Carragee Consulting

Sometimes the most valuable way to contribute to an automation project is entirely manual.

Two pricing projects I did entirely manually that potentially could have been done with machine learning techniques that had tremendous value, but required very little skill.

1)Culling through paper contracts and PDFs to find any redlines or deviations in our standard T&Cs for national and strategic accounts to determine where we could legally apply our Annual Price Increase and at what level 5% or the CPI.

2)Pulling up every retailers and distributor website who carried out brand to track which computer or brands they also carried. I wanted to do pair this with sales and market share data to see if some head to head competition within the same retailers was more detrimental to our sales or pricing power than with other brands.

With supervised learning approaches training data needs to encoded or labeled and sometimes to build base that requires human interaction.

Sometimes the data you need for analysis doesn’t exist and you literally have to build it by hand. Then you need to ask how will this be maintained after its initial capture and how could the process be changed to be more sustainable?

For #1, the contract T&Cs we should have had a way to capture expectations in a database as they were negotiated and probably only have allowed a limited number of predetermined varieties of exceptions.

In the long run we could have found some image processing and NLP to scan documents to speed the manual review process of unstructured data.

For #2, maybe a dealer/retailer survey would have worked instead.

Either way, sometimes a manual task that’s easy to execute is better than automating a one -time or rare effort