Date: 12.10.2025 (2025-10-12)
Nikolaus Müller (12.10.2025) McK vs. Amazon vs. Capgemini He mentioned working with low code platforms • What data do they have to usually pull? and from which systems? What are challenges? Who delivers the data? • What software tools are they using? (e.g. Celonis, Palantir) • Which internal tools do they have? How good are they? • Who would be competition? (e.g. Palantir) Why do they not use Palantir? • What data are they mostly working with? How do they manage raw data (e.g. invoices) if the data is distributed not properly structured? e.g. invoices are not categorized. • What are the most important areas / functions within a company where most things can be improved typically? • Gibt es Abteilungen/Rollen/Funktionen oder Prozesse wo ma den Grossteil der Optimierungen rausholen kann? • Which processes / topics at customers are the biggest challenge for them to do their work? (e.g. data situation, etc.) • How to enable consultants to create recurring SaaS revenue? • Generic: what is the biggest pain point for them in projects? • Hypothesis: when it comes to process automation thinking in functions makes more sense vs. thinking in industries • Product ideas: • API as a Service • Enable to constantly copy data from legacy system in certain format to other software • Invoice → PO → Contract product Notes: • Rider, stack.ai, n8n, Google Agent Space • General: • He thinks low code will be replaced -> no code and code solutions are squeezing them • McK has an agent platform -> super important to educate users • Often people do not know what process they want to do; they do not understand LLMs -> education builds trust • Complex processes are hard to maintain • Define processes, which tools you need, risk approvals • Make it easier for users is good; integrations are important; • CoPilot studio has a lot of enterprise grade integrations -> often APIs are not good enough / not comprehensive • Ryder - it is a bit verticalised • They come from publishing • All others are generic • Model ML -> YC company - check it out • All platforms are all the same and work the same • n8n and Claude -> differentiation is the biggest question • World: people will build much more agents and software themselves • How are workflows maintained -> how often to be updated • Hypothesis: there are always many software -> always interoperability layer to automate across functions • Maintenance: McK leadership has hypothesis that every consultant will build agents for customers • Currently you need people who know the tools • One accumulates complexity over time • If they build an agent for an internal tool requirements change - often technical stuff changes • Agents for customers & internal: • 6 months ago: as little integrations as possible - medium complex - document creation (credit levels, pitch docs, customer service) • Now: AI enabled RPA - create opportunities in Salesforce -> more tool integration, workflows are quite narrow / enclosed - not too much across verticals - still heavy document creation • Everything >10 tools is complex to get into production - very quick to get to 80% • There is a trend to going back to deterministic tools -> e.g. OCR - the closer to production • Patterns he sees in different projects: • Payment reconciliation • Data unification -> centralised data -> new data base • Software development -> definition of requirements & AI based coding • Finance: monthly budgeting - early warning system (classic RAG, daily briefing etc.), Auditing & compliance, build finance models • Day to day operations: meeting summaries, ToDo assignments • Deep Dive Data unification • Project team works in Excel - then later its outdated again • Now you try to do it long term -> product Rationalisierung, clustering, enrichment • SaaS revenue for McK is a hot topic at the moment • A lot of niche topics: field management services • They want to full on productise • They have a lot of tools -> the big value of McK is the combo of all their services • Most of their offering: 40 to 60% could be heavily automated • GTM acceleration • DD for PE • Future: execution will be the value -> Arschkick for people is the big value • SaaS revenue: • A large amount of McK compensation is success based -> % of revenue • Building a product is not their competence • Palantir could take their value • App store for agents to solve problems • Celonis: • Process mining vs. Process change • Palantir is classified as competitor to them -> no engagements together anymore • Super strong data layer -> centralised and structured • You build ontology based on the data • Problem with normal data layers (data lake etc.) -> normally data platforms do not understand the data -> e.g. customer ID -> you do not know what to use or where the data is from • When he worked with them they were focussed on analytics - now probably agentic now • They have a workflow tool as well • Prebuilt use cases • Common process: procurement analysis: compare contracts to market standard and with contract -> that was a blueprint process • They are extremely expensive and strong vendor lockin - companies buy it for one specific use case • Vertical agent platforms: everyone is doing it - what is the differentiation • Procure AI • Zauber -> agentic supply chain (see here) • Model ML -> investment management • If you find a niche • You can charge for savings vs. per seat charging • They do more and more partnerships -> but not a lot of coverage yet • Most agent companies are very early still • Every time you have the processes optimisation: build vs. buy • Accenture does the implementation -> mostly build • Why does McK not productise • 80% of employees are engagement based -> completely different orga for product • Culture: rotating of teams • Data in legacy software • Enterprise integrations are a big topic • Normally they do not build custom integrations - they work based on what the tool gives you (e.g. ETL etc.) • Future: MCP are going to come • Data base is the biggest problem • They have SAP but then teams have Excel lists for a lot of things • Check out Databricks • OpenAI Agent builder: MCP first; A2A by Google • Current APIs: built for old paradigm of old deterministic systems -> 20 API calls to get sth - MCP can get the data • Data lake: • Integration is super important • Customers will not always pay for one off changes but for sustainable improvements • Check out https://www.operand.com/ & Stratos • Allows C-Level to zoom out a lot -> strategic initiatives -> dashboards for initiatives -> KPIs, unit economics -> you can dive deeper etc. • Performance of LLMs • He thinks it will become better -> S-curve - now we are rather flat - then it will go up again • A lot of limits are not technical - the big factors are others -> skills shortage, no maintenance possible, risk consideration, operations model nicht aufgestellt werden, misincentive to collaborate people that will be replaced • Don’t overvalue the opinion of consultants -> tendency to be risk averse • One of the biggest questions: how do they build a comprehensive AI setup in a company -> nobody has a plan atm • They often build a deterministic n8n workflow and then it is called by a LLM • There is not really a need for local LLMs • They start with the strongest model -> if it works -> then price estimate -> then smaller LLMs • Sometimes they finetune • Cost efficiency is most important • He thinks that most companies that do host LLMs locally do it because of cost (but also a bit data privacy) • They currently use models in the Virtual Cloud of the customers