The Hindu Business Line : Room for opportunity

It is easy to classify call centres, transaction-processing centres that cater to the banking and insurance industries worldwide, and the like, under the all-inclusive `BPO' industry. But outsourced business processes that use Statistics? Mathematics? Economics?

Well, before Ramakrishnan, CEO of Marketics, which provides such services, could elaborate on his business, he had to embark on a trip to the US. So we did one better and actually got his client, Wyndham Worldwide, to give us its views on offshoring functions that aren't part of the normal `BPO' bandwagon entering the country.

Sri Raghavan, Senior Vice-President, Wyndham Worldwide, has an interesting job. His company sells resort and hotel accommodation, and it caters to a network of four million-odd members who exchange vacation weeks they own with Wyndham.

Raghavan's job is to ensure that his company gets the maximum yield out of the network every year. He says, "unlike the hotel industry, where the supply of rooms is fixed for a period, my supply of accommodation and demand vary continuously."

He uses the skills of Ph.D.s and masters' degree holders in statistics, mathematics and economics, both in the US and India, to make his job easier. Excerpts from the chat:

Tell us a bit about your business.

Ours is a unique business model. It started 30 years ago when they were selling timeshares in the US. If you got to a resort, he sells you a timeshare. That is a specific week in a specific destination. Unlike hotel accommodation. It is meant to cater to families, with one or two bedrooms and a kitchen. If you own such a `vacation week', you come to us and give up the right to use it for that year. You become a member of our community. Then, you pay us a transaction fee every time you want to exchange it for a vacation week at another location through our network. Our revenues come from membership fees and when members find a match.

The business model is that people own leisure vacation assets and they want to swap. We facilitate the swapping mechanism. Over the years, we found that there is also opportunity in providing non-traditional accommodation. We have a very large business in Europe, with 55,000 such owners across the continent - owning and wanting to swap castles, country cottages, villas. We have a 12-bedroom castle for instance. .We will contract with them and they will list the property with us, exclusively. For a period of time during the year, mostly in summer. So we list 20-26 weeks of the year. It is broken down by the week and we sell those assets to others by the week. It's like service apartments in Bangalore.

Here, because it's membership-based, our revenues are like an annuity stream. Once people get into it, they are in it for a while.

To sell those rentals in Europe, we use a massive customer database, to market through catalogues and direct mail to a massive audience.

So, we bring all that together between people who want to rent and those that want to exchange. It then creates a huge network. Say, if you are in the US and own a week in Orlando, and you want to go to the middle of Tuscany in Italy. We may have a rental property there. So you give us your week and we'll put you on to that accommodation and we'll rent your week out.

All these arbitrage opportunities we manage. Essentially, our business size is $1.1 billion. And, we've been seeing double-digit topline growth in the last 5-6 years. It's a good business to be in.

A few words on your job?

This growth is unique to the travel industry. Revenue management in the hotel industry is different from the same in ours. Hotels have a standard supply of rooms for a given period. Then, given a stream of demand, you decide whether you should raise or lower the price and maximise total yield from each of those rooms. They use revpar, an industry term. It is the product of average daily rate with occupancy percentage for that day.

It's a fairly easy problem to handle since your supply is fixed. You cannot overnight create new hotel rooms. In our case, supply is not fixed. It's variable, as is demand. We can go out and source more homeowners. More people can choose to come in and deposit their week to exchange. You have to predict all of that, each year.

That makes it complicated. It's one more variable to predict. You then worry about if demand is greater than supply, you are going to push up prices, else push them down until occupancy goes up.

In our case, when demand goes up, I can take prices up and I can go bring more `inventory', or vacation weeks. So, that's an option for me to grow the business more effectively. Importantly, when supply exceeds demand, a hotel is stuck. Not I.

I just have to stop contracting those inventories. But, making those decisions is complicated. Running a business as large as this, the number of bets you have to take based on analytics becomes that many more. You are taking bets on the supply side and the demand side.

I'll explain how I tie in revenue management and the analytics portfolio. On the exchange side of the business: if you come in and give me a week of your vacation, then I need to understand the value of the week and give you something in exchange, which is equal or lower in value. So, to figure out the value of the week, I have to forecast the demand for your week. We also allow members to deposit a week that starts two years from now. Meaning, they want to give up the use rights to the week they own, not this year, not next year, but two years from now. They can come back to us, two years from that time, to reclaim the week at another location. So that makes it four years from now.

You can come in now and say, I want to deposit the week I have for 2008 in Coorg, but in 2010, you can reclaim that week and use it for an exchange vacation. Obviously, if there were only 100 people and if they all came and surrendered their week in 2008 and reclaimed it in 2010, it wouldn't be so complicated. What I need to do now is to forecast both the demand and the supply for that type of week in our system for that time frame that you are depositing.

