Introduction: Why strategic intelligence needs better internal sharing

I'm working closely with Anna but also other colleagues to trying to make sense of the future, trying to find trends, opportunities, threats in the market and creating intelligence insights out of that but then obviously we also need to share that.

One way as Anna showed is our newsletter, the reports, the research that we're doing.

However there's also companies like Lufthansa Group but also other companies, they need also other means to share intelligence internally internally and we've tried tons of different tools however we have never really found something that

also met our own specific needs because we are very deep into methodology and how we believe innovation and intelligence insights kind of should be done and we're also kind of very opinionated on how this kind of also look like and be shared within a group and obviously we're

all budget constraints but also we're all curious so when uh kind of like there was opportunity to

Building an internal solution: What “Cortex” is

play around we started building what is today called cortex which is our intelligence our strategic intelligence solution that we've built internally and that it's available for Lufthansa group peers internally for their domains and also we're testing it on the external market

it so how does that actually look like somewhat like this but we will jump into

The metaphor and mission: An “infrastructure of insights”

the live one in a second right but kind of what does cortex do right so cortex is the part of the brain where actually all the senses come together right so where perception of sight smell everything kind of comes together where we more or less connect dots in the human brain however it's also the part

of the brain that we do not use anymore when there is an attack when there is a fire and we have to run away right then amygdala kind of comes back in place so we are running so

that's why we've tried to build cortex this is our system or even said like it's the infrastructure of insights where we try to sense change in the industry across technology and also humanity so social trends consumer trends everything out there and that kind of goes into our system where

Data model and methodology: From signals to synthesis

where we add context to it kind of by tracking signals or applying different signals using our indicator methodology, tracking across different domains, actors, and trends that in the end kind of is more than a set of 40 different relational databases that we connect to each other and that is kind of growing as we speak.

And then we follow our methodology across analysis, synthesis, and genesis, how we kind of monitor enrich but also curate certain data right because yes you could always fire up hhpt what's the next big opportunity in aviation we all know that this will not spit out the best one even if we hit the do the extra research the deep one please right so it's about the data and

the context kind of we believe we put into the system and also adding our methodology methodology that we have gained over the past 10 years into that that is kind of like a mix of information inspiration and in the

End what we kind of set out is not only giving out intelligence and information But really helping make better strategic decisions and setting direction this is kind of how we hope to support our peers and

How Cortex evolved: From side project to scalable platform

the journey on how we kind of like ended there is a phase of what today is also called AI maxing and Kind of guiding you through a little bit in

Phase 1: Airtable as a visual database for research

In 2024, I started that as a side project actually using Airtable to make, right? So Airtable kind of started as a, well, visual database tool where you could just link up different databases, save your data, and actually kind of where everyone else was still using Excel to kind of store a lot of important information. and we still, or all know that kind of information dies in Excel.

Kind of trying to automate certain flows from our research and kind of storing the data in Airtable and kind of making that available at least internally, but also already using the API

to kind of flush different front ends on top of it. Very manual. You could use some Airtable automations.

I'm not a tech guy, right? I used to be a designer. I'm a, well, futurist.

Phase 2: No-code front end MVPs—and the limitations

So kind of moving into the next phase, we then hooked up Glide, which is a no -code builder, was kind of, we tested tons of different tools, kind of making a better front end that allowed us to kind of really brand it, kind of give a certain user experience, really guide and let the users use the intelligence that we want to provide it in the way we intended to, right? And kind of ran a couple of different MVPs with different teams internally and externally.

