Skip to main content

Decision framework for choosing a knowledge solution

This is the fifth post in our guest blog series by Esteban Kolsky of ThinkJar. See the original post for an introduction and a list of other posts in the series.

An almost perpetual problem with AI technologies is deciding which one, or which ones, to embrace as an organization. I am constantly being asked about one or another – or worse, being asked to compare solutions from vendors that are quite different. Comparing a machine learning solution to an NLP engine is not fair to either. 

There are many aspects to making this decision, but I found five elements that play into choosing one solution over another. The following is a decision framework – it allows you to make the decision but does not name any one solution.

  • Outcome Sought. There are five customer-centric outcomes you should be seeking: understanding, resolution, insights, expectations, and engagement.  If you are trying to solve an issue with, for example, understanding your customers better, a tool like NLP may be better aligned as it does precisely that – deconstructs the language into manageable components. If, however, you are more concerned with resolution, a classifier and automation tool may be better suited for that. Understanding which outcome you are seeking and how it correlates to the processes you are trying to improve by using AI is the first question you should ask yourself.
  • Expertise Required. The answer to the problem or question, or the resolution workflow that AI will enhance, is rooted by expertise and wisdom. Knowing where that expertise and wisdom resides, how to access it, how to leverage it, and how to maintain (optimize) it over time are the questions you should be asking next.  While an answer may reside in a knowledgebase and be (mostly) static, at times it might reside with a subject matter expert (SME) and require a different path to acquire the information. Knowing the likely path(s) to information will give you the second clue as to which tools or technologies will work best.
  • Deployment Time. Many things go into understanding and making decisions on deployment time but the two most commonly used and that you should understand are training time and maintenance time. If it takes months of training, as is the case in a few NLP or some neural networks components, to reach an entry level, and then more months to get to a better place, the tool may not be what you’re looking for. On the other hand, if a tool requires constant supervision for maintenance yet you are short on resources, that may not be the best choice. Each tool will require different levels of preparation and maintenance and you should understand these. Deploying “within weeks out of the box” is not a metric, it’s a marketing statement. Understand what’s behind the “few weeks” and how it works on an ongoing basis.
  • Enterprise Adoption. Culturally speaking, most AI tools will die from lack of adoption. While the technology may be there, and all the other elements may align, if the people in the organization don’t use it or promote its use, it won’t live for a long time. These tools all require extensive follow-through and maintenance in addition to support and understanding how to achieve their objectives. None of these tools will be there over the long run if the ability to maintain them is not there, and the maintenance won’t happen if the corporation does not adopt it and support it with budgets, time, and other resources.  Ensure that whatever you are trying to do can be supported within your organization to determine if it makes sense to deploy.
  • Cost. The “hidden” project-killer. While most of the tools available as services today are relatively cheap to implement, most costs of using them don’t come from the tool itself but from the organizational costs of implementing and adopting. AI is all about clean and useful data, and if that is not available in your organization, the path to getting there is complex and expensive. Further, maintenance can cost as much as deployment – more in some cases – and require resources (people, budgets) you are not allocating today. As an example, traditionally KM tools require maintenance whose cost increases by 8% every year – is that cost something you can afford? How about double that (possible with more advanced tools that require more maintenance)? It also requires data people, knowledge administrators, and potentially other people who are not in the organization today – can you get them?

These are the five areas you should understand. They all combine in different ways for different tools and different enterprises. Therefore, there is no “one size fits all” model for decision making. Further, each tool and technology you will consider will have different answers to these questions. 

It is almost impossible to have a chart or table that will guide you, and a matrix to highlight all tools and all these questions would be unwieldy. But knowing what questions to ask, and where to focus the strategy building, will go a long way to making a decision that will benefit your company.