Forecasting is the first thing we do in our business. We forecast demand and supply all the way down to a specific resort, for a specific time or week of the year. The weeks of the year are numbered 1 to 52. So, I can tell you that for resort A in Phuket in 2008 in week 43, for a two-bedroom unit, I expect to see this much demand and this much supply.

The valuation is based mostly on historical data, but mixed with judgement, to some degree. We have to take local business knowledge. For instance, when the tsunami struck, we didn't know how long the resorts were going to be unavailable and were under repair. All that matters.

We also use other possible variables. Predicting demand using broader macroeconomic variables. How well is the economy doing? Say, during the hurricane season, last year in the US, that season was very bad. we need to take into account data outside the system. To a large extent, it is historic data.

After forecasting, we look at demand and at supply. If supply is greater than demand, then the value of the week you are giving me is going to be low. Else, high. We ascertain that value, and lock it in. Say, value ranges from 0 to 1,000, we decide on 850 and put it into the account. While I lock it in for you, I take the daily risk because, daily, that value will change. It's like a stock exchange.

On a daily basis, we value all the assets in our system/portfolio.

The third step is allocation.

We look at the demand and supply forecast at a given resort, at the region level: there's Orlando Disney resort and there's the Orlando region, we look at the amenities of the resorts - is it on the beach and so on.

These are constants for most resorts, right? Where then the question of volatility?

Those are constants but 80 per cent of the value is determined by just the forecast, mismatch, etc. Valuation is a relative valuation and not an absolute value given to a specific destination. How that destination compares to everything else within the system - because that's what they have to exchange and supply. That process is called valuation. of taking an individual resort or a region's forecast of demand and supply and comparing it with everybody else and assessing the mismatch. Let's say, for Hawaii, we predict 200 people who want to go there, but there are only 100 weeks available. In Orlando, 200 weeks are available but only 100 want to go there. That's the mismatch. So Hawaii will be ranked higher because of comparison to Orlando - that process is called valuation.

You don't see this kind of thing in the hotel industry. You don't need that.The hotel doesn't exchange anything like that. It only takes cash.

Since value changes everyday, we assess it everyday and class and group it together - very high value, high value, medium value and low value and we'll put your week of Orlando in one of those buckets.

Why everyday? Factors that determine valuation aren't that volatile, are they? This periodicity seems too frequent.

It is. But that doesn't mean that for every resort, the values change everyday, but in some cases, a few values would change.

But these values you calculate today could be for a few months into the future.

Every day, it's calculated for a rolling window of two years. Obviously the values don't change for every resort every day, but they could. You want that flexibility. Today, you may have forecast of supply of 100 weeks and forecast of demand of 200 weeks for the next year in July.

Tomorrow, you would recognise and realise that one of the suppliers is giving you 500 more weeks for July of next year because he is building a new resort. Which means, your supply has suddenly gone up. It brings down the value of that week, in the current week. Also, the farther out it is, less the change right now.

Looking at the next three months, things change dynamically. It does, because, news of a hurricane will affect demand within the next 30 days or 60 days.

Then we allocate where we bucketise everything. Going back to the example of Rs 850, the person giving in the value of Rs 850 will have access to everything from the very high to the very low, for an exchange. On another day, Rs 850 may not have access to very high, but will have access from high downwards. Depending on that, we adjust the system. That's allocation.

Let us go back to the example of 100 people making a deposit and then coming back four years later for an exchange. In order to get them something, we rent the weeks they give us and then we buy stuff when they come back. So, we rent and buy all the time. In order to do that, just like the hotels, we start to do pricing. That part is similar to the hotel basis. That's revenue management.

This requires a lot of technical, analytical expertise. I have got more than a dozen Ph.Ds in the department. Most of them are in Operations Research (OR) or Statistics or Maths or economics. Our outsourced vendor Marketics is a provider of such talent for us. We have a lot of statisticians with Marketics in Bangalore.

On the analytics piece, we have consumer analytics. We have over five million members as part of the rental and exchange businesses.

We have a wealth of data on those members, across rental and exchange businesses. We know the demographics, psychographics and attitudes. We know their transactional behaviours with us in the past. We do a lot of datamining from this.

We build `predictive' or consumer response models. Say, we have a whole lot of inventory (resorts), say we have 5,000 additional weeks for Orlando in September and we want to know who is most likely to go to Orlando or exchange a resort elsewhere for Orlando for a price point. We figure out the probability or likelihood that he'd go based on all the information we have about him through a statistical model. We target anyone with a probability greater than 0.5. This we do with a marketing campaign through e-mail or whatever and back it up with a catalogue or promotion, say $50 off your vacation and send it only to that targeted audience.

