Appendix: Raw Dictation Interview Summaries

Date: 18.10.2025 (2025-10-18)

Appendix: Raw Dictation Interview Summaries (18.10.2025) I conducted a lot of interviews in the last few days and I'm going to tell you now the main insights in no particular order and it would be great if you could structure them and basically create a summary across all interviews. The more often I mention something, the more often it was mentioned by the people that also said it. So that's also kind of important to represent. So if I have a similar point five times, maybe you can find a way to say that this was mentioned often versus something different that was only mentioned once. In general, in accounting and bookkeeping, there are a lot of companies doing stuff. Pre-accounting is a threat because you have to prepare all payments and invoices for the DATEV export. DATEV has a bad API to get data out of it, so like analytics and end of month and so on is difficult. Rory would like to have a working student that can do all the work that the current working student is doing and reduce the time to 30 minutes for him, so the AI does everything and Rory only has to invest 30 minutes to review and so on. That's what he would pay for. DATEV in Germany has a monopoly situation, they're very cheap. In general, finance and accounting people work a lot in Excel from Microsoft. A tool he would love to have is AGICAP, you spell it A-G-I-C-A-P, plus the capability of uploading a forecast to compare actuals with budgets. He thinks the finance SaaS industry is saturated, we should build a digital finance manager with the goal to replace a working student. At Yababa, his former company of Rory, they have a data warehouse with a web shop and like the automation processes was very expensive, like that could be an interesting field. Sending invoices to customers is a problem because there are not a lot of tools for SMEs. Service processing could be an interesting field via Slack, so using the user interface where the users already are not introducing new user interfaces. He cannot use GGBT due to data privacy reasons for certain topics. At Schachermeyer they started, now this is the second interview by the way, at Schachermeyer they started doing automation six months ago. They're also having a few in the pipeline and they have one person doing that. Currently they are not automating platform cross, but only within platforms with Office and Power Automate. Then third interview was with Dimash. They have a middleware via SFTP to sync the accounting tool with .DEV. They have a big gap for legal documents. They tested the solution Think, which has great value proposition, but it doesn't really work. But they would buy that thing. In accounting they're currently working with N8N, a process where the invoice is booked to .DEV, they check database for expense line, they add it to the cost center, and so on. That's what he would like to have. They don't have the process yet, but they are thinking about working with N8N and build that process basically. Sorry. And N8N has a .DEV integration. The categorization of expenses is a big topic, a manual topic. And then they have to do 500 invoices that they need to put into their system. Each takes a couple of minutes, so that's a big task. And also he would say onboarding of new employees and offboarding could be an interesting field. He thinks doing something with .DEV and integrating with them could be valuable. They are using an outsourced service for accounting. He would like to have a tool where he can check what is the, what did we pay for a certain supplier or something, and then he can query that. So almost like a check, but he do get like, let's say, the cost for a certain supplier in the last month or something. So he would like to have a tool where he can check what is the cost for a certain supplier or something, and then he can query that. Next interview is with Stefano Sorrentino, IT admin. He started automating processes, but only three. He has a couple more he wants to do. They're using make.com. He's not super experienced with LLMs in general. He's not that technical, I think. Everyone is automating their own stuff. There's no central automation role yet, but he wants to become that. He thinks the onboarding and offboarding process could be interesting. Marketing department does a lot of automations. It's not as complex. Then Narendra from Grammarly. He isn't aware of any automations within Grammarly. Then the interview of Harsh with an engineering manager. There seems to be a lot of approval processes, which could be interesting. Then second interview with Rory. They have to manually match the invoice to the transaction. That's once per month, four hours. They also use an accounting company, an external one, who does the accounting and taxes and so on. The working student checks the invoice if all the data is correct in the system, and then Rory has to double check. He would like to have some automatic approval if, for example, an invoice matches the purchase order. Then he would like to include budgeting and forecasting so that he can match for the financing invoices with budgeting. He would like to have AI diagnosis. For example, why is there a gap between actuals and budget? It could be, for example, that a supplier charged more or something. Investor reporting is a big topic for him. He spends eight hours per month on it, so that would be 10k per year that he could save in terms of salary costs. Also, expense claims could be interesting. Making sure that people are uploading the correct invoice. Then a lot of financials have two-factor authentication. Some even have three-factor. That's something to consider. Then interview with Eloisa. They create invoices to their customers manually in the Microsoft suite. She would like to automate processes, but she doesn't have the time to do it. Like one process could be, for example, the invoice process again. Double payment is a big problem for them. They're also using data, which is a problem. Then investor reporting is also a big effort for her, and they don't have a travel expense software yet. Now interview with Jan. He would be willing to pay 1-2k per month if that frees up time for his CFO. He wants a development partnership. He is skeptical with AI due to hallucinations. Then next interview with Marco Littel. His biggest challenge is making insets into the processes transparent. Because RPA traditionally is running on machines so you never know what's happening and so on. He spends a lot of time on requirement gathering with people. He works