Startup Explore

Kieran Taylor (00:00)
We are building AI for financial auditors.

We're really aiming at getting it to that first point where instead of having to have a human go through and prepare that first draft,

an AI agent is able to fully prepare the first draft of the work paper so that someone at the senior level can go through and start reviewing the work.

So I think there's some really interesting things we can do there in that space. And yeah, I don't want to give away too much, actually, but I think that's definitely one to watch out for.

Ali | Startup Explore (00:28)
Meet Kieran Taylor, the co-founder of Midship AI, a seed stage YC backed startup that is reimagining how financial audits are done. They're building AI specifically for the audit industry, which has long lagged behind in AI adoption. So let's jump in.

Ali | Startup Explore (00:44)
Kiran, thank you so much for coming. Very excited to have you on the show.

Kieran Taylor (00:48)
Absolutely great to be here.

Ali | Startup Explore (00:50)
Thank you. Well, let's dive in. Can you tell me what you're building and what is that big problem that you're trying to solve?

Kieran Taylor (00:57)
Yeah, for sure. We are building AI for financial auditors.

So if you think every public company in the US requires yearly, usually quarterly audits, where you have an external accounting firm come in, basically go through all of your financial statements, sample some of the transactions that happen in those statements, and then look for evidence to ensure that everything was correctly recorded as it actually happened in real life.

We're building tooling to automate a lot of that manual process that goes on during a financial audit and basically accelerate those procedures that financial auditors do today.

Ali | Startup Explore (01:29)
Nice, nice. And by the way, as someone who worked in the audit for seven years myself, I know a lot how manual the process is and how labor intensive it is. That's interesting. Now, can you give me a high level overview of the audit process and then where in that process you think midship provides most value?

Kieran Taylor (01:52)
Yeah, so it's broadly broken down into two categories of you have a test of controls and a test of details. So the first thing in order to look for is are there processes in place in the business to ensure that things happen as they should do? If you require a manager's approval for certain transactions above $10,000, can they actually prove out that that happens when a transaction is processed? The second part, which is a test of details is look through all of the historical statements that have happened in the last quarter or the last year.

and actually vouch and find the evidence to say that these happened as they were claimed by management. So really what we're focused on is that second part for the test of details, going through and actually collecting the evidence and running some procedural checks to make sure they were properly recorded. So what that could look like in practice is when you are recognizing accounts receivable, accounts payable, you'll go through and find invoices, orders to actually make sure that...

the documentation is there to prove out that what you have in your ledger is correct and accurate. So what Midship actually does is it takes in those working documents that are provided by the client being audited, understands all of the content inside of each of the documents, and then is able to run basic procedural checks to say if this is an invoice which is vouching for this financial transaction, do the amounts match, do the dates match, are all the details here accurate and correct? And that is what we're really focused on automating today.

Ali | Startup Explore (03:16)
That's really interesting. And I feel like if you guys are helping auditors with testing the samples, that's a huge manual work that can be automated. And that will be a significant impact. Now let's double click on that process. So essentially what you're doing is you're automating that lower level of work and you're putting the auditor still in the reviewer seat, right? And the auditor still can see how AI came up with the findings.

and be able to basically know and if need to, can basically track back into the logic. Is that correct?

Kieran Taylor (03:55)
Yeah, absolutely.

So yeah, every step that is taken by one of the AI agents, we link it back to one of the source documents to ensure that it's reviewable by an actual human at some point. So really the process we're looking to get to is a junior auditor, their first day on the job, just graduated from university. One of the first tasks that will be put on is basically going through each of these documents from a client, matching it up against the samples that are taken and putting it into an Excel work paper.

which is basically every row corresponds to one of the transactions being audited and they are linking out from each of the cells here back to the evidence which is the invoice, purchase order, delivery receipt. So once the junior auditors put that stuff together, a senior will come in, go through line by line, look at each of the source documents, match out each of the values and ensure that everything was prepared correctly. We're really aiming at getting it to that first point where instead of having to have a human go through and prepare that first draft,

an AI agent is able to fully prepare the first draft of the work paper so that someone at the senior level can go through and start reviewing the work.

we have on the two surfaces, have our web app and an Excel add-in. And so from either of those two surfaces, at any point in time, we want you to be able to very quickly look back and make sure that the documents were there, they're expecting, and then linked out from each of the values that we're actually using in each of those steps.

