Performance Appraisals with Machine Learning
Part 1: Use Machine Learning To Improve The Employment Appraisal Results
Objective
In this multi-chapter paper, I explore and discuss techniques you can use to validate and change the outcome of your appraisal using AI-ML.
Pre-requisites
This paper assumes that you use at least one of the following tools for appraisals — either as an appraiser or appraisee:
· A word processor or a spreadsheet wherein you populate your inputs as text, or,
· A home grown or a 3rd party software system wherein you populate your inputs as text.
Common Appraisal Steps
Give or take some steps, most organizations follow an appraisal process that looks like below:
1. Human Resources send e-mails to their workforce on forthcoming appraisals.
a. A calendar of important dates is shared so Appraisee know what to do and when.
b. Dos and Don’ts of the process + training material
c. Validation on team alignment and project allocations
2. Appraisee are asked to re-look at their goals that they may have set at the start of assessment period.
3. Appraisee review their goals. They modify or alter the goals if there was a missed change. (Goals changes are rare just before the appraisal but not uncommon.)
4. Against each parameter of goals, Appraisee are required to rate themselves. For example, productivity, financial goals, team work, sales targets etc. could be the parameters against which Appraisee will rate themselves. These could be on 5 points or a 10 points scale and invariably weighted — in that some parameters are more important than others.
5. Appraisee write supportive summaries and examples of the contributions they made against each parameter as part of their self-assessment. (We will specifically address this part in the machine learning implementation. In subsequent chapters of this paper, we will also look at other supplementary components that are part of the appraisal process.)
6. Appraisee summarize their contributions. The assumption appraisees make is that they highlighted their work positively and that it will have the same positive impact on the reviewer too. Unfortunately, this is rarely the case because:
a. Appraiser have a perception of what appraisee did and not.
b. Appraisee have an imagery of what they you did well and not.
c. Choice of language may not necessarily bridge this gap that inherently exists in the process.
7. Appraiser and appraisee meet on the designated date to review the contributions and performance.
8. Appraiser reviews appraisee’s performance.
a. If a rating system is followed, appraiser rates the appraisee on each parameter and often an overall computed rating.
b. If a rating system is not followed; the appraiser writes qualitative notes on appraisee’s performance after discussions. (Disagreements are more common!)
9. The data is input in the system of records and reaches HR.
10. HR shared the appraisal outcomes as per the calendar.
When appraisee is filling a self-assessment form, they have their interest in mind and write about the contributions they have made in the best possible intent, words and expression. This creates following appraisal scenarios:
Scenario 2 and 4 are not what the appraisee hope for.
What appraisees want is to reduce the probability of a negative or neutral outcome. In fact, scenario 1 and 3 are not guaranteed to be positive if appraisee fills the self- assessment in a way that creates a negative impact on the appraiser while appraisee has filled the form in the best way possible.
The truth is that we, as humans, tend to leave a lot to our choice of words but do not account for the ambiguity and confusion that might come in.
So, what I give you is a method to test the responses in your self-assessment forms before you submit them for final appraiser review. I hope you find it useful, for has a long-term impact on your career (and compensation.)
Applies To
· Appraisal process (and any other process where human prejudice may be involved).
Benefits
· Better appraisal outcome. Need I add anything else! 😊
When to Use
· Whenever you are writing something extremely sensitive.
Code
For implementation, you need to:
· Create a Microsoft Azure account. Most services are free and you will get by without paying!
· Create a Language resource in Microsoft Azure. The endpoint and API key are automatically created by Azure when you create resources.
o Copy the endpoint and API Key 1 into your code or read from Azure if you are familiar with how to extract them programmatically from Azure.
· Before you run the code, please install pip install azure-ai-textanalytics==5.2.0 and pip install pandas. I successfully tested the code on Python runtime version 3.8.
Note
Microsoft’s Text Analytics has been rebranded and incorporated into Azure Cognitive Service for Language.
Appraisal Data
Let us look at the simple dataset used in this paper. The dataset has just one column where you can paste your responses, save the file and then the code to view the outcome.
· I have used Microsoft Azure’s Sentiment Analysis. So, what is Sentiment Analysis?
· Per Wikipedia: Sentiment analysis (also known as opinion mining or emotion AI) is the use of natural language processing, text analysis, computational linguistics, and biometrics to systematically identify, extract, quantify, and study affective states and subjective information. Sentiment analysis is widely applied to voice of the customer materials such as reviews and survey responses, online and social media, and healthcare materials for applications that range from marketing to customer service to clinical medicine.
· The hypothesis is: we want your responses to have a 100% positive sentiment (read 1) or impact on your appraiser. So, if the Sentiment Analysis does not yield a 100% score, you repeatedly change the language, words of your self-assessment (Neutral and Negative ones) until you reach a 100% Positive assessment.
· Side benefit of this exercise: appraisal is a confidential process, you will typically not get a reviewer to help you frame good, positive answers. With machine learning, you have a companion that reviews your responses and shows you how far you are from a 100%.
· Paste your appraisal answers in the CSV file. (You can easily create a CSV in any of the popular spreadsheet programs, mostly with a Save As or Export command).
· For now, assume that the dataset contains the responses from your self-assessment form. The file has just 8 records.
· Look closely and you see that these are actually just 4.
· The alternate records in red colour are a mere variation of the preceding record minus the words of optimism, achievement and positivity. This helps you model which response to test, pick and paste back into the form before you submit your appraisal for review.
· After you run the code, you will see ratings of your responses.
· The red coloured output shows that some change is needed until Positive Score becomes 100% (read 1).
· The curtailed versions either lack the flavour of achievement or appear incomplete. Even though, the reader can decipher the same meaning, the red coloured ones do not have the same ring to them.
· This is a case in point. What we write in & as words has a direct bearing on how it is seen, understood and then acted upon. We think one thing, but may end up projecting different, a part of it (or worse), a negative connotation (sentiment).
· Since machine learning algorithms are a close mathematical approximation of reality, Sentiment Analysis gives a reliable process.
· Paste your (real) responses in the CSV and run the code to see the rating.
· You can add variations of your responses in the file and then finally choose the one for appraisal that has the highest Positive Score.
Tips
· Use words and phrases that carry optimism when filling self-assessment.
· Better to write more than less words — as long as you don’t write words that sound negative.
· Chose words that depict victories. For example, “Delivered revenue successfully” instead of just “Delivered revenue.”
· Do not use words that express negative clauses. For example, while “I could not complete project on time” is honest and that is how you should be. Instead consider writing the same sentence as “I completed 4 components of the project on time, and 1 later than the planned date.”. Retain the honesty at all costs but be kind to yourself. You will be surprised how dramatically scores change from negative to neutral to positive!
Ethical Perspective and Responsible AI
While appraisal settings are context sensitive, I am not advocating this implementation to alter the appraisal processes or to create a cheat sheet.
I also do not want you, the reader, to expect a radically skewed outcome in your appraisal. For example, if you have done well during the assessment period, then this implementation at least helps in ensuring that you don’t make mistakes in writing your assessment form.
If you have not done as well against the planned goals, then you cannot dramatically change your assessment but it is still worth a shot so your scores don’t take a beating just because of your choice of words. Savvy?
Disclaimer
All copyrights and trademarks belong to their respective companies and owners. The purpose of this paper is education only.
