Caveat Lector: This article is based on my actual experiences. While I usually try limit my opinions, I remind the reader that opinions are like noses – everybody has one. The interested reader would be wise to consider other opinions and experiences when deciding their approach.
While this blog focuses primarily on topics related to Procurement & Contracts, I am taking the liberty of discussing some principles that apply to other areas as well.
I read the MIT article stating that over 95% of GenAI initiatives have failed with mixed emotions. Finally, a prestigious institution stated a fact known to those who of us who have worked in this area. Having driven significant digital transformations successfully, I can confidently assert that there really is no excuse for these failures. With the possible exception of integration, the primary technical & technological limitations – processing power, computational speed, storage, network connectivity and speed, analytical libraries – have been largely addressed. IMHO, there are two main reasons – other informed sources will give you between 5 and 7 – viz. data quality and user education/adoption. Address those early and the scales tilt dramatically in your favor.
With regard to GenAI, let us state some basic facts
- Agentic AI / GenAI work on relationships primarily within a textual content. Simply put, it identifies combinations of words to make sentences, paragraphs, posts & presentations. Thanks to the incredible underlying technologies, it builds layers upon layers of relationships and amazingly produces high quality content which appears to make sense. Note however that it is not trying to do a traditional optimization nor make a true business decision i.e. caveat lector applies to all recommendations.
- The content that GenAI generates is dependent upon the information that it is fed. As was seen with Grok, it generated some extremely antisemitic content since the corpus of information inadvertently included some very inappropriate content. On the flip side, the higher the quality of the corpus content, the better the recommendation.
- GenAI is not free. The more extensive the search, the more expensive it is. As information proliferates, the cost of tokenizing the content – a geeky way of saying searching for and incorporating content – will at least partially offset the efficiencies due to improvement in technologies.
Having said that, I am extremely optimistic of the value that can be realized through GenAI and Agentic AI. Both are extremely well suited for text heavy business functions such as Procurement & Contracts. Using a very rudimentary version of the AI agents (think primordial Agentic AI), my team was able to save over $1.5 Million per quarter and it took only 3 months to set up and to start generating the value. Just FYI, this number was validated by the CFO. I am certain that today’s initiatives will realize significantly larger benefits.
A couple of quick points to consider a possible approach for your GenAI / Agentic AI initiative:
- Focus on the problem not the technology – (This is surely a trope by now). Most initiatives focus solely on leveraging a single capability or technology. That is similar building a house with only vertical pillars. A holistic or complete solution may also incorporate ML/AI (the quantitative part of AI), visibility, optimization, etc.
- Address the aforementioned bottlenecks by:
- Educating the end-users at the onset of the project … or earlier. Increasing the user involvement significantly improves user-adoption, a key factor for success.
- Limit the data scope and focus on high quality data. As much as possible, focus on data relevant to your business objective. Adding additional data has diminishing returns and usually brings additional issues such as data quality.
While developing and deploying such solutions, I had some very pleasant surprises.
- Value realization is actually quite rapid. Bottom-line benefits show up in as soon a month. Set up KPIs to track the realized value.
- Data quality improves quickly. A virtuous cycle is created where data is used more when the data quality is high and the increased usage improves data quality.
- Business silos start to break and cross-functional collaboration increases significantly. A second virtuous cycle occurs when the high(er) quality data and information is used by other business functions which in turn drives improved analytics and data quality.
- The users directly improve the solution quality. An educated user has an intuitive idea of what “good” looks like and where the value lies. They take it upon themselves to suggest modifications and actively participate in issue resolutions.
Regardless of whether you are planning to start a GenAI / Agentic AI project or planning to pause one, I recommend an excellent (free) book about building AI agents by Sam Bhagwat. It gives you several ideas to incorporate into your own efforts. If you need any assistance, please do not hesitate to reach us – Naresh Rao or Chand Sooran. We would love to connect.