Ali | Startup Explore (05:11)
Awesome, nice. And what would be some of the other key features of MedChimp AI today?

Kieran Taylor (05:17)
Yeah, really what we're focused on is being hyper-focused on getting that experience right. We're still young company making sure that we have the highest value out there. But I think the biggest thing which a lot of customers come into engagements with is thinking in kind of pre-built modules that we have. So like, can you do AP? Can you do AR? Can you do revenue testing? What we've really built it for is configurability, where whatever the methodology that you're applying on an engagement, we can configure that in midship just using natural language. So if one of the checks that you need to perform

for a certain procedure is make sure that the amounts match across all of the documents. We want that to be very easily configurable. So that actually opens up a huge amount of possibilities. That can be someone's full-time role. It's just running those procedures there. So even though it's just test of details, it's just getting documents, lining them up, running some checks, actually what you can achieve with that configuration is pretty endless.

Ali | Startup Explore (06:08)
Got it, nice. And when it comes to the audit industry, they are kind of professionally very skeptical about new things. And how do you ensure the reliability and the accuracy of AI and its findings?

Kieran Taylor (06:21)
Yeah, I think in the early days, we just had to get very, very good at dedicating as much time as possible to evaluations and just having actually one of the most difficult parts of that. actually to rewind a little bit, myself and my co-founders met at previous company, which was an AI knowledge assistant, very similar to a chat with GPT-like experience, but plugged into enterprise data. And there we were very, very focused and kind of got into the rituals of running evaluations as often as possible, getting everyone on the team to chip into that process.

So we of had that muscle memory for actually doing that process itself. Coming into this fresh, the really difficult part of that was finding representative document sets that we could use for these evaluations. So a bunch of time had to go into investing into searching the internet, finding some public releases of basically PDF dumps that we could use for that. But once you have that process and you can kind of day by day chip away at that, you get a pretty representative view of what you're going to see for a client engagement.

And that is really, really just the cornerstone of getting it right and having some confidence that what you're claiming to someone on the pitch deck is actually going to be able to materialize in real life.

Ali | Startup Explore (07:24)
Nice, it. Going back to you mentioned you guys built the add-on for Excel. What can auditors do from working in Excel and having that native, Excel native add-on from Misha?

Kieran Taylor (07:37)
Yeah, so really right now it's kind of a one-way experience where we have the procedure is actually run within the web app. You export it out into a preliminary work paper and then you can view all of the work from inside that Excel add-in. What we're really focused on in the coming months is building the bidirectional relationship where if you need to start editing and marking up the work paper and making the edits, actually ingesting that back into midship so that you have a consistent view across the web app and the Excel add-in.

I think the really difficult part of that is what makes Excel so powerful is it's completely customizable. At any point, you can add rows, delete columns, change the format however you like. And understanding each of those intentions in a reliable way and making sure that that fits with the data model that we have in Midship is really what we're thinking through from a user experience. So like I said, right now, very much one directional in terms of how it's used, but hoping to make some big strides in the bidirectional nature of that.

Ali | Startup Explore (08:30)
And one of the obvious questions that would come to my mind is now talking about the security of client data. How do you ensure the security of the data? And then do clients get an option to pick whether they want to store their data in on-prem versus cloud VPC?

Kieran Taylor (08:48)
Yeah, so think we have oftentimes talking to customers, they immediately jump to, we need everything to be on-prem. We don't want anything to be run. Like if we could run the LLM on-prem as well, then that would be preferable. I think a lot of that is actually around managing expectations from this new kind of standard architecture for LLM based applications and making sure they understand that just because something is being sent to an opening eye provider or anthropic or whoever it is, that there's not actually any inherent security risk in there as well.