However, we hit technological limitations across time. Most important, it only allowed 50K data rows synchronized from Airtable, which was still our choice for the data set because we wanted to visually interact with our data and not kind of just hook up like a SQL database or Postgres database.

right so kind of there are some limitations in there where we come like

Phase 3: Moving to a more flexible stack (Cursor, Vercel, cloud routines)

once you go and then yes all the white coating explosion kind of games I went through both I went through rapid I went through literally probably every white coated it was out there and currently landed on a stack with cursor that I

actually prefer over cloth code however everything else is really happening in cloth but I just somewhat prefer a cursor over that what I will show you is

actually hosted in Vercel and our data is still in Airtable right so we've overcome the limitations by ditching light we're also replacing make now more and more with actually cloud routines and some of the other agentic workflows

in there so we've actually moved from data set and automation to no code platform to is somewhat let's go I always in bracket wipe code a platform right because there is already some more sophistication in there than what a

typical bold lovable kind of like would give you uh however i'm not deaf there is so far never a deaf has seen my uh github repository it was probably good uh so let's see on that however

Impact: Cost reduction and less manual curation

with all of that we've been able to drastically reduce the cost of the data tools and the other tools that we typically used in the very first phases by more than 80 we've also reduced the the need to curate the data, right?

So we kind of, we bundle and ingest data from all different data sources.

And in the past, it kind of used, when we tackled a new topic, kind of we had to ask different colleagues to kind of help us manually ingest certain data sets

and kind of maintain and curate those. This has been drastically eliminated.

And this day, and this is kind of like a snapshot from this week, actually is that

Custom AI skills: Orchestrating ingestion, cleaning, and analysis

I'm heavily relying on Claude's skills and conformed it to Claude Parkins so this is kind of run out I think around 14 to 15 different skills that are actually orchestrating a lot of the data ingestions the data cleaning and kind of

certain analysis for that so kind of really none of that kind of is coming from an existing github repository but kind of like they're all custom built on our methodology right so this is kind of really where like well those are not hard to build and once you know the methodology and the things that you want

to do you can build those things around they are all connected they all hook up to master orchestrator right and then you can just build from there and this had to really scale this kind of site project into something that we are now

meaningfully scaling throughout Lufthansa group so this is kind of one part that the visual portal I'm going to show you in a second something that we

Outputs: Document creation as a discussion starter

also have in there is document creation I will show you that kind of how we ingest our own data and actually then have a cloud skill kind of creating editable work documents that use our template right so that's kind of really

ready to use data more or less it's it's not a full -fledged white paper because that would not sufficient our own research quality gates but it's kind of a real like a discussion starter I will show you some examples a second and one

Making intelligence accessible everywhere: MCP and direct data querying

thing we are actually actively working on is also adding an MCP on top of the data sets and the interfaces that we have because it's a new means to interact with data right if you want to embed intelligence everywhere where people are you need to have an MCP to kind of query the data right so this is

kind of one of the things we're working next

What MCP is and why it matters for curated datasets

Yeah, so MCP stands for Model Context Protocol, right? And what this more or less allows you within Cloud or anything else is you can directly speak and talk to the data without the need to go to the platform, right? Because there is an API that is connected to a certain kind of database, which is a graph database, right? And this can more effectively and faster retrieve the data that you have in there.

Because in the end, the benefit that we bring is we have curated the best data. We have enriched it. We have built in our methodology. So we want people to base their information, their research on that data.

that they don't use let's go like the averageness of the results that chat GPT and the others kind of get from the same slot that they find online right because it's getting worse and worse kind of like the things that they quote online because they don't have the access to the data right so but kind of how to give people access at scale we believe MCPs might be a great addition to the the interface that we've built.

Product walkthrough: The Cortex front end in practice

But let's jump into that front end and get out of the slides because that was supposed to be a demo, right? So let's, yes.

Where is the question? Yes.

So basically the MCP or what we call it? Mm, what we call it? It's like a vetting tool. So you're vetting the data so you can weigh.

Yeah, this is kind of like, this is what we kind of like do with the data before right that we kind of apply the best standards in terms of how do we source the data how do we enrich it and kind of like what do we kind of bring into the right this is kind of the it's got like the expert bonus also why we are doing that right so this is kind of

people come to us because they believe us that we have the right methodology the right data and the right reasoning and curation for the future of their certain domains right and kind of want to I say use that at scale right so we try because we only

have limited resources we cannot work on any every consultant project that we want to so we are trying to distill this methodology and the data that we kind of have into those portals but also MCPs to kind of bring it in and make it available yes yeah I would like we will we'll get some things in here

