This paper was written by Valyn Perini, CEO of OpenTravel, and Bonnie Lowell, Specification Manager of OpenTravel.
Larry Smith of Thematix has previously written several excellent posts in Tnooz about semantic technologies in the travel industry. In particular, their potential to create truly personalized offers by adding distinct value to travel offers based on customer data.
At OpenTravel, we’re turning the abstract into the concrete by building a semantic ontology for the car rental industry, including thinking through what it takes to actually plan for a semantically enhanced booking experience, what we call semantic offers.
So I thought we’d share some of the lessons we’ve learned by using a basic online transaction as an example.
I should note that this example is based on the supplier company not knowing who the customer is, so it is not able to present a “personalized” offer, that is, an offer to a known individual. This example is based on non-personal but still actionable data passed from one trading or traffic-referral partner to another.
For all the talk out there, personalized offers in the travel industry are a future phenomenon. It’s much more common currently for partners to exchange information about a transaction, not about a person.
Semantic offer technology: a not-too-distant scenario
Although there are lots of ways to implement this technology, let’s look at a semantically enhanced web booking experience that shows how semantic offers might be integrated into availability processing on a travel supplier’s branded website.
In this example, an airline supplier has partnered with three car rental suppliers, but only one of them, ABC Car Rental, has implemented semantically enhanced services in their booking path.
Ted Smith needs to take a one-day business trip to California but has decided to extend the trip by a week and bring along his wife and children for a family vacation.
Ted starts his trip planning by going to the XYZ Airlines website and entering his dates and airports. Ted reviews flight availability options, makes a booking and selects four seats in proximity of each other.
Included with his online booking confirmation is the option to “add a rental car” to his trip.
Ted needs a rental car so after selecting the option, XYZ Airlines sends an availability request to three of their car rental supplier trading partners to request offers, which the airline will present in a single car rental offer page.
In addition to core information about Ted’s flight, the availability request contains other information that a semantically enhanced system can use during car rental offer processing.
The two car suppliers without semantic systems use the core information to determine availability of their inventory, and their availability results include three classes of vehicle with varying prices.
When ABC Car Rental receives the availability request, their semantic search engine takes charge and looks at all the data in Ted’s reservation request, including the number of tickets booked, the class of the tickets, loyalty program participation and the fact that there are two adults and two children.
An assumption is made that this is a family trip and accordingly their three car offers include an “Ultimate Family Vacation Vehicle”.
Ted looks at the nine rental car offers returned by the three rental car companies and immediately selects and books the “Ultimate Family Vacation Vehicle”, which is a minivan with a family entertainment package – even though this offer has the highest rate.
So how did ABC Car Rental, which basically has the same fleet inventory as the other two car suppliers, create the offer that most appealed to Ted?
Planning for semantic search technology: adapting your inventory to appeal to your customers
A diverse and complex collection of factors influence consumer trip planning and purchasing behavior, and while semantic technology can’t provide a crystal ball for each one of your customers, it can help you sell more product by categorizing and describing your inventory and services in a way that is aligned with your customer travel segments and scenarios.
The process of adapting your inventory to appeal to your customers does not mean that you’ll be making any physical changes to it; instead you’ll have to re-think the features and services associated with the inventory and how they pertain to what you know about your customer and their trip.
Keep in mind that a successful semantic technology implementation needs data – the more the better. For travel suppliers, this can be challenging for two reasons:
- Your data is not neatly packaged in one central repository; bits and pieces of your data are spread throughout disparate data sources and systems
- The majority of information you are storing and maintaining is in an unstructured (or unknown) form to a semantic engine.
To address the first challenge, travel suppliers may want to consider third-party companies that create centralized data warehouses.
The second challenge can be resolved by implementing an ontology, which is a structured way of specifying a variety of data elements (such as rental cars or hotel rooms), and identifying the abstract associations and relationships between them.
Step 1: Understand what you sell and how you sell it
This first step is a great example of why marketing and IT teams should collaborate from the onset of the project. You will need to know both the details about your inventory and the complexities of how the inventory is sold through different sales channels.
- Corporate Purchasing Restrictions: You may be contractually obligated to offer specific types of inventory and services to your corporate clients due to negotiated rate agreements.
- Locality-Based Purchasing Restrictions: You may have limitations on what inventory and services can be sold locally versus internationally.
- Loyalty Program Promotions: You may have a loyalty program that your customers participate in that guarantees upgrades and/or preferred pricing based on program levels.
- Marketing Initiatives: You may have ongoing (time-sensitive) marketing initiatives that offer a specific type of inventory and services at a promotional rate.
- Trading Partner Agreements: Similar to corporate purchasing restrictions, you may be contractually bound to offer specific types of inventory and services to various trading partners.
So why is this exercise necessary? Your semantic engine will need to know about these types of restrictions so the appropriate rules can be applied before, during or after semantic offer processing.
ABC Car Rental IT and marketing teams collaborated to make an “inventory to sales channel model” that described specific data fields that could constrain offer processing by either including or excluding inventory, and in some cases, such as time-sensitive marketing initiatives (perhaps identified by a promotion code), bypassing semantic search processing entirely.
As part of this exercise, they determined a rules engine component capable of invoking semantic processing constraints was required and its basic functional elements should include:
- A GUI-interface to allow the marketing team to enter time-sensitive promotional information
- An automated feed from the sales and marketing system to manage rules associated with corporate accounts and trading partner agreements
- A real-time interface to their loyalty program system to determine benefits associated with various program levels
How did this step affect Ted? He did not enter any information that would have triggered the rules engine component, such as a promotion ID, corporate ID or car rental loyalty program number, so he was eligible for a semantic offer.
