying the
Speaker:fundamentals to the agents
Speaker:because the agent should be part of
Speaker:the team. And then you can still
Speaker:have, okay, but who's thinking about
Speaker:strategy? Who's thinking the design,
Speaker:like the small level tree as you
Speaker:grow in the ladder, right?
Speaker:Those are still there, but instead
Speaker:of like, okay, how do you create
Speaker:like the organization environments?
Speaker:How do you think about August
Speaker:structure?
Speaker:You should actually be thinking
Speaker:about systems that's compound.
Speaker:You should be thinking like about
Speaker:process that I mean, you have the
Speaker:proper feedback loop that are only
Speaker:getting better.
Speaker:You shouldn't be thinking about how
Speaker:you break down a problem into
Speaker:a hundred micro steps.
Speaker:Because if you break it down in
Speaker:many, many tiny steps, you can
Speaker:actually have like the agents
Speaker:building those, which in the past
Speaker:was just not possible.
Speaker:So I think for me,
Speaker:I'm also trying
Speaker:at Matomic to transition
Speaker:us into being an AI native company.
Speaker:And on the engineering side, that's
Speaker:a lot easier, partially
Speaker:because engineers understand more,
Speaker:but also because the technology has
Speaker:been built more for them and cloud
Speaker:code and context
Speaker:and the software around the
Speaker:models exists already.
Speaker:I'm finding it more difficult
Speaker:on the commercial
Speaker:side Because.
Speaker:Somewhat because of people, but more
Speaker:because of technology.
Speaker:And some of the software just
Speaker:doesn't do what I want it to do yet.
Speaker:And I don't think it does what the
Speaker:teams want them to do.
Speaker:Yet. So what are you recommending?
Speaker:What software do you think are
Speaker:things that exist are actually good
Speaker:so that I should think.
Speaker:Because it's not only about the
Speaker:software itself.
Speaker:Cloud code is great.
Speaker:I think there is a fundamental
Speaker:change that you need to do around
Speaker:the talent and say, hey, now you're
Speaker:expected.
Speaker:You shouldn't be expected to do it
Speaker:yourself. You should be expected to
Speaker:leverage in AI and how you do that.
Speaker:A simple example, and let's
Speaker:use support because it has
Speaker:a very customer-facing role.
Speaker:Everyone talks about, okay, everyone
Speaker:thinks about using the software,
Speaker:which is the chat bot.
Speaker:To really reduce contact rate.
Speaker:I think beyond that is now
Speaker:when we train the new
Speaker:support class, we train
Speaker:them without an instructor
Speaker:in the room.
Speaker:We train them truly with AI.
Speaker:In the past, you had to read a bunch
Speaker:of things, watch a bunch videos, be
Speaker:trained to follow someone
Speaker:else that had done the job.
Speaker:Now, you're actually trained by
Speaker:an AI agent because that's
Speaker:what you should be doing.
Speaker:These AI agents can put you to
Speaker:any knowledge management.
Speaker:If you are able to ask the questions
Speaker:very quickly, you are to
Speaker:instruct and gather the information.
Speaker:So if the training starts already
Speaker:different, you get a teammate
Speaker:that is much more proficient
Speaker:with AI to resolve a
Speaker:customer challenge.
Speaker:And then when you go deep, okay,
Speaker:Camila, but what's the software I'm
Speaker:trying to do that is not as simple
Speaker:because then people start using.
Speaker:I think that's where I
Speaker:think Brex had made the right
Speaker:investment.
Speaker:We invest in ourselves building
Speaker:a lot of it, building a little
Speaker:like the infrastructure.
Speaker:So there was a lot of like
Speaker:engineering investments to
Speaker:getting the whole company ready and
Speaker:operations ready too.
Speaker:So I don't think it's just ops doing
Speaker:it by itself but it's actually a
Speaker:very close partnership into
Speaker:getting this done because you need
Speaker:to create like the tools, the
Speaker:systems and the first layer that
Speaker:allow people to build on top of
Speaker:that. Now, maybe what I think if
Speaker:I was going into detail here is,
Speaker:I remember that I had the same
Speaker:question as you, Beth, and
Speaker:I was gonna propose this whole
Speaker:change to our board.
Speaker:And I was nervous because I thought,
Speaker:okay, the board will say, Camila,
Speaker:why are we gonna invest on
Speaker:reinventing operations if we can
Speaker:invest in having AI within the
Speaker:product?
Speaker:But when I presented,
Speaker:the feedback was actually very
Speaker:positive. Everyone was like, that's
Speaker:amazing, because if you do
Speaker:invest internally, and you have the
Speaker:right tools, you build the taste
Speaker:that you need to build the products
Speaker:that the customers want.
Speaker:It was actually everyone agreed,
Speaker:let's just double down and
Speaker:invest in revamping operations
Speaker:with AI, also from an engineering
Speaker:and R&D perspective.
Speaker:What did that look like?
Speaker:Like?