The rules would be like, two years ago, in this month, this family went to Orlando.

Correct. But it's more than rules-based. It's actually a predictive model. So we might say, the dependent variable is the likelihood that you'd go.

What factors determine that?

The independent variables could be the number of times you have visited Orlando. The distance between where you live and Orlando, if it's a short-term, 30-day promotion, could be key. We know that airfares are expensive, the closer you get to the start date. Something in the model says: look at the geographic distance between the resort and where the member lives to where the promotion is applicable. All that is taken into account as independent variables in the equation.

When you crunch the numbers and run it, it'd say, "this guy has a probability of 0.63, if so then send him the campaign." That sort of modelling happens. Much of that happens in Bangalore with Marketics.

What is the kind of involvement of Ph.Ds in determining the probability? I understand that software tools and predictive modelling come into play. But these rules keep changing. the factors going into predictive modelling could change anytime. What is the kind of human involvement that takes place?

It's not. What the software allows us to do is to specify the variables in the equation. If you think of demand as a function of price, income and your gender, for instance, you specify those three variables one time. The next time, you might say, demand is a function of income, price and where you live. Who makes that call? The Ph.Ds do this. The software tool then uses data against those variables and runs it. It is only a really large calculator. But it gives us a lot of power and flexibility. It'll give you co-efficient of correlation. It gives you a lot of tools to decide what variables to put in that equation and eases up that process. That's where the tool helps. It crunches numbers and the computer efficiency is obvious.

Is the tool built in-house by RCI?

No. On the analytics side, we use SAS (Statistical Application Software) for all our models. Our Marketics team has access to that and to our database.

Also, we also buy a lot of inventory for our members. We also want to add value to members by giving them inventory in places where they don't have inventory to exchange. Developers of resorts build in only some destinations, usually it's on the beachsides, on the hill, or on a ski mountain.. not in city centres such as London, Paris, New York,. So we go out and acquire hotel inventory, i.e, we purchase accommodation. It's a more recent phenomenon. We don't purchase hotels but do mass bookings of rooms. We use inventory analytics here. This requires OR-related work.

When you spend millions of dollars buying room nights, you want to do it methodically and smartly. So, we have data about room rates - the average daily rates of the area, we have the occupancy of different hotel chains, we also have the revpar. We know the demand from our members. We have fairly non-seasonal demand, meaning, members are willing to go to places such as San Francisco where no one else wants to go.

Hotels build accommodation to service two types of clients: Business and leisure in many of these places. Hotels may have good demand from business clients at certain times of the year. However, during the holiday periods, those hotels go empty, for, business clients are not coming at that time.

We know our members would like to go during holiday seasons to those places. So, we developed an optimisation problem or algorithm. We would figure out the maximum number of rooms at the lowest cost, which satisfies our demand, subject to constraints that our demands impose.

The algorithm is called, in OR terms, a knapsack problem. If a burglar goes into a house, he has a big sack he is carrying, he has different items that he could take away from that house. But he can only take as much as the sack permits and as much as he can carry. He wants to maximise the value of the items in the sack, subject to his constraints of ability to carry, size of individual items and how much the sack can carry. We do pretty much the same kind of analysis.

We know the rates and all that information and we know the demand for a certain place, and we know the value of the demand (how much our members would be willing to pay). I would also know the hotel rates. I'd say `go pick those places and those days of the week when those rates are low, and when my demand is also high, and assemble those in the knapsack.' That's what is getting me the maximum value.

That's an OR problem. We solve that. That's inventory analytics to us. We solve other problems like this, working with our developers.

The two of them put together constitutes the analytics operation and then we have the revenue management operation.

If what you do is high-end kind of value add, why outsource and also, why offshore?

A couple of big reasons. When we started out, we didn't have an analytics department. It's ground-up from two-and-a-half years ago. I still have 120 people in the department globally - analytics and revenue management put together. But only a portion of that is with Marketics, not all of it. Over time, it will build strength. They have between 20-25 people working for us. As a percentage of our total manpower strength too, in addition to absolute growth in numbers, the Marketics team would grow. I am expanding more and more with Marketics than internally.

Capabilities (in India) matter. The reasons why I offshore: It's because I need very high-end skill sets. I can get a Ph.D or master's degree in statistics, mathematics or economics in this country at a much lesser cost than in the US. The differences are very significant for these skill sets. Differences in wages and on the IT and software/tech programming side, even though there are rate differences between the US and India (differences not as large as it was five years ago), you still have a lot of local talent in the US. But the high-end kind of talent is precious. You don't see a lot of people in the US with a masters' in statistics. It's not a hot discipline in the US, the universities don't churn out so many as they do on the computer science front. There is genuine shortfall in that kind of skill and you can take advantage of the availability here.