Step 2: Determine what you know about your customers
This step is a two-part process that includes:
- Understanding what data you already have about your existing customers and where that data resides
- Understanding what data you can derive about unidentified customers and how you obtain that data.
Let’s assume your company already has some type of data about your existing customers – stored booking and itinerary data, customer profile data and/or loyalty program data, etc.
From a semantic technology perspective, this is known as explicit data because it is based on information that a customer has either given you or on their past purchase behavior.
Your semantic engine will want to know about explicit data for two reasons. The first is that past behavior and purchases can be a great indicator of future behavior and purchases.
The second is customer profiles that include travel preferences are a goldmine of information that can be used when processing a semantic offer.
For example, a customer’s past purchases may indicate they typically fly economy class, but their preferences may show they prefer business class and have opted-in to marketing offers from an airline supplier.
From a semantic offer perspective, a quick check can reveal that when the customer has flown economy class there has been an associated corporate ID in the reservation, indicating their company has a negotiated rate and restrictive travel policy.
From a semantic processing perspective in this example, if the customer has not included a corporate ID in an availability request, no constraints need to be applied to the processing and they would be eligible for a semantic offer.
For the second part of this process, the task at hand involves identifying all the methods your company uses to receive inventory availability requests and what data is contained in the request so you can determine what data may be significant for semantic engine processing.
If you’re using XML or another structured data format, you have a head start as each of your sales channels will be sending data in the same format. From a semantic technology perspective, this is known as implicit data because it is based on dynamic information (versus stored or factual) that an unidentified customer has provided.
Why is this exercise necessary? Your semantic engine will need to know how to weight or influence the information that it receives to create a semantic offer.
Explicit data typically has a higher weight on the semantic offer processing, but in cases where explicit data is not available, the semantic engine needs to rely solely on implicit data, and it is important to the engine to know if some portion of the implicit data is more significant than others.
For this step, the ABC Car Rental IT team:
- Identified all sources of stored (explicit) customer data, including:
- Determining whether or not they had a pre-existing interface to the data or would need to create one
- Creating a consolidated map of all of the available data fields with assigned data field weights between 1 and 3, with 1 being the most significant because these contained the most relevant information that would influence semantic offer processing
They then identified all of their internal and external web services that dynamically request car rental availability, including:
- Creating a data map of each service’s XML data fields
- Assigning data field weights between 1 and 3
As a part of this exercise, the ABC Car Rental IT team needed to create an ontology to help them formally represent each of the explicit and implicit data fields, weights and relationships between them if appropriate.
How did this step affect Ted? He did not enter any information that would have triggered the semantic engine to look for and process explicit data, such as a car rental loyalty program number, so the offers were processed within the ontology using data fields in the availability request passed on from the partner airline.
Step 3: Rethink (and classify) your inventory and services
Now that you know what you sell, how you sell it, and what kinds of customer information you have access to, it’s time for some serious thinking about your inventory and services. This step involves creating “synthetic classes” for your inventory and services.
Although it sounds complicated, it’s basically an exercise to identify what may motivate your customers (or prospective customers) to buy a product, and then map and/or assign new categories (or classes) to your inventory.
Why is this exercise necessary? Your semantic engine wants to think out of the box and present offers based on an abstract collection of information.
For this step, the ABC Car Rental IT team:
- Created new synthetic classes for their inventory that would appeal to their customers personalities and trip goals, including “Power Car”, “Green (or eco-conscious) Car” and “Family Vehicle”
- Identified known inventory and service characteristics (data fields) that would pertain to their new synthetic car classes (such as transmission type, fuel type, mileage per gallon, horsepower, seatbelt quantity, luggage capacity and entertainment packages)
- Mapped these data fields to their existing inventory through their ontology
Step 4: Make the necessary changes to the descriptive information presented to your customers
The last major step is to review how you would describe your inventory and make adjustments to descriptive information to be consistent with your new synthetic classes.
For example, if you have created a synthetic class called “Family Vehicle”, you may want to adjust how you describe this vehicle by including an offer title such as “The Ultimate Family Vacation Vehicle” and include some description of the features that put it into a family vehicle class, such as seating capacity, car seats and other similar amenities, entertainment packages, etc.
Why is this exercise necessary? You created a great semantic engine, but your customer’s perception of and response to your semantic offers will ultimately determine the success of this initiative.
For this step, the ABC Car Rental marketing team created new offer titles and descriptions based on their new synthetic vehicle classes so the customer would see this alternate descriptive information if the semantic engine determined that one of these offers was appropriate to an availability request.
As a part of this exercise, the ABC Car Rental IT team needed to:
- Create a GUI-interface that allowed the marketing team to enter the alternate synthetic class descriptions for their inventory and services
- Update their application presentation (and availability results) logic to include the targeted semantic offers and ensure that they appeared in priority order in availability results
How did this step affect Ted? He was offered three vehicles from ABC Car Rental, with the first vehicle titled “The Ultimate Family Vacation Vehicle”. The other synthetic classes defined by ABC Car Rental did not meet Ted’s criteria, so two standard vehicles were also included in the results.
We’ve used a fairly basic scenario to illustrate these steps but the structure can be applied to almost any type of description and inventory in the travel industry.
The benefits to the consumer are clear – his shopping experience is much better.
The benefits to the travel company are clear – generating higher conversion AND incremental revenue – and customer loyalty.