Speaker:I mean, without revealing any trade
Speaker:secrets, additional engineering
Speaker:hired, what's the infrastructure
Speaker:like?
Speaker:Share as much detail as you can.
Speaker:Then I got everyone on board.
Speaker:I sent them the memo and
Speaker:an email to the whole company.
Speaker:That's what I expected the road to
Speaker:be. And this became part of
Speaker:our roadmap.
Speaker:But the reality is I didn't feel
Speaker:things were moving fast enough.
Speaker:And then this was maybe what,
Speaker:May last year.
Speaker:And I was like, oh, all those new
Speaker:companies, they are able to move
Speaker:quick. So what's the difference
Speaker:between what they're doing and what
Speaker:are you doing?
Speaker:And then I think it was more like
Speaker:a joke in the beginning, you said,
Speaker:okay, what if we put people
Speaker:together? What if we say we have
Speaker:to operate as a seed company, and
Speaker:we have a group of people that will
Speaker:do that? And that leads it.
Speaker:So I remember that I was like, okay
Speaker:let's do it.
Speaker:I was sitting here, let's let's
Speaker:recruit your engineers within Brex
Speaker:to do this project.
Speaker:And then, I think in two or three
Speaker:days, I basically said, okay,
Speaker:this is our budget from a- very
Speaker:underfunded seed
Speaker:company.
Speaker:We're going to work from this
Speaker:office.
Speaker:We're gonna cancel all the meetings.
Speaker:We're not going to join like regular
Speaker:breaks routines.
Speaker:And including myself, I cancel all
Speaker:my external meetings.
Speaker:And I said, I will spend
Speaker:four to six weeks in a
Speaker:group with like everyone joining
Speaker:the office, going to the office
Speaker:together, spending a whole day
Speaker:with those problems.
Speaker:What we did, we chose
Speaker:great engineers.
Speaker:That have been dealing with those
Speaker:problems that have MIMO-like AI
Speaker:proficient.
Speaker:And we got like the subject matter
Speaker:experts with them.
Speaker:And then it was like ops and ends
Speaker:together to truly
Speaker:try to rebuild all those solutions.
Speaker:To think, okay, this is like what
Speaker:SOP says and then engineer,
Speaker:okay. But what do you do when you
Speaker:get that use case?
Speaker:Okay, so that's what you have to
Speaker:codify to the agent.
Speaker:Okay, and then like building like
Speaker:those multi-agent systems that
Speaker:allow us to like make better
Speaker:decisions when you're talking about
Speaker:fraud.
Speaker:Which was like high risk for us.
Speaker:So having this room to like put
Speaker:people together and say, you don't
Speaker:leave until you find a new
Speaker:way of like this to operate was
Speaker:actually very helpful.
Speaker:In four weeks, actually in six
Speaker:weeks, we launched like our
Speaker:first new
Speaker:reinvented onboarding flow,
Speaker:which was for a very specific
Speaker:segment of customers,
Speaker:fully automated, which was also the
Speaker:foundation for a lot of like
Speaker:investments that we made after that.
Speaker:And then when you circle back to
Speaker:that board pitch, so this is kind
Speaker:of, I guess, earlier in the cycle
Speaker:slightly, in the board pitch like
Speaker:what was your pitch?
Speaker:If you can describe it a little bit
Speaker:like, because always in these board
Speaker:meetings, there's this question of
Speaker:AI, you know, are you using AI?
Speaker:How effectively are you are using
Speaker:it? Why are you not doing more of
Speaker:it? Why are we adding all this
Speaker:headcount, et cetera, et cetera.
Speaker:So for an operator, for a CO to come
Speaker:into a board meeting to kind of
Speaker:pitch them around.
Speaker:Kind of the concept of like
Speaker:reinvented workflows and job roles
Speaker:and whatnot.
Speaker:Like what was your, what was a bit
Speaker:of the detail around the
Speaker:presentation itself and how you kind
Speaker:of pitched it.
Speaker:I'll break it down in three parts.
Speaker:The first one is I needed
Speaker:a business outcome.
Speaker:I need to convince them that if
Speaker:I was getting investments, that it
Speaker:would be real ROI.
Speaker:So here we measure cost
Speaker:to serve.
Speaker:So we have this whole operations,
Speaker:how much cost
Speaker:to serve the business.
Speaker:If you're doing the same project
Speaker:with more commercial functions, you
Speaker:could think about CAC.
Speaker:But whatever it is, for me, this
Speaker:is the cost to solve of breaks
Speaker:today. I think I can bring this
Speaker:down by half,
Speaker:which makes Brexit by half.
Speaker:That's what I mean.
Speaker:Well, no wonder the board was like,
Speaker:here, have some money.
Speaker:I think we're like already.
Speaker:So I said half in 24
Speaker:months.
Speaker:We are thinking like halfway through
Speaker:that in less than 12 months.
Speaker:So, I actually think we are going to
Speaker:be able to deliver like shorter
Speaker:than I actually promised.
