Technology quotes

During a transformation, organizations show clear patterns.

For the past decade, PRAIORITIZE helped consultants and their clients to gather hard numbers about transformations. As there is usually little or no data in the corporate administration about the transformation itself, one has to ask people. Still, one can get hard data when asking people; you just have to take some precautions. Then, their answers power algorithms. Algorithms lead to patterns. Transformation patterns. Finally, patterns lead to predictability, control, success, and better sleep.

Scale makes these patterns 'universal'; science makes them thrustworthy.

After studying hard data from 23,000 team case studies in over 1,000 transformations in more than 15 industries in more than 30 countries, the PRAIORITIZE people academically underpinned fourteen of these transformation patterns in scientific, peer-reviewed journals. Half of these patterns showed clear reasons (not: all reasons) why transformations fail. The other seven patterns showed how artificial intelligence brings predictability and peace of mind back into the game.

Patterns turn into algorithms and into advice.

Bayesian probability (also known as evidential probability) is the process of adding prior probability to a hypothesis and adjusting that probability as new information becomes available. Say, 3,500 teams behave according to one of three patterns in their ambition to improve. Then, team 3,501 will highly likely behave (very) similar to the probabilities derived from the first 3,500 teams. Algorithms capture the patterns and their probabilities. The algorithms use these probabilities to analyse and advice any next situation: copy the successes and avoid the mistakes from 23,00 team case studies.

Universal patterns occur irrespective of the topic at hand.

Stuyding the 23,000 team case studies, it was quickly noted that these transformation patterns occurred irrespective of the transformation topic at hand. Just put different content in front of the algorithms and the advice now extends to this new topic. Generative A.I. (like GPT-3/ChatGPT) is perfect to create the textual elements with which the algorithms, in the form of the Virtual Consultants, give advice.

Train A.I. with consultancy white papers.

What kind of content should GPT-3/ChatGPT then create? The company scraped 11,000 whitepapers from the 100 most-noted consultancy firms. Dedicated editorial A.I. read the white papers’ 100,000 pages and 1,5 million sentences and ranked the white papers among others according to, e.g., originality, contrasts and comparisons (STORY). And the use of numbers, page content diversity, and other parameters (STRUCTURE). Quickly, the more promotional white papers were tossed out and eventually the best 20% remained. The content of these remaining white papers was extracted, grouped and fed into GPT-3/ChatGPT. In summary, team case studies powered the shape of the advice; white papers and GPT-3/ChatGPT its content.

Providing A.I. with examples gives much more nuance than just instructions.

GPT-3/ChatGPT uses instructions to create content. Other than instructions, GPT-3/ChatGPT can be finetuned with training examples (usually hundreds or thousands) to add much more nuance and predictability to GPT-3/ChatGPT’s outcomes.

Training GPT-3/ChatGPT is like making soup.

GPT-3/ChatGPT has been criticized for its "long-range semantic inconsistency”: the longer its output, the further it strays away from the user’s intention. Conversely speaking, GPT-3/ChatGPT gives excellent output on the short range. So, the PRAIORITIZE people kept it short and looped GPT-3/ChatGPT’s output back to GPT-3/ChatGPT for many times, every time sifting away some ingredients and adding others until they reached the required quality level to power the Virtual Consultants.

Virtual Consultants need more A.I. than only GPT-3/ChatGPT.

For example: document classification, keyword extraction, sentiment analysis, TF-IDF, topic modelling (LDA), hyperparameter tuning, proximity scoring, k-means, clustering (unsupervised learning), Bayesian probability, random forests, and proprietary algorithms to prepare the Virtual Consultants for giving advice as well as proprietary algorithms for target setting.