Appendix: Raw Dictation Interview Summaries

Date: 11.11.2025 (2025-11-11)

Appendix: Raw Dictation Interview Summaries (11.11.2025) Samir Jola: In larger companies there are sometimes buying specialists and also purchasing agents and purchasing agent is basically an external company that does the buying for the company and then just forwards and invoices at the end of the month but they take care of all the legal stuff and the admin work and so on. They also do the reconciliation payments and so on. He has to confirm the Wareneingang, so receiving the goods and if it's correct. They have a procurement department, it's in their big company. In general finance runs a lot behind people to get the invoices especially for expenses. In their case their international company and their legal entities have all their own CRM and ERP systems. There's a lot of redundancy. Next interview is with Søren Hornoff. They have a lot of manual work. He's working at a government authority in Germany and they do a lot of manual work and just copy-pasting stuff from left to right. And when they buy software that's important for public companies, they have a procurement process and a tender process where it has to be approved. Next interview is with Finn Killam. He says every company in USA uses QuickBook for accounting. Next interview is with Tobias Norlund. He basically said browser automation is good because you can do it in the background, but that's not possible on Microsoft, only on Linux. There's some complexity in accruals. For example, is something HGB, so Handelsgesetzbuch, conform? And stuff like Bauleistung versus Werkleistung versus Lieferleistung, there's a lot of specifics. They have their own tools, for example LexOffice. It's very important to do accruals for creditors and debitors. So in German, Kreditoren und Debitoren. There are a lot of gaps that happen during the year, but you only realize them at the end of the year, and then you have to think back many months, and often it's very difficult to find out where to get an invoice or why some numbers are how they are. It would be perfect if reconciliation per month being perfect, because then you don't have a lot of effort at the end of the year. Then also VAT has to be done per month, that could also be a value to do that correctly. In general, there's a lack of talent in the tax advisor space, because nobody wants to do it. Reconciliation is a big problem, there's a war of talents. If they would be faster, they could do more customers and thus generate more revenue. Then interview Julian Hüllimann. Legacy systems are everywhere. They do RPA basically never because it's unstable. And they don't do a lot of agents yet. There are some POCs but nothing really in production. The next interview John Marco Brenn. He says the portfolio companies have a lot of mess with accounting because he's working at a VC. The next interview Rainer Schwarz who is a fractional CFO. He says UK is saturated. There are some tools that can get data from data and run analytics on top of them. He says on the revenue side there's no tool that manages the revenue value chain. It's very fragmented so you have HubSpot as a CRM, you have ChargeP as a billing tool. That could be a big potential. He says we should not try to replace .dev and if you build something it has to be compatible with .dev. In UK QuickBooks is very good. Check for fraud could be interesting and check for abandoned credit cards. Tax advisors also do accounting tax and payroll in .dev. They have a very high sick leave rate and high fluctuation and there are no new talents for tax advisors. 20 to 50 percent of invoices are missing for expenses. Tools have reminders but people are not really reacting to it. In many tax advisors only the partner is good, the rest of the team is not really good. Tax advisors are very dependent on what data they get from their clients. Then interview with Corbinian Flock. He works at Celonis and they have a data model that consists of objects and events and maps the whole company. They rarely use RPA but they do it via a third-party tool. Then interview Florian Christoph. He says that non-technical people are not good at building workflows because they don't understand what's possible and what not and how to even define a process. Then interview with Thomas Smetana. He says some invoices have to be downloaded from some websites. That could be a good use case for browser agents. They give every team a credit card so it's easier to track and monitor costs. He says MOS and DATEV integration is painful to set up. And it needs a process documentation by the tax advisor. Integrations to bank accounts are important. It's important to check actuals with budget and see why there are differences. But sometimes, so they have a tool which flags discrepancies but it's like too stupid because for example it flags stuff like costs going up because they hire more people. So that flags are basically unnecessary noise so it has to be a smart flagging system. For example when you hire a new employee of course some costs will go up to a certain degree. Same on the customer side if there's a new customer then also some costs will go up. For him it's important to have AI that scales with the company. He says there are three areas in finance accounting controlling FP&A and all three areas have different tools so it's very fragmented which is annoying. One financial home would be great but it also has to grow with the company. Different countries have different formal requirements. For example