Speaker:But anyways, I had the number.
Speaker:I had that's I want to do this
Speaker:is what we're gonna this is why
Speaker:And the reason behind that wasn't
Speaker:because I would cut roles or
Speaker:anything. No, it was because I would
Speaker:be able to grow as fast as
Speaker:we wanted without investing
Speaker:into all these ops machines, because
Speaker:that's where I would see all the
Speaker:AI benefits.
Speaker:So then, okay, first, go business
Speaker:outcome.
Speaker:Second was, okay.
Speaker:How are we going to change the team?
Speaker:Which was what I explained to you.
Speaker:That was my first thing.
Speaker:This is how the roles would change.
Speaker:This is the type of investments that
Speaker:we do in like L1 roles and that's
Speaker:why it scales so quickly with
Speaker:revenue. If we don't need to do this
Speaker:anymore, this is one of the reasons
Speaker:that we're gonna get true to the
Speaker:outcome that I mentioned.
Speaker:So when I explained what the roles
Speaker:would look like and how we'd get to
Speaker:the business outcome, I was like,
Speaker:okay, but what is the level of
Speaker:investments we need?
Speaker:And then was there needs to be a
Speaker:priority for engineering too
Speaker:because we need to build those
Speaker:systems, the platform that allow
Speaker:Ops should be a contributor that
Speaker:allow ops to do the evolve
Speaker:themselves, that allow ops to
Speaker:codify with the prompt in a safe
Speaker:way.
Speaker:And that was the third piece to
Speaker:close the loop.
Speaker:So if I would go back,
Speaker:the business would come, how
Speaker:we would change the team,
Speaker:and what were like the investments
Speaker:from an engineering capacity.
Speaker:And then how did you figure out what
Speaker:technology to use?
Speaker:Did you left that to engineering or
Speaker:did you already have some ideas?
Speaker:I had an idea of how the system
Speaker:would look like,
Speaker:how that we should have to break
Speaker:down the problems.
Speaker:But in the end, this was
Speaker:in partnership with engineering,
Speaker:right? When we did this first,
Speaker:we call internally Hacker House,
Speaker:when we did the first Hacker house,
Speaker:we already had some development,
Speaker:okay, how we were investing to
Speaker:automate the whole KYC.
Speaker:But the reality is we didn't know
Speaker:how we could underwrite a customer
Speaker:without any human touch.
Speaker:So it was us all going together,
Speaker:OK, how do we make this even
Speaker:possible?
Speaker:But I wouldn't say there was a lot
Speaker:of like new software
Speaker:or new things.
Speaker:It was a little like, OK.
Speaker:Using cloud, using like
Speaker:all the APIs with like, I mean,
Speaker:we tested Gemini.
Speaker:We tested Shared GPT.
Speaker:We tested cloud and we saw
Speaker:all the performance of the models.
Speaker:And we still do.
Speaker:And using tools that we already
Speaker:have, we still use like Retool,
Speaker:which is a tool that we use.
Speaker:A break today, but the
Speaker:building honestly was just with
Speaker:the models that are available.
Speaker:We didn't need to retrain the model,
Speaker:there was no RL or anything extra
Speaker:advanced here.
Speaker:There were instructions around how
Speaker:you build the agents, how they
Speaker:connect, which the engineer
Speaker:obviously had a ton of input there,
Speaker:but it was not any specific
Speaker:software.
Speaker:Well, I guess it's more, or if
Speaker:you've built things in-house,
Speaker:because the areas where we've ended
Speaker:up needing to build is
Speaker:our data structure
Speaker:so that we can access data in the
Speaker:right ways, some amount of
Speaker:sharing skills or sharing
Speaker:information between spaces.
Speaker:MCP servers aren't always great.
Speaker:The actions aren't always great,
Speaker:like there's a lot of engineering
Speaker:needed around the edges.
Speaker:So it's not the model so much
Speaker:anymore that I think matters as.
Speaker:How do you get the best out of the
Speaker:model?
Speaker:And that's the part that we're
Speaker:focusing on building at the moment.
Speaker:So we use Clay.
Speaker:It was one of the softwares that I
Speaker:think it was great that we had to
Speaker:integrate that gave us a lot of
Speaker:access to LinkedIn data.
Speaker:When I'm thinking, we use like some
Speaker:starting companies, we use Accent
Speaker:to spread financials, which was
Speaker:actually very helpful into how
Speaker:we collecting.
Speaker:We use a lot like data aggregators
Speaker:that had been adopting AI,
Speaker:but that have been in the market
Speaker:even before the last
Speaker:three years.
Speaker:So...
Speaker:We use a bunch of them, not I don't
Speaker:want to pretend here that we built
Speaker:it all in-house, but it was
Speaker:a mix, but the infrastructure of how
Speaker:we would build the agent, we just
Speaker:mute our own.