All of that to say is that our standard architecture is cloud-based SaaS. We bring the storage ourselves. We use external API providers as well. We will definitely work with clients if they have strict requirements around data residency in particular. But a lot of the time is from the data storage. Most of the work issues goes into working with the client to make sure they understand what the posture is that they're dealing with. And on top of that, we went through the SOC 2 certification process.

to make sure that we have the rubber stamp stuff there as well. I think a combination of both of those is spending time with the client, having some external view of the security posture helps out a lot with giving them that confidence to go into this world.

Ali | Startup Explore (09:59)
that's helpful. And speaking about LLMs, I believe LLMs are really good at extracting, searching and analyzing the data regardless of the file formats. Would you agree with that?

Kieran Taylor (10:13)
Yeah, I mean, it's a little bit of a it depends in terms of there's this. I mean, really the biggest innovation which we've kind of built the architecture of the product around is the ability to afford LMS and VLMS to understand unstructured working documents, PDFs, Excels, etc. And I think that that is it always is very much it depends. It's actually very difficult to put guardrails around. How do you understand at what point is the LM going to fall over when it's looking at a certain PDF?

Ali | Startup Explore (10:17)
Okay.

Kieran Taylor (10:42)
What I'm trying to say is there's certain tasks that you think should be very, very easy for a VLM to look at and for a human, passing a document is very obvious. But for example, if you have a table which spans over five different pages on a PDF, actually building a system which is able to reliably look at the page breaks, join a table together across each of the pages, actually becomes very difficult and subtly hard to achieve.

All of that to say is that making big strides in it as context length increases, this problem again, a lot easier as well. There's definitely still some very subtle ways that it falls over today.

Ali | Startup Explore (11:16)
Now zooming out at the professional services industry as a whole, I can think of many industries where AI has made significant traction. Like for example, if you take legal professions, you have tool called Harvey, but there's no such tool yet in the audit industry. And why do you think audit is basically lacking AI adoption?

Kieran Taylor (11:41)
very good question. I mean, I think a lot of it goes to what I'm saying around there are some very recent breakthroughs in understanding of working documents to a certain level of reliability that is required to actually make something useful. So that is to say, I don't think that a year ago that we could have made the current iterative midship and make it actually useful for customers. A lot of the breakthrough in understanding PDFs being able to reliably extract that data.

actually enables the way we've built the system today. And so I think without that, there is a big barrier to entry for that. I think a lot of the adoption in professional services has found usage in the traditional knowledge management. So kind of rag-based applications, you have a chat interface, you're asking questions which usually you'd have to go and mine some internet resources for. But in terms of actually processing reliably working documents, it really wasn't available until, like I said,

the last 12 months or so, and that has opened the door for products like Midship to come through and actually break into those industries.

Ali | Startup Explore (12:43)
Got Now, in terms of the use cases, can you provide some examples of use cases and maybe pilot programs, even in general terms?

Kieran Taylor (12:53)
Yeah, for sure. So you mean with midship today, like what are the kind of use cases we come across for? Yeah, for sure. Yeah, so I think I've touched on some of them already, but it's really focused on test of details for external financial auditors in particular. And the external piece is really key there because for someone who is internal at a company going through accounting records, they may have access to some of the internal ERPs and accounting systems and kind of plug directly into the source of truth.

Ali | Startup Explore (12:57)
Right, right.

Kieran Taylor (13:21)
and makes it very easy for them to go through and find certain information. Inherently for an external auditor, because the data which they're looking at is almost always working documents like PDFs which get sent across by emails, kind of the class of tasks that we're looking at for sorting through lots of dumps, finding the correct document for a sample, going through and actually matching that up in a way which the auditor does today, that's really the kind of niche in the ICP which we're going after. And that use case...

It varies a lot across different industries in terms of the type of the procedure that we're actually going to perform. But in terms of the broad use case of test of details, external financial audit is really, really what we're focused on today.

