The market noise around Gen-AI at the moment is deafening, every CEO is asking their teams and CIO’s what they should be doing and where can they jump ahead of the competition using this capability. Every board is asking their CEO how they are keeping safe in the wild west of data extraction and accidental sharing, which, in a highly regulated market dealing with personal data, is a very real concern. Everyone is playing with ChatGPT and other easy to access AI tools to see what is really available. 

Having tested in the mea lab what can be achieved with GPT, having spoken to many of our consulting partners and having spoken to many of our clients, we have collated a view on the key decision points around build vs buy for data extraction. 

It should be noted, and this is important, that many vendors in this space are actually build options, albeit they come with information and expertise from working with many clients. 

Drivers to build yourself on GPT (or an equivalent Large Language Model) 

  • Competitive advantage 
  • Kudos vs competitors (friends in the market) 
  • Faster benefits? As you will see later in this paper this is likely a myth 
  • Avoid working with difficult to manage vendors 
  • Control of your own destiny 

There are many many areas where Gen-AI could be applied to Insurance, data extraction is one step, the first step in the underwriting and claims process. This is not a step that alone drives competitive advantage, what you then go on to do with the data – inform pricing, inform which risks to bind, apply risk appetite rules, generate data for actuarial teams, and many more – are what would drive the competitive advantage. 

Drivers to build with a tech vendor 

  • Faster to deliver (than going it alone) 
  • Do not have the internal capability/ expertise 
  • Understand competitive advantage is driven post data extraction step 
  • Use a vendor to skill-up internal team 

One word of caution with a build option, ensure that your tech vendor/ partner has feedback (that you have verified) from multiple other clients – have they delivered this before successfully? Do they have clients who being 2years+ into a relationship will talk positively about the accuracy of extraction and speed of deployment? Are they planning on charging you for the never ending ‘AI training’?  

Refer to for more thoughts here: Selecting a Gen-AI vendor for data ingestion

Drivers to buy the capability from a vendor where it is proven to works 

  • Faster to deliver 
  • No risk of endless AI training 
  • Competitive advantage is what we do with the data, not how clever we are at extracting the data 
  • Low/ zero project cost and lowest run rate cost 

Based on analysis of time and cost feedback from clients this approach can save up to 70% of cost and time from an implementation of this capability.  

Critical Self-Reflection questions 

  • Do we see the data ingestion as a capability a competitive advantage or a commodity? 
  • Do we have the time and budget, and risk appetite for a build option? How critical is the need? 
  • Where does our business case deliver most benefits? Data ingestion vs triage vs data augmentation 

Watch this space for our next article on – Implementing successfully data extraction with Gen AI 

See our existing articles on, 


Transforming the Insurance Industry: Solving a $400bn+ Problem


Selecting a Gen AI Vendor for Insurance Data Extraction: Where to Invest and Where to Be Cautious 


Accelerating ‘data-first’ in pre-bind processing to unlock the potential of Blueprint 2

Mea in the news

AI training is dead, beware this bottomless pit 

Mea in the news

PR News Wire: mea Platform Ora Continues to Lead the Market


The Digital Dilemma: Decoding the Challenges in Commercial & Specialty P&C Underwriting Automation