Speaker:I think it's just because I'm so
Speaker:deep in it right now and I was like
Speaker:what's the silver bullet how can I
Speaker:do this in a way that's easier than
Speaker:what's happening because some of it
Speaker:is culture change but a lot of it
Speaker:I'm finding right now is people are
Speaker:ready to go and the technology
Speaker:isn't always
Speaker:there without needing to
Speaker:build.
Speaker:Yes, one thing that we
Speaker:definitely saw as an example,
Speaker:I wanted to automate the full
Speaker:dispute process, right?
Speaker:Dispute, in my words, here is
Speaker:a customer use the cards, they
Speaker:don't recognize the transaction,
Speaker:they come to brex and they want to
Speaker:dispute. So there are mood steps
Speaker:here. We have like
Speaker:teams and teammates like walking
Speaker:behind the scenes.
Speaker:So the first step to do that
Speaker:honestly started fooling operations
Speaker:without any engineering resource.
Speaker:We basically, as a teammate to say
Speaker:hey. What can you do
Speaker:because you're not going to be able
Speaker:to prioritize from an engineering
Speaker:perspective like this flow here?
Speaker:Then they created the gems, which at
Speaker:the time was like the agents that
Speaker:you could build with Gemini.
Speaker:Just by then creating the agents,
Speaker:being able to record, putting all
Speaker:the knowledge management there
Speaker:and the hundreds and hundreds of
Speaker:pages from all the master cards,
Speaker:books of how you file the disputes,
Speaker:we were able to automate
Speaker:50% of the
Speaker:flow. 50%, I always say
Speaker:that that's when people stop and get
Speaker:to the main. Mistake, because if
Speaker:you automate 50%
Speaker:of something, it will still,
Speaker:you may get 10% of
Speaker:efficiency.
Speaker:Because unless you fully automate
Speaker:the whole flow, and
Speaker:maybe this is the silver bullet for
Speaker:me, unless you're fully remove the
Speaker:whole floor, it's not enough.
Speaker:So, okay, this feels great.
Speaker:So then we lay in
Speaker:the engineering resource to
Speaker:build the other layers that
Speaker:need to be done, to actually
Speaker:integrate this in fully.
Speaker:Finish the automation of the
Speaker:process. Now this is
Speaker:going live.
Speaker:This has been tested and going live
Speaker:and with this, this is like one of
Speaker:the main things that is helping us
Speaker:to deliver our cost to serve goals
Speaker:because before I had
Speaker:more than 30 people working today
Speaker:with like a lot of our
Speaker:teams working overnight and
Speaker:now I don't need any of the other
Speaker:our teams anymore.
Speaker:So we've just raised a series B.
Speaker:We have a ton of cash.
Speaker:We're now spending it to hire a
Speaker:bunch of people.
Speaker:You know, we're obviously on the
Speaker:software development side, things
Speaker:are, we're very cognizant of
Speaker:like reworking a lot of our flows
Speaker:and kind of the software developers
Speaker:themselves were be hiring probably
Speaker:half of what we would have
Speaker:classically done, expecting the
Speaker:software development team to be much
Speaker:more productive. And we can see
Speaker:we're on a good trajectory for that
Speaker:part of it.
Speaker:I guess what I'm wondering, is on
Speaker:the back office side of things.
Speaker:We need to create space for the
Speaker:organization to be able to have
Speaker:space to think about their jobs,
Speaker:think about what they're doing,
Speaker:think about our flows, start to
Speaker:think through how best to rework
Speaker:them. We have a program or
Speaker:initiative in the company where
Speaker:we're pairing engineers
Speaker:with other people in the companies,
Speaker:like back office as an example, to
Speaker:work on particular problems
Speaker:of SOPs or whatever, similar
Speaker:to what you're describing where...
Speaker:There's a particular issue or
Speaker:workflow that we have that's highly
Speaker:repeatable, work with the engineer
Speaker:and that individual, reimagine
Speaker:it and automate it using
Speaker:AI effectively.
Speaker:So we're starting on that journey a
Speaker:little bit. But what I'm wondering
Speaker:about is the acceleration
Speaker:part of it.
Speaker:And right now we all have day jobs,
Speaker:including myself.
Speaker:So, you know, in my back office
Speaker:team, whether or not it's worthwhile
Speaker:or sensible even to
Speaker:hire somebody that has
Speaker:100% capacity.
Speaker:To work wholesale on some of
Speaker:the operational issues that we have
Speaker:with the software development team
Speaker:as opposed to piecemealing right now
Speaker:where we have bits and pieces of
Speaker:myself and some of my team, some of
Speaker:team working with the engineers on
Speaker:spots of a problems on a limited
Speaker:time basis.
Speaker:Should I hire somebody with my team
Speaker:to help?
Speaker:Should I actually work with the
Speaker:engineering lead to hire more
Speaker:engineers perhaps to have more space
Speaker:on their side to work with our side
Speaker:internally to do more of
Speaker:this? What do you make of that?
Speaker:Excellent question, and I think
Speaker:that's spot on.