Ali | Startup Explore (13:59)
Got it, nice. And what has been the customer feedback so far and anything surprising that you've learned recently?

Kieran Taylor (14:07)
Yeah, the customer feedback has been very positive so far. I think some of those times when we're going into pitches in particular, there's kind of stuff that we gloss over which doesn't seem super impressive to us because we've been working on this problem head down for a while now. And there's certain things which even just the ability for high quality OCR PDFs and to pull out an invoice ID from the top of a document.

I can sometimes, it's not totally obvious that that maybe wasn't always a simple task to do. So I think really it's like aligning our expectations with customers has been something which I've had to align myself with a lot as well. I'm trying to think of something more like surprising than that, which I can give you. I feel like that wasn't a super surprising insight, but I'm sure there'll be something which comes to me.

Ali | Startup Explore (14:54)
Yeah, yeah, no worries, no worries. OK. And now, in terms of pricing, how do you approach pricing today?

Kieran Taylor (15:05)
Yeah, I think the advice which we got throughout YC is just like, keep it simple stupid, where it's really just if you can give a flat fee to customers, which is greppable and kind of reflects the value which we're actually adding, that is the easiest way to do it. And I think a lot of early startup pricing is very much finger to the wind where it's like, hey, you throw out a number, the next customer comes along, you double that number, you keep doubling until someone pushes back on it. And definitely like in the early days, that's I think that's how some of that goes.

Ali | Startup Explore (15:26)
Yeah.

Kieran Taylor (15:34)
But really what we coalesced onto is per seat pricing. Because you were accelerating a human auditors work today, it's very obvious how that scales as you add more seats to the product. And that's really how we think about it. And then in terms of the factor that we multiply that by to get to a dollar figure, it's really working backwards from how much time do we think we can save for an average auditor on a given day, taking some percentage of that which we think is fair, and then getting to a final number with the customer.

Ali | Startup Explore (15:58)
Got it. Got it. Well, I feel like if w from what you told me, what midship can do today, I feel like a lot of external audit firms basically can replace outsourcing to probably a midship because those firms have been getting a lot of pressure on their margins and they had to become more efficient. So on one hand, you have this increased complexity of audit, accounting standards and tax systems that increases your workload. And then on the other hand, you have like somewhat flat.

staff or maybe even declining. So the combination of two puts a lot of pressure for them to become more efficient. And from my knowledge, I feel like they've been trying to outsource a lot of the work to countries like India, but then you have some challenges now with the language, with the time differences and so on. And yeah, from what I've heard, I feel like Midship can basically replace maybe a lot of that work that the firms have been trying to outsource.

Kieran Taylor (16:58)
Yeah, yeah, it's definitely something which we thought about a lot. I think going back to the pricing question, one of the things we were hesitant about was if you do per seat pricing, and you're doing something where you're possibly like eliminating the need for increased headcount, does that kind of affect your model for like, hey, like you're pinning yourself in a year's time once you've saved all this money to actually reduce the value which you're charging. So I think from like, the way which we've really seen

Audit standards from a regulatory and industry standard perspective as well is that as new tooling is introduced and auditors can move faster and do more samples and actually increase the quality of an audit, you end up saturating whatever the capacity is within the workforce. So we do definitely see a world where the increased efficiency per head doesn't lead to any head count reductions. It actually just leads to high quality audits and more thorough procedures.

But I think in terms of the offshoring question, it's definitely interesting. I don't think we've seen how that has affected other professional services industries enough yet to see how the pattern is going to represent itself across other industries as well. But definitely a possibility and something which will be interesting to see how that one unfolds.

Ali | Startup Explore (18:11)
Now, I know we've talked about some of the existing features. Now, what are future product developments and some of the key features that you are most excited about and users can expect?