mainly with the Dutch government who are not open to LLMs at the moment. They want to have everything locally hosted and they use a lot of Power Automate. Then we have the written feedback of Anton Hosank. He says more code is better than no code. It's missing. The field is crowded. He also thinks that the biggest challenge is understanding which process to automate and how. Then interview with Florian Wirth. They have make.com with which they connected two CRMs. One use case is taking a in German Bauanordnung and check it with some requirements. There's an automatic warning on certain topics. They're using N8n for that but it's not really adopted yet. They have an AI expert group within the company. They also built something between HubSpot and FOSS manager with make.com but they killed it again because it got way too complex with 200 nodes. They wanted to automate the procurement but they decided not to do it because it's a shadow process and better to optimize within SAP. They use Microsoft Fabric and Power Automate. They often have incomplete APIs with which they struggle. At his former company, Planted, they had a lot of invoices and approvals which was a lot of manual work. Then the next interview with Renan Duarte. He says there's tons to automate in accounting with what the external accountants are doing. Then matching of invoices is a nightmare. He could do an intro to a guy who has an accounting software. The intro part you don't need to consider. That's not a big deal. He thinks the hotel industry could be interesting. There are many old platforms and large chains. Shipping logistics could be interesting. Then an interview with David Frederiksen. He says your IPAF is very expensive. Microsoft Power Automate is becoming very popular. It has also RPA capabilities with Microsoft Power Automate is becoming very popular. It has also RPA capabilities, but at the moment they're poor. He says most processes get complex. The Sim AI demo was not really useful to him. He doesn't see a value for him as an RPA developer with it. Then an interview with Pablo del Ser. He sees that the interest in RPA is going down because he feels the companies who want it already have it and it's super expensive. Due to the AI wave, a lot of CTOs are waiting for projects to not with RPA, but wait how AI can solve it. He's mainly working with insurance, logistic utilities. He also says the difficulty is that people don't understand processes and you have to do a lot of requirements. Gathering and processes always break in unexpected ways. It's frustrating for non-technical people, so quality has to be high. Next one, interview with Martin Ruby. Banks in general do a lot of automations. Most robots take data, do some simple data processing, and put it into another system, mostly the ERP. He says approval flows are very important for vacation, onboarding, offboarding, and so on, so that should be something we think about. Then interview with Monish Revuri. He had to get Slack admin access to authenticate the account. In general, credentials of accounts and setting up basic accounts for automation is a challenge, or has to be thought about. He integrated a third-party tool on Salesforce. If invoice is generated, there's an automated invoice number that is created in Salesforce. Then he pushed that into a Google Sheet to later match it with the Conto transactions. So he matched with an Excel. Could be something to think about how to solve that better. Then interview with Paul Brandstetter. Actually, he didn't say anything relevant at this point. Then interview with Alexander Lichtmann. For him, most processes are pull the data from a finance tool to Excel, then to PowerPoint, to email, to show to some stakeholders. He thinks billing, purchase order, steal approval could be interesting areas. Then interview with Philipp Arbeiter. He says know your customer could be an interesting topic because it's a lot about collecting documents and so on. Then interview with Paul Anstetter. When they have to get data, they make a data request via an form in Excel where they define what people should send them. And then the client people export stuff from their systems, mainly ERPs probably, and send it back. Some systems, some clients have decentralized systems, some have centralized systems. Sometimes they have to categorize expenses to then run analysis. They merge all the excellent templates, they get them to one big excellent and run some analysis there. And they really get it, they get very granular data. So for example, not only invoices, but really purchase orders. So every kind of item is one line. It's more detailed than invoices, invoices are often monthly. All the software tools they're using are not really good. They have a lot of internal tools, but there are probably better tools in the market. They benchmark supplier purchase or procurement categories with other companies and also number of employees to see if the client is above benchmarks or not, to see where to focus on. They have a tool that checks invoices and if approvals are needed or not, which they sell to customers and some of them are using it. Then they have a tool that calculates where, like it's a supply chain tool and it calculates where warehouses should be built and where bottlenecks are. It's very complex and then when they actually want to build a warehouse, they have to do some manual process to really see how it would affect the supply chain. They also have a tool contract screener, which summarizes contract and you can query like whatever is in the contract. Same for invoice, it's an invoice screener app. Interpersonal topics are often tricky because people think they're replaced, so that's something to manage. If they want to lay off people, they send forms to the managers for them to describe what the employees are doing, so they can then pick who to fire and where to save costs. If they want to get new suppliers, they have a tender and then they negotiate with the suppliers and they have their own people who are pro-negotiators. In general, the project is often like supply chain, which Inverto does, then BCG comes for sales, ops. And then they also have SG and ASUM, more like the general administrative stuff. These are kind of the big teams normally. The big problem is when they leave, the data is still fragmented again and it's very difficult to have a continuous high level, like high quality process. In logistics and supply chain, like tracking and so on is also super important. A lot of data is very distributed, so it's very difficult to get life insights.