Speaker:Ideally, I would recommend you to
Speaker:get someone that
Speaker:is doing the back office
Speaker:job, that is leading it, that
Speaker:has a lot of expertise
Speaker:in the subject, whatever you want
Speaker:to automate to be part of the
Speaker:process. For two reasons,
Speaker:they know best, they know
Speaker:the institutional knowledge.
Speaker:If you bring someone fully external,
Speaker:they would have to learn all of that
Speaker:and try to codify.
Speaker:And normally that's when people
Speaker:make a ton of mistakes.
Speaker:So if you're able to bring this
Speaker:someone to lead
Speaker:the initiative and pair them
Speaker:up with like the engineers,
Speaker:I think that is very important.
Speaker:I would say that maybe hire
Speaker:a few engineers that know and
Speaker:have built in an environment like
Speaker:more like AI native does helps.
Speaker:So maybe I would invest there, but I
Speaker:leverage operations subject matter
Speaker:experts.
Speaker:And the reason why I would push
Speaker:for that or like I would advise
Speaker:for that is Because you're
Speaker:not going to need just one flow.
Speaker:You're going to needs multiple
Speaker:subject matter experts.
Speaker:So you want the talent to be like,
Speaker:OK, you kind of create one.
Speaker:You prove like points with this.
Speaker:And you want this to be replicable,
Speaker:right? The engineering team can go
Speaker:and move on to resolve all the
Speaker:problems. But then, whoever was
Speaker:able to create this new flow is who
Speaker:you'll be working towards how
Speaker:you actually QA, how you'll
Speaker:actually keep improving the
Speaker:agent itself.
Speaker:Because it's not a once and done.
Speaker:The models keep getting better.
Speaker:You need to always keep testing the
Speaker:response that are done.
Speaker:And if you are able to be part of
Speaker:the build, then you create the new
Speaker:L2 of my crazy matrix
Speaker:that in my head does make sense and
Speaker:has been working.
Speaker:But then you created this new L two
Speaker:because this person can actually
Speaker:train in this new workforce model.
Speaker:Does that make any sense?
Speaker:Yeah, it does.
Speaker:What would you suggest in terms of
Speaker:the time allocation?
Speaker:Right now, similar to Bethany,
Speaker:we had a company
Speaker:offsite. We spent a day doing this
Speaker:kind of activity.
Speaker:We now have this initiative that
Speaker:we've set forth, but we haven't
Speaker:really constructed the company in
Speaker:some form as to, yeah, you should be
Speaker:spending every second Friday doing
Speaker:this very specifically with your
Speaker:paired engineer as an example.
Speaker:So, if you had to recommend kind
Speaker:of space-time allocation for...
Speaker:Of in-house experts, whether it's
Speaker:the operations team or it's GTM team
Speaker:members to work with their paired
Speaker:person to work on
Speaker:workflow issues like this.
Speaker:What would you suggest in terms of
Speaker:allocation of time?
Speaker:I wouldn't do like once a week.
Speaker:I think this needs to be continued.
Speaker:So I would highly recommend at
Speaker:least a whole week, fully
Speaker:focused on that.
Speaker:Ideally, you're gonna need more.
Speaker:Across the company to be nervous.
Speaker:No, I don't think any.
Speaker:No, no, no. I cross the company.
Speaker:No. Because then you can't stop to
Speaker:only do that.
Speaker:I would try to choose, okay, let's
Speaker:prove that this model can work.
Speaker:Choose like a group of engineers
Speaker:that someone should lead,
Speaker:that will push back on everything.
Speaker:And then choose some, a
Speaker:few subject matter experts and say,
Speaker:hey, this is your day job.
Speaker:Your group for the next one to
Speaker:four weeks will be fully focused
Speaker:on this.
Speaker:And you got to deliver.
Speaker:And Don't you create them.
Speaker:To choose the most complex workflow,
Speaker:but choose something that you can do
Speaker:end to end.
Speaker:It did help that I was in the
Speaker:room because you normally have
Speaker:like leadership trade-offs.
Speaker:And even though the team was
Speaker:absolutely amazing and they did all
Speaker:the work, I think
Speaker:being there in the world allowed
Speaker:them to move quick, to not having to
Speaker:go to other teams to get feedback on
Speaker:how to do that because I was just
Speaker:breaking down like the decision
Speaker:right there for them.
Speaker:Oh, but coming up we do A or B.
Speaker:So instead, they're like, OK,
Speaker:so now I have to wait this
Speaker:Friday meeting to get everyone on
Speaker:board. No, we made the decision
Speaker:quicker.
Speaker:And even if you can't allocate your
Speaker:full time, today's all like you
Speaker:don't have a leader that could
Speaker:allocate four weeks to
Speaker:this, allocate all the mornings.
Speaker:Allocate a block of like three hours
Speaker:a day.
Speaker:That's then all the decisions that
Speaker:this group needs should be done.
