Consultancy Technology Outlook - #2

You cannot find the answer to most strategic or non-routine business issues in the corporate data warehouse. You have to ask people. And, therefore, you have to make a questionnaire. Roughly 99% of consultants' surveys are in the Likert format, making them unfit for any serious application of Artificial Intelligence. To help consultants detox from these Likert-based surveys, combined with the need for tailor-made questionnaires in a format that A.I. does accept, A.I. comes again to rescue to produce these questionnaires in the first place.

Simply put: a consultant uploads a text (e.g., an article from the Harvard Business Review, a strategy summary from the client, the paragraph in that offering's explanatory PowerPoint), and A.I. converts it into a questionnaire. Initial curation of that output by consultants will feed a self-learning algorithm to produce even better surveys. Spiders will then look for white papers, winning articles and successful books, and translate these into an ever-growing library of assessment topics, in turn feeding the need for scope bots.

The A.I. reads the uploaded text by looking for business keywords. "Objectives" is such a word; "furthermore" or "define" are not. And the A.I. has been taught that some words go together: "goals" and "objectives", "mentor" and "mentee", "organization" and "company". In technical terms, these are 'named entities.' Named entities are instances of an object. For example, New York City is an instance of a city.

Comparing the keywords with a reference database of questions creates the resulting questionnaire. For example, at our company, Transparency Lab, we have a reference database of over 22.000 questions in 6.000 groups.

Then, a keyword map shows which keywords if the source text the questionnaire has covered and which have not. Say, the resulting rate is that 45 out of the 65 keywords in the uploaded text can be found again in the questionnaire: a 69% keywordcoverage. Obviously, with such a coverage percentage, a consultant will have to do some further curation. Once coverage percentages move beyond 90%, clients can do their curation. Beyond 95% coverage, there is hardly any need for curation. The source text quality, the question database' size and the richness of the named entities all contribute to the keywordcoverage.