What fails AI ?

 The entire world talks about the hallucinations of AI ( incorrect answers), opacity of its workings, threat of job losses. But have you pondered what fails AI? what sabotages it and ensures its failure? When the models and tech come out of labs to the real world , what awaits them?

A very small percentage of AI projects get completed and deliver value. What comes in the way?

1. Big 'D' : For a ChatGPT, the billions of parameters get fed with mountains of data from a number of open source/ commercial data. For a business model, the data resides in multiple locations owned by various stakeholders - commercial / regulatory/users/customers : almost none offering the data easily. The sources are difficult to access , data is of varying quality and trustworthiness. Connections are unreliable . Compliance /Geopolitical /Privacy concerns can shut off those data pipes without notice.
2. Survivorship: With all the media talking up job losses, it is natural for people to get defensive. Exceptions and individual judgements are over-emphasized, early mistakes are magnified and model adoptions happen with varying levels of reservations.
3. Trust : Data legacy and past experiences do not always contribute to users trusting the base data much less the decisions /predictions /recommendations.
4. The invisible demons: Efforts to establish data pipelines, enhance /standardize data quality are underestimated and under staffed. For businesses looking to see quick results, these pose some real big obstacles.
5. Dynamism of business : Market Place is not static. Customers/competition is changing continuously , Newer nuances come in that do not have history. Newer processes/ process attributes and market dynamics may mean the project contours may become drastically different.
6. Costs of a right/wrong decision: Not every process needs to be as precise as a Japanese bullet train. Some times the manual processes / semi-automatic/automated processes are efficient enough and AI may not add incremental value to the extra cost.

While every project has its own challenges and surprises, AI projects tend to encounter unique situations . Planning and managing them could make a difference between real impact and a statistic.

The 'Moat' myth

 Are moats sustainable anymore? Are they just glorified differentiators that are transient? Can they help you identify a long term winner?

Since Warren Buffet talked about moats for corporates, everyone in the investing world latched on to this buzzword to sell investments. Moats are the ditches around castles in olden days to protect them in case of wars and in a corporate context, they are the big advantage a corporate has over its competitors in the market place to keep winning customers at high margins.

Over the last 10 years, companies highlighted business models, scalability, Product design & pipelines, R&D, Intellectual property, Customer orientation, superior cost structures, sovereign support, technology etc as their 'moats' that help them maintain unfair growth and profitability.

Things seem to have changed. Some of them:

1. Geopolitical concerns mean R&D needs to be localized and is no longer as 'hidden' as before.
2. Talent is lot more globally mobile , thanks to remote work and tech
3. Governments no longer allow businesses to collect/monopolize data
4. Algorithms can no longer be opaque, thanks to ethical/political concerns
5. Underlying business components are now available to anyone , be it funding, technology on cloud, open sourced programs and models, skills, work places and even supply chains
6. Business boom/bust cycles are much shorter making accumulation of anything a double edged sword
7. Thanks to explosion of financial markets and technology, everything is funded ,sometimes on opposite sides.
8. Hence,replicating a successful idea now is much faster, easier and cheaper.

What now is the 'Moat' that is defensible , at least in the medium term ?

Is there anything that qualifies to be a 'Moat' any more?