Speaker:Someone is unblocking right away,
Speaker:and everyone in the company will
Speaker:respect that. I think so then the
Speaker:leadership, subject matter experts.
Speaker:And the technical
Speaker:background is the
Speaker:perfect combination.
Speaker:Everything you're sharing is so
Speaker:valuable that I feel like I just
Speaker:want to ask you, what do
Speaker:you, it's a bit of a reflective
Speaker:question, but like what are the
Speaker:five biggest tips you have or the
Speaker:five things you learned that
Speaker:everybody should focus on?
Speaker:I think we definitely have
Speaker:leadership needs to be involved.
Speaker:You need intensity of work
Speaker:and do something
Speaker:end to end.
Speaker:Do a process end to and rather
Speaker:than trying to just optimize a
Speaker:process. What are your other pearls
Speaker:of wisdom.
Speaker:Expert with the technical team
Speaker:that really helps.
Speaker:And then we already talked about the
Speaker:end-to-end. It's slightly different
Speaker:but very similar.
Speaker:I would say have
Speaker:a very clear goal.
Speaker:What do you want to achieve?
Speaker:Not only like automate something but
Speaker:what actually you
Speaker:expect? Is it a full
Speaker:automate process end- to-end or is
Speaker:it 80% less
Speaker:of something or you want gets a
Speaker:much higher conversion
Speaker:rates, but having a very clear goal
Speaker:that helps with the trade-offs
Speaker:helped us here too.
Speaker:So I want to be able to address
Speaker:the markets that we are not able to
Speaker:to address.
Speaker:I thought this was easier because
Speaker:for this first experiment that
Speaker:we had, I didn't want
Speaker:the whole company to stop.
Speaker:I wanted to do something that was
Speaker:incremental because if I chose like
Speaker:something to prove a concept that
Speaker:was part of like the day-to-day,
Speaker:I would be running into a lot of
Speaker:roadblocks. So I chose a segment
Speaker:that in the past we had failed to
Speaker:serve, so then it would be crazy
Speaker:if we fixed.
Speaker:So I choose the segment and said,
Speaker:okay, we're going to find a way to
Speaker:serve the segment.
Speaker:But then it's kind of like a proof
Speaker:point. Okay, if you're able to solve
Speaker:this, now let's test with something
Speaker:else that scores the business.
Speaker:And the second one was like, okay.
Speaker:Actually, move from this little
Speaker:tangential thing to the
Speaker:largest flow that we have, to the
Speaker:largest volume, and we said, now we
Speaker:need to fix onboarding.
Speaker:So before onboarding was take five
Speaker:business days, I want to do this.
Speaker:In five minutes.
Speaker:So this five to five, it was
Speaker:what led us to
Speaker:believe in what it means.
Speaker:So did you actually look
Speaker:through the business and identify
Speaker:your biggest wins and
Speaker:the most painful areas and
Speaker:then pick them off?
Speaker:The way that I prioritize what we
Speaker:need to invest was different.
Speaker:So the first time we always had
Speaker:the prioritization and then I went
Speaker:then, hey, I need to deliver this
Speaker:business goal. I need you to drop
Speaker:cost to serve by half in 24
Speaker:months.
Speaker:So that was very clear to me.
Speaker:And then I had to prioritize, okay,
Speaker:what's driving cost to server and
Speaker:like, what are the things that I'm
Speaker:gonna tackle? So that's how
Speaker:everything started.
Speaker:For the first operating model
Speaker:experience, which was like how we
Speaker:created the Hacker House, I tried to
Speaker:create something that was actually.
Speaker:Not a business goal for that
Speaker:fiscal year, because if I
Speaker:was able to do that, I could move
Speaker:quick and deliver upside
Speaker:that then everyone would say, okay,
Speaker:we have to invest on this new
Speaker:operating model.
Speaker:For the second one, since I had a
Speaker:very proven and clear model,
Speaker:it was more like, okay this is the
Speaker:way that make things happen and
Speaker:quicker. So let's choose like a real
Speaker:problem that will deliver
Speaker:the business's results for this
Speaker:fiscal year and tackle the cost to
Speaker:serve goals. And again,
Speaker:we did it!
Speaker:Now, I don't need to be in other
Speaker:rooms anymore because people have
Speaker:learned how to operate.
Speaker:This has been spread out in
Speaker:multi-blocks of multi-areas
Speaker:of the business.
Speaker:What were your biggest
Speaker:surprises?
Speaker:The remaining.
Speaker:For the first one, I was
Speaker:truly surprised that it worked so
Speaker:well and so fast, in a sense.
Speaker:I was like, okay, that's impressive.
Speaker:That's amazing. I was really
Speaker:surprised to see how
Speaker:Everyone was adopting to AI
Speaker:in different ways.
Speaker:It was so funny for me
Speaker:to be in the room and some engineers
Speaker:be talking with AI to code,
Speaker:also seeing operations being able to
Speaker:create evolved data sets that they
Speaker:didn't even know what it was,
Speaker:I myself connecting to an API
Speaker:to be able to do.
