Technology Behind Awfis’ Highly Profitable Managed Aggregation Model

What is the role of tech, digital and tech partnerships in identifying properties where office space can be offered for commercial purposes in various cities ?

Awfis operates on a Managed Aggregation Model (MAM). In the sense that we do a revenue share with the landlord. For a revenue share, I have to get two things very accurate: I need to get the pricing and demand information accurate. Awfis will be able to strike a deal, where it's profitable both for the landlord and us as a company. 

For the pricing information, we keep track of the ongoing rental prices. Put that into a database. It helps us do normal predictive modelling in terms of how the price will increase. We've added AI layers on top of predictive modelling. This practice has given us good results. For example, when a particular area goes into development—take Nariman Point (in South Mumbai), the oldest part of Mumbai commercial. Nariman Point’s real estate has historically seen price appreciation. A linear trajectory of prices in Nariman Point would always show an uptrend. But beyond a point, it stopped increasing because other commercial hubs came up and then demand started moving to other areas in Mumbai.

The interesting part in our demand forecasting is not about what is happening in that locality, but what is happening across the city. That is where a lot of predictive analytics changes into AI analytics. It helps to identify where the next high growth micro market will play out. This is where there is a gap in terms of the supply price being X and the demand being at X + Y. That’s where I have an arbitrage opportunity.

Another example could be - Vashi in Navi Mumbai. The rentals there in a normal demand scenario would be ₹100, then 105, then 110, then 120, etc. However if a new micro market emerged, taking away customers from Vashi, then that 110 to 120 will not happen, but 110 will actually fall back to 90. So that’s where we get our whole modelling, where we figure out across a city which are the demand pockets that are going to emerge in the next year, year and a half.

What is the role of technology in arriving at profitability while signing deals with enterprise or SME customers?

There is a client we have just signed at a particular location, for which the average price is ₹10,000. However one company offered to pay ₹10,500 for the same location, while another company offered ₹9,500. The latter gives us more profitability. How?

What’s interesting is, our modelling does not look at the price alone—it looks at many other variables. The price signed for by the client (be it ₹9,500 or ₹10,500) is just the base price. 

Now, the company that is paying me ₹9,500, because of a bulk deal, also gives free lunch to its employees, which is consumed in my cafeteria, where I make a certain percentage margin. So while they pay me ₹9,500 per seat per month, every day their employee spends around ₹100 on meals. Spread over even just 20 working days, that is another ₹2,000 of revenue that I’m getting. 

Now, ₹2,000 is at the top-line level, but I still have a margin play in that. Certain companies also give incentives for people to stay late. How those incentives are consumed is that they ask my cafeteria to provide evening snacks after 7:30 pm. Some companies even give free morning breakfast.

What we do in our modelling is, when we engage with a customer, we try to figure out what the HR policies of that company are and therefore price the seat accordingly. That way, I can increase my overall yield per seat, per employee, and make more money overall. This is one example where, in this case, the enterprise customer—even though they’re paying less—is more profitable for me versus the SME customer.

So how do you use Agentic AI in this use case ?

We use Agentic AI while scoping a customer, trying to understand their HR policies. Agentic AI is used to look at the publicly available information the company puts out.

It could be on their website, or in the job descriptions they post on Naukri.com or LinkedIn. What the agent looks for is if companies - as a part of their HR policies - are offering any incentive to employees like free lunch, etc. We use the agentic part to build a database. which then feeds into our CRM.

This information is then looped back in the price range quoted to the company by the concerned sales executive. He’s made aware of the possible quotes to be offered to a particular company - ₹1,000 or ₹2,000 extra per person per seat, and therefore the company can afford a ₹500 discount on the seat price. The reason is that overall, we’re going to make more money from this customer.

The sales person will not know the HR policies of the company. So we use the agentic part to empower the sales team with that data, which flows into the CRM.

Landlord partnerships and AI, what’s the connection ?

As far as landlord partnerships are concerned, we have to figure out two essential pieces. One is whether I can offer the landlord a pricing structure that is in his interest by saying, “I will give you a certain yield on your property,” which comes from my demand estimation—if I get it accurate.

The landlord is happy with Awfis if I get it accurately. If I don’t, he may give me a notice and ask to vacate. That has a ripple-down, negative effect on my brand. But if you look at my quarterly results, it’s very rare that a landlord asks us to vacate a property.

In fact, we may sometimes vacate a property ourselves because we think that a particular micro market is going down and therefore we don’t want to continue in that market.

Here also, a lot of AI is used because demand prediction is critical. It’s not a simple, linear estimation like: 100 will go to 110, then 120, then 130, then 140. WIth the projection going false, the 140 may actually crash to 90 because a new micro market has emerged nearby. That is where the whole demand estimation work happens, and we’ve managed to get that fairly solid.

However we haven't got it right every time. In some cities where Awfis don’t have a very large presence—particularly Tier 2 markets—our data is sketchy. And if the data is sketchy, then obviously the modelling is also not as strong.