So now interview with Nikolas Müller from McKinsey. He thinks the low-code will be replaced by either no-code or coding solutions. Copilot Studio has a lot of enterprise-grade integrations but often APIs are not good enough and not comprehensive. But Copilot seems to have a high quality. He thinks there will always be a lot of software and interoperability layers needed to automate across functions. He has quite some experience building agents for clients and he says everything every agent that has to call more than 10 tools is complex and difficult to get into production. It's easy to get to 80% but not 200% quality. There's a big trend going back to deterministic tools. Patterns that he sees in different projects are payment, reconciliation, data unification, like building a centralized data pool or system, software development like definition of requirements and AI-based coding, and then in finance monthly budgeting, early warning systems, auditing and compliance, and building financial models. In day-to-day operations more like meeting summaries and to-do assignments. Then they work a lot in Excel so that a lot of the databases and processes they build within the project team is lost when McKinsey leaves. SaaS revenue for McKinsey is a hot topic at the moment but they're not really able to do it because they're not a product company. He thinks most of the McKinsey offering could be, let's say 40 to 60 percent of the McKinsey offering could be easily automated, so go-to-market, acceleration, due diligence for PE, and he also thinks execution will be the value. That will come later up again in an interview later from another McKinsey consultant. They think Palantir could take the value and in a competitive situation with Palantir. The problem with normal data layers is that they don't understand the data but Palantir does. Procurement analysis is a standard process which has high value. Palantir, so the procurement analysis was a standard process within Palantir but they're expensive and there's a strong vendor lock-in. Companies often buy it for one specific use case and then already think about how to getting out of it again. I think that vendor lock-in is an important point we should highlight. So often people are just using Excel for data although they would have a system. One of the biggest questions for McKinsey is how do they build a comprehensive AI setup within a company because nobody has a plan at the moment. They often build a deterministic n8n workflow and then it's called by an LLM so it's basically a tool for an LLM. For LLMs cost efficiency is most important because like the big models often get expensive quickly and he thinks that most companies that do host LLMs locally do it because of the cost not because of data privacy. Then interview with Varun Mundra. Companies take siloed decisions because departments automate only within their departments. They don't communicate a lot department across and it's impossible to kick out legacy software and also decision-making companies is often local for example an Indian HR buys the different solution and global company for to have a local solution for example. Then interview with Alvaro Aubele. They use power automate it's not super user-friendly and a lot of things are blocked by the IT team due to authorization. He sees a lot of potential in the product team. ERP integrations to build them are expensive it's too high of a cost to generate ROI it often takes three to four months to build and they often work with CSVs files instead of integrations. Everyone has different ERPs. Then interview with Sebastian Torres. He says build a workflow tool for cursor. Then interview with Florian Christoph. Their finance has a USB key in order to log in so it's like TBD how this can be solved. Then he says in reporting a lot of their processes kind of getting numbers from a system and putting it into a Google Sheet or Excel so it's a lot of copy-paste. Data is very fragmented because they for example have different tools for different countries due to price and taxes in different countries. I think that's also a good thing that like like an important thing that local entities often have local solutions because of like local policies and so on. Then Alexander Anserud interview. They built a workflow basically a UI use tool as well. They had several paying customers but it was not good enough in terms of quality. They built like for the computer like low latency. It was basically building a small macro. Building macros on your computer based on AI. Next one Paulios Latounas. They use Microsoft environment. They have three legacy systems. It's a Estonian or Lithuanian railway company. They have like a few bots in finance taxes marketing mainly for two legacy applications that are not really supported anymore. Speed of RPA is a problem. He also uses Power Automate. I think that's also one pattern that like people are now using Power Automate a lot and more and more. Then interview with Anthony May from Kearney. They are doing RPA as well in the project. He sees the pain point with legacy systems, especially for airlines, that it's massive the problem. The buying circle is quite complex in these big players. That's a very interesting insight. He says that a lot of the legacy systems in these big companies, in departments, are like specialized decision-making engines, how he calls it. For example, the planning of a crew schedule in an airline. But the problem is that the users are not smart enough to use it in the best way, because it's very complicated. It's not that they're dumb, but it's just very complicated. It's just a very, very