Speaker:So I think I was surprised that
Speaker:when you put it there in the
Speaker:challenge and you remove
Speaker:distractions, everyone.
Speaker:Got excited to learn something new,
Speaker:and even if you didn't, that was
Speaker:a good surprise.
Speaker:On the not as great surprise,
Speaker:on the second one that we did,
Speaker:we chose a much larger problem that
Speaker:was already core for how the
Speaker:business ran.
Speaker:The first one, we could wait
Speaker:and release the
Speaker:code release, go live all together.
Speaker:For the second, we released
Speaker:in batches.
Speaker:As an operator, I'm always
Speaker:pushing back on engineers when they
Speaker:release things and causing stands if
Speaker:it's not fully ready.
Speaker:So for the second one, actually
Speaker:this, this group that I was leading
Speaker:created a ton of incidents,
Speaker:which is a operator nightmare.
Speaker:So like, I was like, almost like
Speaker:when we would go to the stability
Speaker:meetings, I like, Oh my gosh,
Speaker:this was like a, there's this
Speaker:process that was leading, but I
Speaker:mean, you don't build until you
Speaker:break things.
Speaker:Right. So it was interesting
Speaker:to see when you're really
Speaker:like speeding up.
Speaker:You need to be careful.
Speaker:So sometimes doing something fully
Speaker:on the side is much easier.
Speaker:That's why companies that are
Speaker:studying is normally much easier
Speaker:when you have to do things on the
Speaker:core, you're gonna need to trade off
Speaker:and you're going to need to understand
Speaker:that the mistakes will happen.
Speaker:You need be there to fix quickly.
Speaker:And the outcome is real.
Speaker:And so you were just brave about
Speaker:making mistakes and accepting
Speaker:they're going to happen as long as
Speaker:you're there to fix them as fast as
Speaker:you can.
Speaker:At the moment, I got some critiques
Speaker:on that and I criticize myself,
Speaker:but after seeing this,
Speaker:after the outcome was
Speaker:so great, yes.
Speaker:So there's a lot of bravery there.
Speaker:How many sleepless nights did you
Speaker:have? Oh, many.
Speaker:It's funny. I was like leaving the
Speaker:office midnight, like those days
Speaker:when we were doing the Hacker
Speaker:Houses.
Speaker:We are having lunches and dinners
Speaker:in the office, but it was fun.
Speaker:I'm not joking. Like, after
Speaker:that, it's like, yes, you're looking
Speaker:for things to go back to normal.
Speaker:But everyone involved on that was
Speaker:like, oh, this was like one of like
Speaker:the best moments of my career.
Speaker:I learned so much into this.
Speaker:So it's a trade off, but I think it
Speaker:was fun. I think that's continuously
Speaker:you can do this every day,
Speaker:but that's why I think like
Speaker:intensity for like a short period
Speaker:and that, you know, beginning and.
Speaker:Is really helpful.
Speaker:So here's a question for you.
Speaker:So now that you're no longer in the
Speaker:room and there is something you
Speaker:inferred this slightly but
Speaker:throughout the organization you've
Speaker:kind of it's now self-perpetuating
Speaker:to some extent where folks are just
Speaker:doing this.
Speaker:Organizationally as a CEO have you
Speaker:done anything to support the
Speaker:company to ensure that it was going
Speaker:to be embedded?
Speaker:So any kind of like you know
Speaker:showcasing of demos type
Speaker:situation or what was your way of
Speaker:vetting it.
Speaker:Two different things.
Speaker:One is from an engineering
Speaker:perspective and keeping the
Speaker:ops partnership with engineering
Speaker:because I think it's key.
Speaker:I want to make sure that it wasn't a
Speaker:road map and I'm still very
Speaker:close to overall cost
Speaker:of the road map, how we're going to
Speaker:prioritize and why.
Speaker:So we continue to deliver the goals.
Speaker:On the other side, I want make sure
Speaker:that I have more and more SMEs
Speaker:to get ready for that.
Speaker:So we have monthly ops on
Speaker:hand where we always celebrate
Speaker:the AI spotlight.
Speaker:During the whole month, we
Speaker:incentivize all the new,
Speaker:everyone in the team to showcase
Speaker:and make sure that they are creating
Speaker:their prototypes or things that are
Speaker:helping them to be much more
Speaker:productive so we can roll out across
Speaker:the organization.
Speaker:And we define change recording.
Speaker:The case studies for joining
Speaker:operations are completely different
Speaker:now.
Speaker:You are expected to use
Speaker:AI to resolve the case study and
Speaker:I actually asked you to see the
Speaker:prompts.
Speaker:Of what everyone uses as part of
Speaker:the case study, and I joined
Speaker:the final round.
Speaker:So we have this entry level that
Speaker:everyone helps join this
Speaker:entry-level, and we have
Speaker:final rounds of interviews, and
Speaker:that's when we receive the prompts
Speaker:from everyone.