Second, companies haven't advanced in the area of analytics. It's new even in the US. It's the first time they are starting to recognise the power of the information that they have amassed with all this data and databases. They are starting to extract value now. They do this to improve the process. For instance, the process of order-taking improved when they put the information into the database and set out to collect it.

But little did they realise that that is the wealth of information about their customers. Inventory analytics is new from the standpoint of US companies. Davenport wrote an article in HBR as the first article on analytics and it appeared a couple of months ago. It's new for them as well.

For us, having an entity like Marketics allows us to experiment with Analytics on a low-cost basis. I can hire 15-20 people here in India, put them on a job and show value to my company. If I can do that, I can expand. It is easy to experiment - power of experimentation and power of trial is key. Because of the inexpensive nature of the exercise, the company permitted me to experiment and see what value we get. We made it worth our while.

What is the difference you see, say, for a 25-member team, between the US and here?

Indian wage costs are cheaper - on a fully loaded cost basis - it's savings of between 50 and 60 per cent. That means my budgets are that much less diminished at the end of the day.

Stats and math skills as well as software programming skills required for this job?

It's fused. You can't do this job without knowing statistics, or without knowing SAS or programming. You need to use databases to extract information and then be able to manage databases. You do need knowledge of all this and Marketics has these.

How long have you been with them?

All of the time we have been doing analytics - about two-and-a-half years. We were their first client. We started off with four people from their side. Now, they are 200 or so.

What is the pricing model?

Manpower per hour. We are comfortable with hourly rates. Our work is continuous and ad hoc at the same time. There is a stream of work that is fairly continuous and at any given point in time there are five models we are building, so it's fairly continuous. There is analysis work that we do every now and then - for instance . last year we had a very strong hurricane season - we need to know what was to be the impact on travel patterns this year, same time, because customers are going to be very worried about travelling at that time of the year. That is off-hand analysis. That stream of work cannot be predicted.

One of the key things we need to manage from our side is having a stable set of resources understanding our business and understanding our data. We cannot afford churn in the kind of talent we have at Marketics.

We expect individuals to be with us at least for a period of two years. It takes that level of investment for them to learn the data, learn the business and then apply the statistics knowledge to the data and then go one step ahead and provide business insight. So, if they only come in for three months, there's no point.

If we do it on a project-by-project basis, then, we may not be able to get the same resources and talent pool.

Comfortable with attrition rates now?

It's been a challenge as with any shop. We try to work with Marketics to reduce the churn. We treat our consultants like our employees - talk to them about our strategy - they sit in our offices. We work with Marketics to keep them motivated with a career path, internal to Marketics where they learn and grow. For, these are very ambitious folks who want to keep learning new things. Even though the churn is high - it's not as high as the industry average.

You saw value - but on an annual or periodic basis. What value have they brought? Do you define an error band beyond which they cannot go?

Process metrics are harder to get on the analytics business. On the bread and butter side of that business, it's just building marketing campaign models. Under consumer analytics, I told you how we build models, where we get out the likelihood that a person would do a certain transaction. We have done many marketing campaigns, for which these people build models and we figure who the targets are.

We don't hold Marketics accountable to the marketing campaign response rates, but in general, my company has shown to our CEO that there has been a substantial increase in the response rates - or campaign effectiveness. There are no hard and fast metrics but the value is very palpable.

There are campaign-by-campaign metrics that shows the lift we have received from a given campaign supported by the model over a controlled group. We observe a controlled group that sees no campaign and also observe the campaign group. We measure the difference and have seen that in many cases, campaign groups have gained a considerable lift over the controlled group. That's how the value is palpable.

Also, service level agreements (SLAs) are built on turnaround time. We give business problem statements to Marketics. We measure how long they take to turn it around. It used to take us three months in Bangalore, now it takes less than three weeks to build a model there..that change is because of maturity of the organisation and our contribution.

Together, we found the right processes to put in place.

Your core competence?

Understanding our destinations, where our members like to travel to, and what kind of vacations they want at what prices, and hence bringing the right inventory to allow them to enjoy the vacation.

Domain knowledge with our capability to source within the network and . become an aggregator of demand - that is our core competence.

bharatk@thehindu.co.in

More Stories on : Interview | Outsourcing | Business Models | Resorts & Amusement Parks

Article E-Mail :: Comment :: Syndication :: Printer Friendly Page

Stories in this Section
`Voicing' differences

No fairy to help this Cinderella!

Sprint for the print

Tapping the e-word

`Information Bar' display

Room for opportunity

Unfolding story

`Accent' on trends

Quiz