in the invoice you don't have all the same information like in Germany. Then interview Sebastian Spittler. They are building a PE fund to buy tax advisors make them more efficient. But they don't really know yet what they exactly will optimize. Then Kees Pruim from Racknology. He has his own accounting company where he companies can create reports. He has not seen really good AI tools yet but he's very tech-savvy so tools still seem to not really work. He had a feature idea that basically when they do background checks for companies you can just give the tool the company number or something and then it will escape the internet and get all the information about the company. He says fraud detection especially in the field of approvals is important. For example an employee approving his own expenses. They have a big method AR and AP. He says we should tackle not the companies but rather the accountants and tax advisors. He didn't believe the full focus. Back then I pitched AP, AR and closing and controlling and he said it's too broad like unrealistic that we can do that. So he said we should tackle the accountants and tax advisors. He didn't believe the full focus. Back then I pitched AP, AR and closing and controlling and he said it's too broad like unrealistic that we can make that work. He says AR so accounts receivable are less problematic when it comes to automation because they're more standardized on the AP side like you get different formats and invoices from different countries, different suppliers and so on. He says e-commerce, producing company, export-import companies, businesses with a lot of new customers like automotive companies with not fixed customers have a lot of paperwork so that could be interesting. Also big retail brands. He says e-commerce, producing company, export-import companies, businesses with a lot of new customers He says if you solve 80% it's not that interesting, you have to solve 98% of the invoices. He says we should not pitch no new UI and UX because it's not realistic and also not believable and it's actually good to sell a new UI, kind of an admin dashboard or a steering dashboard. Then interview with Julius Westphal, he's an VC. He says reconciliation is a big problem and ARAP was a hype two and a half years ago. He says for pre-seed 50 to 60k ARR signed is good, for a 4 to 6 million seed we need 300 to 500k ARR. Then interview with Max Müller, he did some automation within his company BITS. He used make.com mainly for copy pasting from one system into the other one or triggering an action in tool one when there's a data change in tool two. The problem is that he's the only one knowing how to automate and that's a problem. They use MOSS but there's still manual steps. Then ARR is very manual for them. Then I talked to Jonas Menesklu from AskUI, that's the name of the company. They use Playwright and they do their cache pages to make the agent more efficient. They also use a visual regression library to compare screenshots, so basically a pixel-based comparison. They mainly work with larger companies. Then interview Kimi Peverinta. He says Finland is very innovative with APIs. Integrations is a big pain to build. They use LLMs to map Excel to other Excel, so transferring the format. He says he would not like to have RPA except it gets better with the self-healing features he got interested. Then interview Simon Furrer. Then interview Simon Furrer. They started automating AP with Odoo, the ERP system. He says it's pretty accurate. So there is a problem with splitting invoices to multiple POs or the other way around. Foreign currencies is important and transfer bank accounts. The Atul landscape is very fragmented and for FP&A he uses Google Sheets. Interview with Jaime Gomez. Yeah, he basically didn't say anything important. Then interview with Mark Vendromet. He said if there are no APIs to try to go to the database, RPAs, last resort option. He says Foundry of Palantir is very good. They have a lot of connectors, pipelining tools, web app tools, dashboarding tools. It's very nice. Then interview with Sacha Fuchs. He says like a lot of the times the ERP doesn't really have the data you need. They do mainly analytics in Excel. They have a lot of fragmented IT landscape in his company because they have many locations, so they have to get a lot of the data manually. Then interview with Juergen Maurer. They do the budgeting first in the subsidiaries and then in the holding company. They don't have AI at the moment. Their accounting is distributed in their subsidiaries, but they start thinking centralizing it in the next year. He says he's very skeptical when it comes to getting the data in a clean format. He says it's too messy. They have an invoicing tool. It's important to have a clear ROI for getting a new tool. The biggest pain point is consolidation of the subsidiaries. It's completely manual. They don't have a lot of insights into the subsidiaries because of that. Then interview with Melanie Richter. They have a tool for automation of AP. They have a separate tool for AR. It's different per country because they're local requirements or things that have to be done and it's better to have local tools sometimes. They don't have a lot of pain with AP. They do accounting in-house. Their integration with bank accounts does not really work. She says reconciliation is done well by many tools but currently they do it manually because the bank account is not connected. They wanted to automate taxes but they found out that there's a lot of edge cases so