Thematic intelligence and personalization across domains

So, that's our front end, as I said, it's kind of built in with Cursor, it's hosted on Vercel, people can log in, and the main thing that we do is what we call thematic intelligence, so we are getting intelligence on all sorts of different domains, and in the airline industry or travel industry, we kind of round about have more than 90 different domains.

we have personalization layer so you can come like come in and then see okay well example we are working with ESG with the sustainability team of Lufthansa they are interested in a lot of different topics that they need to cover from their strategy point of view so we are watching those right and and we are ingesting data so what you kind of see here the signals those are

Signals: News ingestion, taxonomy, and noise reduction

are scraped from news. So we have one tool, it's called Feedly, where we have different news aggregators coming in where we get the data and label it, categorize it based on our taxonomy and stuff.

So we will see a couple of examples here.

The framework: Now, New, Next

So when we come in in our methodology, we look across things around the now, the new, and the next.

So when we come in in the now, this is kind of like what I just said.

we for circularity and waste we have around 260 signals 10 of the within this month you kind of see a little bit the momentum you also see the share of like okay is this like a general news is this a project launch is this partnership right so we are tracking different type of signals and I'm kind of like assumed

on maybe zoom in a little bit you kind of like then just go in and kind of well get summaries you can get to the original source right so it's kind of of really more or less first of all like a news reader taking out the noise so really understand what is happening in this area without the need to read the news right and if you kind of think

oh the is g team kind of needs to look at all of those at the same time this is already somewhat of a hub well obviously as the editors they can go in and say like it's noise delete so there is an interactivity built into the system so it's not just a news reader you can then go in so we

We extract who's actually working in those areas, who are the actors, who are the startups to potentially partner with, so this is kind of what we then extract across a lot of different domains in the now, the new and the next.

What does that mean for the new?

Trends and clustering: Turning signals into actionable themes

We are actually then scanning those signals, trying to find trends and cluster them. them right so these are then some of the skills that really go through it and kind of where we then go in and say like okay what are the trends and the signals we are seeing here for airport

coffee waste to biofuel conversions it's very niche obviously because it is in we are still we are very deep in the circularity and waste right so if you are really working in that field this is the depth of detail of intelligent information you would like to work with right this is not all circularity sums up it's a good topic no this is not the level of

information that you would want to have right and also like funny anecdote when we were building and glide there was obviously no way of making this trend radar round but this was a clear ask from a colleague within the group like like we had them as kanbans we had them as metrics and they were like can you make the trends not in this can you make them round and we said yes we can also make it round

we looked at different tools um and the one that more or less looked very much like this quoted us 40k a year just to make our data round what is the natural response of someone with entrepreneurial mind like a builder mindset okay i take a 600 euro boot camp from my learning budget and I just start by coding this so this is kind of also one of the reasons why we said it was like restraint budget and saying like no I do not want to have it

okay yes I was comfortable using a no coding tool with light however yes it has limitations like rows but also in terms of custom front -end okay then let's make the jump and kind of went into cursor and so so this is kind of them where we bring in some of those data where you kind of come in and so this is a little bit

Trend library and scoring: Readiness levels, linking, and implications

like on future and then what we then do for example is we track those things where you also have a trend library right so if you kind of go to all trends and loads it's more than a thousand different trends and use cases you can follow your most favorite trends obviously we are all into

MCPs which we just learned about so the question is which airlines are doing what in the MCP area right so we can go into that trend we have those so we have special skills that are trained on the methodology on what's technology

readiness level what's business readiness level and we do AI assistant scoring of those things and then also do well obviously we link all the data right as I said the system has more than 40 different relational databases so everything is linked against to each other we say oh we see the signals that's why we believe for example if we do an adoption curve

analysis we are right now in the early adopter phase right and then you can run that against the strategic context that we have and say like oh what might this have implications for Lufthansa group right so and nowadays with all this technology and all those automations you can do that for hundreds or thousands of technologies trends simultaneously everything at the same time

with a team of one right so this is kind of really allowing you to do that and if we look for example that's let's see if that works because we looked at data source and that's

