Can AI Build Faster Consensus in Business Negotiations?

Proof of Concept built on a real-estate case study

Image from Canva Pro

Knowledge is power in negotiations. One who has more knowledge about the situation holds more chance of convincing the other side.

Not necessarily all stakeholders in a discussion will have equal access to the data, as it depends on various business conditions. The lesser the data access you have, the lesser the knowledge you gain on the subject and it carries disadvantages in the negotiation.

Can artificial intelligence bridge this gap, and bring up the knowledge of that stakeholder on par with other stakeholders? Let’s explore.

This article is for the Deep learning enthusiasts, who wants to code and see, how the neural networks can bridge the gap of uneven data access, by building substantial knowledge from limited data.

We have built a prototype by studying the case of apartment rentals. Will walk you through the problem context, data set, libraries, and algorithm used to train the model, the validation results, and the conclusions made. Here we go.

Case Description

Every year millions of rental agreements happen between the landlords and the tenants worldwide. In most cases, we see a difference of opinion on the rental price leading to longer negotiation cycles.

While the tenants rely purely on the public data (market), the landlords consider both the public and the private data specific to the asset.

If we build a model that learns deeply from the public data and determines the realistic rental price of a specific asset, we can build faster a consensus between the tenant and the landlord and reduce the negotiation cycle.

Data Set

We did this experiment on a dataset of 268 thousand apartments, with 49 features. Feature descriptions are available at this link.

Problem Statement for PoC

Service Charges are varying from 5% to 60% of the Base Rent. By providing a few parameters of interest, a Tenant should be able to find out whether the Service Charge quoted for an apartment is Low, Medium, or High.

Data Preparation

1. Have cleaned the data for NA, Null
2. Retained 8 parameters and dropped the rest 41 parameters.

3. Added column serviceChargePercent, representing serviceCharge as a percent of baseRent.

4. Added column ‘target’: representing the category of serviceChargePercent.

• Low: <=15%
• Medium: 15–30%
• High: >30%

5. Removed the columns: serviceCharge, baseRent, and serviceChargePercent. Their presence will make the model overfit.

Neural Network Model

1. Using Keras Preprocessing library have encoded the 6 features, excluding the target variable.

• 3 numeric features : livingSpace, yearConstructedRange & garden
• 3 categorical features: regio1, condition & typeOfFlat

2. Created a Dense Layer of 32 units with activation function as ‘relu’.

3. Created a Dropout Layer of 50%.

4. Trained the Model and tested it on 20% of the raw dataset.

Result

On test data, the model has predicted the ‘target’ with an accuracy of 71.4%

The accuracy is found to be good at the PoC level. It indicates that in most of the cases, the model can collaboratively determine whether the Service Charge for a given apartment should be <15%, 15–30%, or >30%.

This helps the tenants to make an informed decision on whether to proceed with the quoted price or to negotiate it further down before making an agreement with the landlord.

Conclusion

Neural networks can make an instant deep comparison of the business entities and classify the dependent variables at a high degree of accuracy. This classification can act as a standard point of reference in the negotiations to bring faster consensus among the stakeholders.

Concept Extension

Based on the clue taken from this PoC, I predict that some applications may come up in the near future based on the deep comparison abilities of neural networks. Like:

1. A Property Management Portal predicting the rent & expenses, as per the advanced search criteria entered by the tenant. Not limiting the calculation to a fixed set of rules.
2. An Agent Application notifying the landlords on asset improvements that can bring confident returns.
3. A Recommendation Engine telling the tenants, the likelihood of a vacant apartment getting sold out in a couple of days. Based on the modified price of the landlord.