difficult task. What he thinks could be super valuable would be using the systems better. But then an API is not enough. You would have to, I don't know, add some intelligence to the API, basically. He's building his own internal project for crew planning, and he currently has not a solution yet how to get data out of systems and push it back in again. At Kearney, they often say they have to improve data situation, but often the ROI is not really clear of what better data really brings. So it's very important to always talk from the ROI perspective. And also Kearney wants to sell SAS. Then interview with Gunther Baderer. He says procurement has a huge potential. In construction, it's very fragmented. It's basically per construction site, and there's not really an auditing possible, because it's very manual on paper and just not digital. And the leaders of construction, they hide the delay as long as possible and also fake data until it's too late. He thinks it would be super valuable to find out basically which construction places are the most critical. Then another interview with Rory Ribeiro. He thinks that Relay, for example, the accounting software, they will not be able to get really large customers if they can get someone than like one department so that it can add the brand to the website but not a whole kind of holding company. So they work with external accountants and saving their money could be super interesting business case because Bara pays 2k per month for the accountants at his former company it was 10k. If there you can save something it's a good business case. He would switch to a new solution if it could either replace that or help with investor reporting. The external accountants are doing invoice matching and categorization and also taxes. He says a very interesting process would be the full reconciliation of the general ledger like validation of invoices in both directions suppliers and clients, categorization, purchase order comparison, supplier contract comparison, and the second variable would be financial reporting especially with actual sources budget and kind of analysis why there are gaps. Investor reporting takes him two days per month. Accruals are important to consider could be also supported with AI and we could theoretically sell to accountants as well. Then interview with David Dürr, consultant. He says a lot of times the problem is that people are not trained on tools especially new tools so that's why often the fallback is Excel or they're using the tool not really well. Data is often siloed and it's a big problem because data is not kind of exchanged between departments. He thinks industry-wise telecommunication and mobility is relatively good. Data is often in the cloud whereas like banks healthcare like the regulated sector are way less advanced. He says for example a large German railway company has a very fragmented IT landscape with more than 150 applications most of them are not connected. Building integrations is super expensive and also mirroring access rights and data privacy across tools is a big problem when integrating them. The lock-in effect in SMEs is a big one and general companies and ITs are scared of locking in so that's something to consider also to highlight because it was mentioned a couple of times already. In general his consultancy is not really incentivized to push for to make legacy software better because they want to introduce new software. That's what they're paid for basically. He says often in-house IT is not really good. Then interview with ICOT-MOS-LUOGLU. Sometimes your IPath has problems to control systems because they are layered systems and so it's hard for them to catch the right elements. The hardest part about RPA is working with the people and understand what and how to automate. He thinks traditional RPA will fail because pages often change and there's not a lot of stability. He thinks we should build something that integrates with access rights and data privacy something like that integrates with Excel for example via an VBA script. Or the other way around take data from Excel and push it to a legacy system so making that integrations very simple. Then interview with Andreas Beyer. Travel expenses is interesting it's very very manual at the moment. They have to fill an Excel file send it to an assistant who checks it with the travel plan then signs it forwards it to HR and then it's paid out and accounted for. Then also Bellassungsanzeigen in German of suppliers is basically interesting so if the supplier is overcharged that they have to pay money back. They have a manual process to compare like an invoice with the SAP entry. They're working with EDI connections to suppliers which is like this traditional format and they have like a development partnership with a company called Verkist that does email to ERP entry so for orders. Then interview with Silke Mitternacht. Yeah she's a very good person and she's a very good person and she's a very good Then interview with Silke Mitternacht. Yeah she's in a mid-sized city with 17 000 like people living there. They have like 80 people in the admin department or company. There are a lot of processes online already but also they have a lot of paper-based stuff in the administrative side. Then interview with Markus Borsch. They only have one site. He doesn't know the exact process for invoices but it's very manual. There are multiple approval steps and also controlling is checking the possibility. They don't have a lot of automation yet. There's a person for example doing your travel booking and he has also some processes that he could automate but they seem