it didn't really work. They have to do Rückstellungen at the end of the month which is always a pain and they don't have purchase orders. Then interview Thomas Stangl. They have a plugin for Salesforce to remind customers to pay. The accountant has to do a lot of manual work which he thinks could be solved by AI. All invoices are touched manually. He wants to automate the complete accounts payable process. Reconciliation is often erroneous. In general, explainability and transparency is very important. For him it's important to have a clear pricing structure when he buys a new tool. Then the Watt Registration or in German Umsatzsteuervoranmeldung is supported by a tool but very manual. AI could support there. He says trusting AI is a big challenge. He says Dattev is planning to put their software into the browser but it will take many years. He says we should use GDP DU files to get data from Dattev. He says most important is controlling. There's a lot of space for AI because currently not a lot is done there. And also some of the tools are not yet for AI because currently not a lot is done there. And also suggesting KPIs based on accounting data for example generator and ARR. He says reconciliation is also a big topic. Then Johannes Martens, he said that currently we are in the human resources budget, yes. But once competition comes up it will be a race to the bottom in terms of pricing. Then interview Alexander Illichmann. He says invoicing is a pain, often his data is wrong so his analytics are always a bit limited. They have a big challenge because processes are constantly changing so they always have to adapt their tools and so on. They outsource a lot of the tasks to temporary workers which have low quality and high turnover and make the same mistakes over and over again. But it's always a balance between fixing everything and just living with the errors and correcting them. Then interview Robert Walgemuth. They are pretty automated because they have a very easy business model and basically tailored SAP very to their needs because they are an SAP advisory firm. They don't have OCR functionality but they have some office managers sitting around who anyways need some work so they do it. Accounts receivables is completely automated. The Rückstellungen for bonuses of people is difficult. There are rules but it's like they built it but it's like hard-coded. In budgeting the extras versus budget comparison is tedious. He checks budget versus extras quite often sometimes even weekly. They have diverse reporting formats for different stakeholders. He didn't come across any ad items he found convincing. Then interview with Shane Fuller. They use a lot of Stripe and have a tool that can import the Stripe information into .dev. He didn't know how they do reconciliation. Stripe has a nice customer portal. They don't have a lot of effort in accounting because they're still relatively small. For him as a founder of an early stage company cash flow management is most important like how long is their runway. He would love to have a cash flow coach that triggers proactively if there's a deviation and gives recommendations what to do. Then interview with Pratmesh Tiagan Unkar. They work on AP and AR as well. They develop procure AI. They build across tech stacks from AWS to Azure so they can deploy it on any cloud. Duplicate payments is a thing to pay attention to and also fraud. But he says it's complex because the more smarter AI gets, the smarter the people get. Cash flow management is interesting what to pay and what to hold off. He says highest value has R to R, O to C, and P to P. Then interview Florian Schabus. They are currently looking for a fractional receipt for auto-support. Then interview with Patrick Bunk. He gives instructions and SOPs to the browser agents to make their performance better. He lets the agent complete a task 200 times and defines success criteria. Then he lets an LLM write an SOP instruction based on the successful agents and that increases the quality of the agent. He says changing a process has cost to change 90% of the values by automating it and later you can still optimize it but that's only 10% of the value because automation itself is the big value. Then interview with Dimas Urgal. He says reconciliation is not that important because it's done by the tax advisor. When this conto is not in the text of the invoice, they call the supplier. For China, there is no IBAN, for example, so you have to transfer the money more complicated. There is no VAT for international transfers. That's also something to consider. A check of the invoice when receiving it and sending back an email if something is missing would be a lot of value. He says we should only do invoice processing and pre-accounting. They already have a PO matching between invoice and PO, but they pay 40k per year. They work with a startup that fills out an excellent and the agent uploads it into dot-dev. He says the co-pilot is not important to have because they're not supplier based. They're a deep tech company, but it could be interesting for retail companies that have more suppliers. For supplier negotiation, average payment date could be used as an argument to renegotiate. He would love to have monthly reports with time to payment from customers. They have a supplier management tool. Then interview with Manoj Baskar. He says data accumulation is a challenge from different countries. He works at a big company Siemens. Supplier management is done by the procurement team. They build stuff internally, but