Speaker:So we talked a bit about some of
Speaker:the restructure in terms of removing
Speaker:the L1s or not hiring more L1.
Speaker:But if you found changes to
Speaker:roles,
Speaker:and are you rewriting job
Speaker:descriptions? Are you just letting
Speaker:it kind of organically change?
Speaker:Yes, so it's definitely
Speaker:like rewriting job descriptions,
Speaker:like the job to be done change.
Speaker:In the past, you expect to make a
Speaker:decision, follow a procedure.
Speaker:Now what I'm asking you is to
Speaker:create an agent to do what was the
Speaker:procedure that you wanted and it
Speaker:needs to perform.
Speaker:So the business outcome is still the
Speaker:same but sometimes the skills
Speaker:change.
Speaker:So as much as I want everyone to
Speaker:get there, to get the skills, it's
Speaker:also a different profile.
Speaker:So that's why I need to change the
Speaker:recording to get like the right
Speaker:profile in house.
Speaker:So it goes in a mix.
Speaker:You make sure that you give
Speaker:opportunities for folks to get
Speaker:there, and you keep hiring
Speaker:more like this talent that is very
Speaker:native using all those tools.
Speaker:And like in this mix, you get
Speaker:both, like people that know the
Speaker:business, people that don't know the
Speaker:tools, one teach each other, and you
Speaker:move forward.
Speaker:And have you found, so speaking to
Speaker:somebody else who's measuring their
Speaker:AI adoption by some
Speaker:of the normal metrics, plus how
Speaker:many handovers have been
Speaker:eliminated?
Speaker:Yes, this is awesome.
Speaker:We do have some metrics as
Speaker:handovers elimination.
Speaker:My only concern, and I'll go back to
Speaker:this, is sometimes when you
Speaker:eliminate some handovers, you still
Speaker:have a fee where that's whoever's
Speaker:looking to this.
Speaker:We maybe need to look at the whole
Speaker:history.
Speaker:So I always say that if you
Speaker:remove 70%
Speaker:of the handovers you never get
Speaker:70% off efficiency.
Speaker:You may get 10% to 20% of efficiency
Speaker:with the whole flow.
Speaker:So I really think like the
Speaker:more that you can, not only think
Speaker:about the handovers, but to think
Speaker:about this whole process, you
Speaker:get a real.
Speaker:Effective outcomes.
Speaker:Sorry, so that's what I mean, is the
Speaker:handovers are eliminated because the
Speaker:process has changed and is much more
Speaker:streamlined. So that's the way I'm
Speaker:talking about it.
Speaker:Oh, okay.
Speaker:Yeah. So like then, basically
Speaker:what we mapped is like all the flows
Speaker:that exist that are like operational
Speaker:heavy, and we're going tackle
Speaker:one by one.
Speaker:It's like an underwriting decision,
Speaker:a fraud review,
Speaker:a reward check, a
Speaker:the deal desk flow, whatever it's
Speaker:like, or how you talk with like the
Speaker:customer, how escalation happens,
Speaker:whatever, it is like those are the
Speaker:things that we're mapping to go
Speaker:tackle one more.
Speaker:You're mapping it and you're
Speaker:restructuring
Speaker:it rather than just automating what.
Speaker:Exists today?
Speaker:We are automating or like getting
Speaker:the team to get in the automations
Speaker:themselves.
Speaker:But yes, we're basically
Speaker:re-automating, reinventing all those
Speaker:flaws in a way that they become I
Speaker:need it.
Speaker:And are you finding that the flows
Speaker:are actually changing
Speaker:or the flow is the same,
Speaker:but now AI is
Speaker:doing it?
Speaker:It's a mix.
Speaker:Sometimes for AI to do it, you need
Speaker:to slightly change the flow.
Speaker:But in many situations, if it
Speaker:was a very prescriptive flow,
Speaker:AI can do it.
Speaker:But if it's not that prescripted,
Speaker:sometimes you have to redesign how
Speaker:the flow would look like.
Speaker:So it actually needs to be more
Speaker:prescript if not less for AI.
Speaker:Yes.
Speaker:So when that's why it's the beauty,
Speaker:you have to break down to very,
Speaker:very, very small steps because
Speaker:if it is tiny, it's very prescrptive
Speaker:and that you can fully do.
Speaker:We are running out of time,
Speaker:a fascinating conversation, but
Speaker:everybody has to answer the final
Speaker:question, which is
Speaker:if our listeners
Speaker:can only take one thing away from
Speaker:the conversation.
Speaker:It's possible, intensive matters.
Speaker:Make sure that you have operators
Speaker:working with engineers because
Speaker:this partnership is very powerful.
Speaker:Lovely. Thank you, Camilla, for
Speaker:joining us on the operations room.
Speaker:If you like what you hear, please
Speaker:leave us a comment or subscribe, and
Speaker:we will see you next week.