Kieran Taylor (18:22)
Yeah, for sure. So if you think about what we have to do to get to the test of detail stage, you already have to have a lot of back and forth with the client to understand, hey, here's the samples which I'm going to make, here's the request I made to the customer, you have some portal for them to upload PDFs, they upload the wrong PDF, you have to go back and forth with them on clients, on emails. So there is a lot of work which gets done before we even get to anything Midship can do, and kind of moving up.

moving up the stack, moving upstream to actually own some of that process as well is really like the most obvious next step. I think downstream from that is once you actually sign off an audit, there is a lot of completeness checks, which in order has to perform. something like, hey, here's all of the financial statements which we've positively, I have to say this, this is all the stuff which we've been shown by the client and we can actually go through and sample and test.

Ali | Startup Explore (18:56)
Mm-hmm.

Kieran Taylor (19:18)
But is there anything which we're missing, which the client hasn't told us, which we can pick up on from some other source? So for example, going through board minutes and saying, was there some transaction which was spoken about in a board meeting six months ago, but wasn't registered in the general ledger? It's kind of downstream of all the test of details. Is there anything you can pick up on the completeness side? And kind of expanding the product out in both directions there is the most obvious projection and definitely where we're spending all of our brain cycles on.

Ali | Startup Explore (19:45)
That's very exciting because talking about the upstream work. So if you can create AI basically that will do PBC tracking process or yeah, like creating those requests, following up and kind of going back and forth with the communication with a client and checking the things that they have received and so on. I feel like that's pretty interesting. Cool.

Kieran Taylor (20:09)
Yeah, there's definitely been some other attempts at this. And I think that some of them are built on, like I said, assumptions which you probably couldn't make 12 months ago, which we can make now based on the state of the underlying technology in LMs. So I think there's some really interesting things we can do there in that space. And yeah, I don't want to give away too much, actually, but I think that's definitely one to watch out for.

Ali | Startup Explore (20:32)
Got it, got it. Okay. Now let me take you back actually now and can you talk about your background, your education, prior work history and what ultimately led you to start building MidChip?

Kieran Taylor (20:43)
Yeah, definitely. in a previous life, far too long ago now, I was a biology undergrad student in Manchester in the UK. Out of that, I came into the grad scheme for Deloitte in London. So that was basically a two year program where you get to try out a bunch of different stuff. Deloitte has many arms, many tentacles that reached into. You kind of have a chance to try out some of those different industries and lots of different project work. So I spent two years there, ended up going back to school.

in New York to do a master's in computer science and I had a change in the career there. And then I was at Amazon and then Instacart for a while afterwards. So I've had like a very varied background, seen a lot of different angles of different industries, including like multiple sides of an audit from both a financial audit side, the kind of building their technology to assist in a financial audit, being audited as a technology provider at Instacart.

and then actually building data access controls in response to some of those things as well. So I've seen from many different perspectives the industry both inside and out. think specifically how we came into Midship, we were previously working together, myself and my co-founders, at a previous YC company, Dashworks. So like I said, AI knowledge assistant, at the end of working there together.

We kind of at a very last minute from an extended deadline applied to YC with an idea which we'd been like noodling on for previous weekends. While expecting it all to get in, it was like very much a last minute thing we threw together. Got an interview request back a couple of days later. Had an interview, got the acceptance back a couple of days later after that. And within the space of a week from applying, we found out that we had gotten and had sent in our application to resign from our current position.

It's a very, very whirlwind how it happened in the end with Midship. We went in on a very different idea as well around data extraction. That's really what led us into that space of what is the kind of bleeding edge of structured data extraction you can get to today. We always had a view of how could we verticalize the solution and where would it be most impactful to actually build out something which is built on top of that very solid structured data extraction. And all of that is to say that is all of the confluence of factors which led

to us going into this and building out the solution with an audit. So it's been like a 10-year journey probably into this space.

Ali | Startup Explore (23:07)
Got it. That's awesome. And speaking of YC experience, can you share your maybe the most challenging moments and at the same time, the biggest learnings from that journey?