rather complicated. Then interview with Alex Silea, RPA developer. He says that the idea that we had sounds really big because RPA tools are really good because RPA tools are currently doing automation serially so that could be a big USP. Then interview with RPA developer August Agelich. Blueprint has a better license structure than UiPath so it's a bit cheaper. Then interview with RPA developer. He's not sure if AI will be adopted within RPA. He thinks AI will be more used to accelerate the development and testing of RPA so kind of the co-pilot case. Testing is very difficult so some of them said don't test but just roll it out and see what happens. Then interview with Lukas Arbeiter, controller. There are 30 to 40 people in the controlling department of the group, of the holding. They have a lot of software but it's centered so they're using like it's a centralized software system. They have like the Microsoft suite mainly and some finance tools. He does mainly do forecasts versus actuals. He reviews planning of departments and compares it with numbers of the prior year. The planning is made locally for example the US subsidy does it in their department. They get the data sent in Excel templates every at a certain day of the month. So it's a standard process and there's not really a problem in getting these Excel templates. He's not aware that they're doing automation at the moment. Processes that he would automate would be transfer one Excel to another and a kind of an easy process to understand what to copy and in which format. And then also some for reporting some standard interpretations that you always have to do like for example the revenue development or something. And now a kind of a general note from myself like a pattern that I see is the controlling they don't really have insight into like the operational processes. So they're very detached from the rest of the company. So I guess talking to controllers and accountants and so on it's mainly they're mainly not their area but not really what happens in the rest of the company. Then I talked to John Huck. He's working at a real estate company a developing developer company. They have an approval process where they transfer data to .dev. There's a lot of manual process because the API doesn't really work. He says if we do something that we should talk again. And he also spends a lot of time with tenders to get basically suppliers. And often like one of the biggest problems for him is keeping the overview in projects because there's so many things that affect each other. So for example what happens is that he forgets some clause in a contract and then it has negative consequences. So that would be great to automate like build a tool there. Then interview with Helena Carsten-Hirsch, who is a former McKinsey consultant. They looked at PE area, like due diligence and market analysis. Then kind of building ahead of strategy in mid-market or supporting them for mid-market firms. And then basically all strategic sales for complex B2B companies, kind of like helping them tailor the pitch to the company they're selling to and that got most pull. She's as procurement, she knows a lot of people doing that. So it's a hot topic, which is a good signal, I would say. She sees a lot of potential in building AI-first consultancy. She thinks consultancy could benefit from a tailored ChatGPT, something like Harvey for consulting. She says that McKinsey is already selling their own SaaS, while I heard that from others as well, like she's in the bit as if it's more common than the other set. But in general, I think she was a bit like, she was not completely honest. She was a bit in the selling mode, I guess. So let's see, but still important to document. She thinks that there will be more Palantir in the future, so more tech companies to do data analytics. But companies do not want to be locked in with someone like Palantir. And like one company, like one consulting specific input was people are also getting consultants not only for the content work, but also for like their justification to do drastic actions to profit from the partners know-how or the firm's know-how from competition as well. And also as a safety compared to the board. She thinks execution is getting more important than strategy. That's interesting because it's also what the other guy said, but he was also from McKinsey. So maybe it's like an internal kind of hypothesis within the firm and they're just repeating it. Then interview with Niharika's sister. A big problem with local tools is that most of them are not HIPAA or GDPR compliant. Traditional RPA often processes are stopped when humans have to intervene. And only when human approves, it continues. So there's like a lot of kind of blockage of throughput. Good reporting features when processes fail is important so that you get a screenshot or a video, even what happens so you can fix it quickly or react. Then interview with Marcelo Cruz. He says that a lot of companies have internal apps that are not exposed to the web and you need basically VPN tunnel to access it. UiPath now is exposing workflows as MCP server so that agents can use them. UiPath now is exposing workflows as MCP server so that agents can use them. UiPath is super expensive. You quickly have like 550k of license costs per year if you want to do some basic automations. UiPath also has some document understanding products that are interesting. It's kind of like standardized apps that help you with workflows. As a developer, he wants to review rather than have automatic self-healing if something fails. But he wants to be quickly able to review.

PART 2