it doesn't really work. If the supplier has an ERP, they send data directly from actually the invoice directly from ERP to ERP. But they have a lot of small suppliers where this is not the case. They have some internal planning tools, but he just uses Excel because it's more flexible. He gets a lot of his information for forecasting and budgeting from the local teams and customers. It's not like publicly available data. It's more like sentiment data. Then interview with Alexander Blank. They are not using AI yet for finance. Their mapping of advisor hours to invoices is quite a lot automated already. Dataflow is not able to detect the time period for a service. So it's difficult to do accruals. They have to do it manually. Sconto management is not really relevant for them. Human in the loop and approval process is important also for automation. They currently pay manually via online banking from Dataflow, however. Then I talked to Leon and Likert, so they optimize purchasing. And they are like per delivery, there are four to five relevant documents like the leave-assigned, the invoice, the purchase order, and so on. They get the data from the original document and put it into the ERP system. They also do renegotiation. They work a lot with CSV files and SFTP server. Then Nikolas Padon. They also do purchasing. They also take the, for orders, they take the original document, not the ERP file. Then they do some analytics based on it. And they get invoices and orders from the billing address. For negotiation, they do deep research and format it for the person to process. One big focus of them is deep contract knowledge where they, for example, get punishments out of contracts, which is still limited. It's not working super well yet. One value of VCs could be selling to the portfolio companies, that's what Daria Gelas said. Then Claudio Martai, they do accounts receivable via Google Form. Then the assignment of who has to approve something is manual. They have a lot of spam in their inbox, newsletters and so on. This content is not relevant for them. Running behind invoices is a big problem, especially for the ones where somebody paid with a credit card. And mapping forecasts to extras is very annoying. Timon Oberholtz, he said in Large Corporates, they want to do AI but don't understand it and authentication is a big challenge. At the moment a lot of the big companies want to build in-house. Data privacy is of course very important for them. Then meeting with Marco Borenschlägel from Strabag. They do quite a lot with AI already. They have a product for supply management. They currently also work on automatic invoice detection internally. And they have a very chaotic invoice to purchase other relationships. Then interview with Nils Schneider. He says pre-accounting is working quite well with MOS. But Kostenstellenzuahnung is not working well. It also changed a lot at his company. The biggest effort was adapting the backend in his accounting tools to the new structure. So in Spendesk, Excel, Dataf, everywhere. In Dataf it's even with the tax advisor. He thinks the big challenges are in controlling. Who spent for what? And having a financial overview of the company. It's a lot of manual work. They build scripts in Excel to run analysis. They are producing. So there is a big difference between the accounting numbers and cash management. They build a script to transfer one into the other. Then also transferring controlling stuff into PowerPoints to present and discuss. It's important visualization basically. Abgrenzungen was very bad. The process was very manual. Often suppliers do not send invoices on time. He sees the biggest value in cash flow, planning and controlling. It's not good enough yet in many tools. Then in an interview with Paulina Zapotoksna she said AP people are very much under pressure and a lot of them burn out. They were even scanning invoices at her pharma company, comparing purchase orders and approvals to invoices, but also getting the approval in the first place was a big pain. Often there are multiple invoices per purchase order, especially for services that are across many months. She has not seen discounts in the UK, something similar to Sconto. One downside companies see with company credit cards is that people start spending more. A lot of companies have a lot of different tools and data is very fragmented. Fabian Habler interview, their accounting is very manual and on paper, so he carries a folder to the accounting department. Then Pascal Muster, big problem is that the incoming goods and the invoice are not matching. That sometimes blocks the supplier and they don't get the goods they need. They do accounting on paper, they have to start with digitalization, he's looking into OCR. They have two accountants. For him, most important is the Kredit-Toren-Workflow and Waren-Eingang. Also, procurement has a lot of potential. Sconto is very important. For example, express deliveries are very expensive if they miss something. Then Nikolaus Christe on the procurement side, he says benchmarks are bullshit. They are very difficult to compare between companies, but they have playbooks for each domain, for each industry where they can say, okay, typically a customer of that size in that industry spends following amount on something. He says alone, being able to tell a company that their spend is too high is already value. Then Markus Baumgartner, he looks at AI players actively. Sconto is covered by the SAP system. They have a global company. They have an accounting