Projects / Which employees to approach with a retention offer

Which employees to approach with a retention offer

Course project, Michigan Ross MBA. Logistic regression and a random forest in R, scored by the saving of a retention program. Not client or employer work.

R
R scriptModels and the business result
SituationAbout 1 employee in 6 leaves the company. To replace an employee costs about $25,000.
ComplicationA retention offer costs $10,000 and does not always work, so the company cannot make an offer to all employees. It does not know who will leave.
QuestionWhich employees must get an offer?
AnswerApproach the 15% of employees with a leave risk above 67%. The offers reach 92% of the leavers and save $2.6M for 3,598 employees.

The saving depends on who gets an offer; the model and the cutoff are the two levers

  • Saving of the retention program$2.56M on the test employees
    • Replacement cost avoided
      • Leavers who get an offerThe lever: the model that ranks employees

        Does the model find the leavers?

        Yes. 92% of them, with a random forest.

      • Success of an offerFixed: 60% (assumption)
      • Cost to replace an employeeFixed: $25,000 (assumption)
    • Cost of the offers
      • Offers madeThe lever: the risk cutoff

        Above which risk does an offer pay?

        67%. $10,000 divided by 60% of $25,000.

      • Offers to employees who would stayThe limit: wasted cost

        How many offers are wasted?

        6 of 529. 99% of the offers go to a leaver.

Lever what the decision changes Limit what the answer must respect Fixed outside the decision

Saving from the retention offers

$2.56M

for 3,598 test employees, against no program

Employees who get an offer

15%

those with a risk of 67% or more; 99% of them are leavers

Leavers that the offers reach

92%

523 of 566

A model that sees curves and combinations saves $2.6M; a straight-line model loses money

Assumptions: to replace an employee costs $25,000. A retention offer costs $10,000 and keeps 60% of the employees who would leave. An offer pays when the risk is above 67%.

Three groups hold almost all the leavers, and each has a different cause

Low load, low rating: 2 projects or fewer, low satisfaction and a low evaluation. Overloaded: 6 projects or more, or 270 hours a month or more. Strong performers: 5 to 6 years of tenure with a high evaluation and high satisfaction.

The model

Decision

Which employees get a retention offer.

Objective

The highest saving: replacement cost avoided, less the cost of the offers.

Method

Logistic regression and a random forest in R on 11,991 employee profiles: 70% to train, 30% to test.

  • Risk is not a straight line: employees with too little work and employees with too much work both leave. This is why the main-effects model fails (AUC 0.83) and the forest does not (AUC 0.99).
  • One offer does not fit all three groups. The overloaded group needs less work, not a bonus. Outside the three groups, only 1% of employees leave.
  • The file has 14,999 rows but only 11,991 different profiles. The repeats were removed before the split, so that the same profile is not in the training and the test data.
  • The saving depends on the three assumptions. With a lower success rate of the offer, the cutoff goes up and fewer employees get an offer.