Kieran Taylor (23:17)
Yeah, so I think when we started YC, I think some people from the outside view it very much as it's going to be like a didactic teaching experience where on the first day they tell you, hey, today you do X, tomorrow you do Y, next day you do Z. And it is very, very different from that. And I think the biggest value which we got from the whole experience is just going through the motions of being a first-time founder with 200 other companies alongside you doing that. So I mean, in terms of picking the hardest thing,

Every day was like a huge struggle because every single day that you're doing it, there's a new thing which you just didn't even expect to have to do. And you're having to just react in real time to each of those. But I think that just the community which you have at YC, everything which you've been through, someone very close to you has been through before as well, super beneficial and helped us out a lot with, like I said, reacting to each of those things.

Ali | Startup Explore (24:12)
Nice. Now, also speaking about the backgrounds of your co-founders, are they also technical?

Kieran Taylor (24:19)
Yeah, so Max is also an engineer, more focused on front end and Ahil is a designer.

Ali | Startup Explore (24:25)
Got it, okay. Cool, now let's switch gears to a total addressable market. Now I know it's huge, but can you help me to quantify it?

Kieran Taylor (24:35)
Yeah, for sure. So in the US alone, the total spend on audit services is north of $200 billion annually. The big four actually has like a huge outsized portion of that spend. And then the rest is divided between, I think, about 4,000 different firms in the US. So I don't have the number for the worldwide market off the top of my head. But even just the 200 billion US market alone is enough of juicy target for many people to go after.

Ali | Startup Explore (24:59)
yeah, yeah.

That's for sure. And now speaking about the target customers, are you only targeting external audit firms or you also think the internal audit teams within the companies can benefit from Meshup?

Kieran Taylor (25:16)
Yeah, definitely ICP within external audit today just because from a sales motion is much easier to tell the story of what we're doing from test of details. And I think the reason behind that is that the role of an internal auditor looks very different based on which industry they're working in, which company in particular, and it's just the actual, the functional role of what they're doing varies a lot. Whereas for an external auditor is much more because it's literally a regulated industry, the actual, the day by day actions are much more.

non-unique, I don't know if I'm missing the word for that there. But it's much easier to tell that story of what we're going to actually use the software for. I think within internal audit, there's definitely use cases. And we are inactive conversations with people to explore those as well. But it's just much easier to grab for, think, external audits.

Ali | Startup Explore (25:48)
Ha

Mm-hmm.

I'm always curious to ask this question, but how did you land your first customer? What steps did you take and how easy or difficult that journey was?

Kieran Taylor (26:13)
Okay, firstly, very difficult. Like I said, doing everything as a first time founder is difficult. But the most fruitful thing for us was just having basically cold LinkedIn outreach and just asking for genuine feedback from people working in industry. And it wasn't even like a disingenuous, hey, can you guys give us feedback on your experience of working with an audit? Like we really did need to learn from them because we knew very little at the start. I think once you start off with that,

mindset of I'm just looking for people to learn from, looking for genuine connections within the industry. It leads you both from like, hey, this is the product and this is the word direction we should take that in. And also naturally starts to build the network for people that you can actually sell into as well. And we're really hoping to get to the point where we learn from people enough that we build something that is so useful that it kind of gets drawn out of us from like, hey, someone else is coming and requesting a demo from us, as opposed to us having to push anything from a cold outbound.

So yeah, first customer very much from that process. But like I said, not something which either came naturally to us or was super easy to start with.

Ali | Startup Explore (27:17)
Got it. I'm curious to know how long that process took, the sales cycle, the first one.

Kieran Taylor (27:24)
This would have been, I think from the time we first started talking to them, it was actually about six weeks start to finish. And it was someone who was willing to move very nimbly as well. So that is like an extremely short cycle compared to what we usually expect. But I'm making sure I'm getting that right, but very, very close to six weeks, if not exactly that.

Ali | Startup Explore (27:35)
Okay.