Actor views and benchmarking: Tracking companies and peers

actually good one here let's stop with you so if we look into domains we also can look into everything by an actor right so obviously we're always comparing us against our peers and benchmarking is a very important topic in innovation so we also build data sets and gather data on an actor level right

and this is now the Lufthansa group example where we kind of bring in certain topics you also see where do we innovate what are the project what are the news actually telling about Lufthansa group and then we come to the

additional data so here's for example something that we are currently building building up for the ESG team in a second and one thing I built over the weekend

Case study: Earnings-call transcript analysis in hours

is we got access to two earnings called transcripts right so transcripts when there is like the earnings call so stock market executives stakeholders kind of come together and those calls are public those are transcribed and you can get the data we got them from our colleague did analysis kind of gave me the raw raw data 150 different PDFs 10 different airlines over the past 16 quarters and so kind of in the end those were three different skills kind of getting this in so what we here now have the

different transcripts analyzed against different lenses that we define where we say like ah the like how often do they talk about ESG how often do they talk about AI and automation and how does does that change over the years and we can also make the earnings calls like now loading the transcript so they are all available in the data right so you can also always go back on like where is this data actually coming from right and in the end that's three different skills or even pushing that into a table and then building the well best type of interface on top of it and kind of well

Well, from someone sent me a zip file with the PDFs to having this here was like a couple of hours, right?

Because, like, really, I was on a, let's say, a diamond wedding of the aunt of my wife. And kind of started with that and was like, only having my cell phone, like, okay, what can we kind of do? obviously it does not have then the access to right fire it's kind of stuff

from there so what I did was I defined all the lenses on my mobile phone kind of like on the fly kind of like okay planning those out and then kind of saying okay how do we actually want to do those lenses so when I come like came back then all those line lenses kind of talked through and I kind of just had to run the skills because I did the definitions more or less on the fly and And this is kind of really how this is set up these days.

Spotlight reports: Auto-generated Word/PDF drafts from vetted data

And kind of just jumping back to the example that I just said, going for the personal filters, like circularity and waste is, for example, here the spotlight feature, which more or less is going through the data and kind of then creating a word file in the PDF, which is kind of following our design. always the same structure of okay what is like like doing some assessments kind of looking on technological readiness who are the players that we identified with the data and kind of putting that into a document as a word file which is easy to edit right so this is kind of eliminating this

blank space phenomena where it's like ah i need to write a white paper i kind of need to start start writing a report because I want to exchange with someone on the topic completely taking this out and this is also not a random chat GPT search no this is kind of against the data that we are continuously vetting the methodology that we are putting in here and that we can run on all the

different things and themes and as you kind of see there is tons of different other tools and functionalities in there but let's move into some of the questions that

Data quality and trust: Confidence levels, risks, and expert review loops

I was just wondering, you said you're vetting it yourself, but you're talking about yourself as one person that does all of that. How confident are you that all the data is correct that is displayed here? Not 100 % for everything because we don't check every row and everything.

So we are confident on the methodology and the way kind of like how we put something. things a lot of the data kind of comes through even that we buy some of those data right or that we see it is it's a signal that we're tracking right because it comes through a link and then

the parsing part of it is typically nothing where you see a lot of hallucination right so there are some very basic things where we then use the methodology on rather quantifying the data right and then it's let's say there is low risk of hallucination and then it might be rather a more overly confident interpretation on some of the data but for at least like the

MVPs that we've run and this is also where we kind of then go in into some things people can kind of then let's say comment vet against it what we also have in here for the internal ones are actually so this is something we're

Human-in-the-loop scoring: Comparing expert ratings with automated scores

doing with Lufthansa industry solutions together is we can take the trends and kind of then do our own expert ratings on top of it right so so we are building a second database of expert scoring against the automated scores right also on the scoring that we have and this is

something that the more we roll it out internally the more we will have the data on those things right so this is kind of then really here where we can go in and really do the different experts rating and see and do additional scoring so this is kind of something that we

obviously have in mind that we have the experts that also have different view on what might be in the market out there and kind of then yeah put the data against that