manual, which would have to be understood by the AI. Anomaly detection is a big topic for him. For example, when a company has an anomaly detection and suddenly one payment is much higher than the months before, he wants to really automate bookings below a certain threshold so nobody has to look at them anymore. He's introducing a reporting tool for budgeting and analytics. One example for an anomaly would be a change of bank account number. Accounts receivable is very difficult because there's a lot of text complexity. He tried it but he failed. Then interview Andre Schneider. He says putting invoices into data and doing reconciliation is super painful. He wants to have an AI agent that does the pre-counting. Accruals and reconciliation is super important. Then Andreas Lichtler. He played around with AI and make.com on himself, but he didn't really get things to work sustainably. There was a lot of maintenance to do, and he even annotated 200 invoices manually. They're currently working on supply negotiation. They have a fully automated process for AR because it's very standardized. For FP&A, he thinks AI is not good enough. Then Shaheen Dusty. They have contracted ARR and live ARR. These are two different things, and it's always a discussion what to show and look at between sales management board investors. It's a lot about relevant KPIs. They want all KPIs on account level. Top line numbers per account is relatively easy, but cost is different, especially for fixed costs. How do you distribute them per supplier or customer to then get the margin per account, the margin per product, per product family, per geography, and so on. Some sophisticated analysis. They don't have a big AP problem. He says customers are more important than suppliers. They have a supply management feature, Spendisk. Their supply negotiation process is that the team who needs something negotiates first, and he does the final negotiation and some optimization parameters or payment terms, free features, sometimes the price for the next year or the price for the current year. Renegotiation for SAAS often is related to the employee number. Sometimes it's a bit limited, especially for the bigger companies because they are too small. They don't really do a regular review. It's expected that the owner of a tool does it, but I guess it's not really happening. Then Stefan Schlosserik. Abgrenzung is very important. There's a difference between Kostenstellen und Kostenträgerrechnung, which is important to differentiate. It's important to have a profitability calculation per product. Der Vorkontierung is partly automated, but it doesn't really run yet, so it's not really smart. They plan on Kostenarten. He would like to simulate scenarios. So what happens if I add one employee to the budget? Notes to myself. I think we need to work with Excel and build a data model that lets you work very flexibly with Excel. Many companies do not know where their profit is coming from and where they make losses because it's not fine-grained enough. Companies need profitability on product level, company level, sales manager level, and so on. But the decision-making process is very important. Companies need profitability on product level, company level, sales manager level, and so on. But the distribution of costs is a problem. For example, a service technician, how do you distribute the costs? You have to look at the hourly effort and then split it. Even killing the worst customer, if it's big, could be a big problem. Sorry, supplier. No, sorry, customer, because they are still covering some costs. So they have Deckungsbeitrag, basically. Then Tim Wöller VC said good revenue is 100k ARR. Then Nicholas Libeno, he says OCI is working well. Cashflow, they have nothing, only Excel. For him, budgeting is difficult because they have to predict the behavior of users. That's difficult to do in software, that's why they use Excel. For example, for employee planning you have to calculate when do you increase the salary, is it a standard increase, is it exponentially over time, is there a bonus. The downside of X is that it's very error prone because complex. A killer company would if MOS would really go into automation. Important for them is revenue per sales manager. Their preparation and getting and cleaning of data is very tricky. Has a lot of hygiene required to process the data and make it usable. Datafast already a data model, but it's an accounting model, that's interesting. They have to do abgrenzung very granular and it's too big for Excel, so they have a tool for it. He would delegate pre-accounting, there's still like invoice matching, a lot of time is required and also anomalies and a lot of discussion with teams. Then last interview Moritz Diederich. He says that MOS and Candice are not competition, they are too old and slow. Now the next step is complete automation, bookkeeping is still very manual. He says there are a lot of open accounting positions on StepStone, so people have problems finding accountants. He is also using Excel for FP&A, it's the best tool, very flexible, it's the kleinstmögliche gemeinsame Nenner. He says we should use GDPDU exports to analyze with AI and SUSE data from .DEV. He says FP&A is very crowded. He says the FP is quite automated, for example, the investor reporting, but they often have different data. For example, the CIM has different data than charge B, because in the CIM maybe a discount is not correctly documented. Then it's difficult to find out what the discrepancy is and why and what is correct.