Yeah, it's still good. mean, it's I would consider that pretty short sales cycle. So good for you guys. Nice. OK, now let's talk about competition. Is there any similar tools in the market today? And if yes, what sets you apart?

Kieran Taylor (27:50)
Yeah. Yes.

Yeah, so the competitor who gets brought up all the time when speak to people has by far the most penetration is Datasnipper, who are a team based out of Amsterdam. they are, originally the tool was built from like a very simple Excel add-in, which is just, you're working inside of Excel and you've already completed the work paper and you need to link back to one of the supporting documents, can you do that by basically snipping out a value from the tool?

sorry, using the tool from the document. It then creates a persistent link. Then when someone is going to go through and review that work paper, they can see and compare the values. So that was very much from like, hey, can we have the most bare bones Excel add-in which we can support that functionality? I think after they found this really strong use case within audit and expanded out the product, they've kind of built towards how can we add more and more advanced functionality on top of that? So instead of having to manually go through and do the snipping, can we add features which is automate that process?

And so I think the thing which really sets us apart from then is kind of the base foundations of how we built the product. So they are starting from, a simple Excel add-in, and we're going to add functionality on top of that. We're starting with LLM agents at our core automating these tasks and then exposing that stuff through Excel for your review. And that really opens up different possibilities in terms of what you can achieve. And it's a very different product experience actually using them as well. But I guess the final end product, is...

a work paper with the evidence linked out for each of the procedures. Looks very similar from an endpoint, but how you get there is a totally different story.

Ali | Startup Explore (29:40)
Yeah, that sounds like pretty different architecture decision from the start, right? Got it. Now looking into the future, let's say into three, five years, where do you see Midship and how do you see the industry to shape?

Kieran Taylor (29:45)
Yes.

Yeah, this is something which we think about a lot as well in terms of the, I mean, the core assumption within auditing today is that you have to perform sampling during an audit procedure just because of the limitations around how much time you can dedicate to a single client and a single engagement. So as AI tooling becomes more pervasive, as Midship solves more and more of these problems, that is definitely going to change in terms of the assumptions around

Like how many samples can you perform during a given engagement? we kind of the terminal state of that is that can get to a place where it's not even sampling anymore. And you're looking through every single transaction which has happened in the past year. And that is really what we're thinking towards from a product perspective as well is as this changes from both expectations from auditors and then what is possible from the product side. How do you build something where if that base assumption goes away of doing the sampling procedure, how do you still support something which I guess.

is able to do that complete view of a client's financials.

Ali | Startup Explore (30:59)
That sounds like it's going to be kind of a constant audit of all of the transactions at all of the times, right? Wow. Okay.

Kieran Taylor (31:04)
Right, right, exactly. Yeah.

As you think about AI agents becoming like truly 24 seven autonomous beings that opens up a lot of possibilities from both an audit perspective, IT security controls perspective. So really trying to hit that moving target of how long is it going to take us to get there and what are the possibilities open when we do get there is what we think from a long time perspective.

Ali | Startup Explore (31:26)
Okay. Now as you scale and grow the business, you would need to hire more folks, more people. I'm curious to hear if you kind of look for specific qualities in candidates and what those qualities would be.

Kieran Taylor (31:40)
Yeah, think we have from... It's been very interesting actually from a hiring perspective from our YC cohort, just because the decision to start hiring is actually going to push back further and further. And I think a lot of that is the effect of AI tooling, which kind of amplifies the amount of work which the founding team can get done. So I think in terms of like our view of hiring, we're still waiting for the point when it gets to...

So painful that we really need people to come on board to help us out with that I think in terms of the the qualities that are required now in the age of AI tooling as well is very different from what was probably required Previously, so I think from like a software engineer in particular The most obvious way to go through an interview someone previously was get on to leak code give them a problem See if someone can knock out leak code hard. I think it's all of that stuff that lower level capabilities

much less important now than your ability to actually focus on the business problem, collaborate with the team in person, and kind of move very quickly and adapt to the changing environment of a startup. I think that that's probably going to affect, I haven't thought about as much for other roles outside of engineering, but all of that is very top of mind as thinking through hiring, which is how does someone leverage the new generation of tooling which we have available to be an effective employee? I think that's probably the biggest mindset change from a hiring perspective.

Ali | Startup Explore (33:08)
me flip the question now for future potential candidates out there. Why should they join Mishap? What would you tell them?

Kieran Taylor (33:16)
Yeah, I think the biggest thing is that we have a super exciting opportunity to bring innovation to an industry which, like you said, has been both reticent and slow to integrate new technology over the last X number of years. And I think that just if you are interested in high impact, high leverage role, where there is really a chance to build new interactions with this emerging technology to

effects and actually like engineer solutions, which truly are zero to one solutions, which haven't actually been created before. That is the biggest thing which we can offer from like, what is the problem which we're going after? I think probably for myself as a candidate, when I was looking to leave somewhere like Amazon, that was really what I was interested in looking for is like, what are these really high leverage problems, which weren't possible to tackle a year ago? I think that's really what we're focusing on today and what I would want someone to join us to do.

Ali | Startup Explore (34:15)
Awesome, that's exciting. Now, I know we've talked a lot about the product company, the future vision. Let me ask you a question outside of work. What hobbies and activities do you have, do you like to do?

Kieran Taylor (34:29)
Yeah, I've always been a huge tennis player. So yeah, since very young playing with grass courts back in the UK, that's been like my main passion. I'm also hugely into photography. I just picked up a new Fuji Voigtlander 23mm 1.2 for any of the camera nerds who watching. Yeah, and really just being able to go to cool places which look cool to take photos of them is probably my biggest thing outside of tennis.

Ali | Startup Explore (34:47)
Ha ha ha.

That's awesome. I'm not into cameras, but that sounded pretty expensive camera. I know they're pretty pricey.

Kieran Taylor (35:02)
Yeah, don't Google the price of that one, be sad to look at.

Ali | Startup Explore (35:04)
Got it, okay.

Well, thanks, Kieran. Now, Kieran, my last question is, is there anything else about the product team and any features that you would want to share with the audience that we haven't touched on?

Kieran Taylor (35:19)
that's a very good question.

I feel like, yeah, I feel like I've been talking a lot. It's tough to think tough mind. mean...

Ali | Startup Explore (35:25)
It can be any

word of advice either.

Kieran Taylor (35:31)
Yeah, think that one of the things which we have from a design perspective that's very cool is that I think one of the main things with building AI products today is thinking about how do you integrate human checks into each of the processes that you're building out for. There's a lot of companies who are building offerings for how can you do human and loop reviews for every time an AI agent takes an action. One of the cool things with Midchip

that I've been thinking about is that because of the regulation within the audit industry and the fact that you have to have multiple people going through and reviewing every step which is taken during an audit, we kind of have this process of human in the loop baked in to the actual workflow which we're modeling out. So we have some really cool opportunities, I think, to break a lot of ground in how do you actually expose the work that an AI agent is producing in such a way that it's very easy for a human to...

human to review. I think I just think constantly around the user experience elements that are going to come out in the next few years is going to be just a total game changer in terms of anything that we've seen before. I think a lot, like when Chat Root first came out, the only mode of interaction which everyone had and everyone was familiar with was a chat box and a chat interface. And actually thinking through the possibilities from a UX perspective of...

how that changes of interactions with AI agents is something which we spend a lot of time on thinking about. I can't remember that actually related to your original question, by the way, but I was just trying to think through stuff which I think about when I'm going to bed at night.

Ali | Startup Explore (37:06)
Hahaha.

Yeah, it definitely did. It definitely did. So it's very exciting milestone and yeah, looking forward to see the execution and success. And thank you so much, Kieran, for this interview. I've enjoyed it a lot and hopefully we will meet again sometime soon and you will have even bigger success stories to talk about.

Kieran Taylor (37:30)
Yeah, absolutely. Happy to come back on and share those.