RAMP is complex and sophisticated in concept and design, but is surprisingly simple to deploy. Designed to layer over existing analytics or customer segmentation processes, RAMP can integrate seamlessly with the industry’s leading Computer Telephony and Interactive Voice Response Vendors. RAMP is designed to ‘plug-into’ most CTI and routing environments including Genesys, Cisco and Nortel.

The solution is SOA (Service Oriented Architecture) based, can run on industry standard infrastructure (e.g. WebSphere, MQ, DB2 and AIX) and is designed to be deployed with minimal operational disruption. Matching algorithms consider location specific attributes such as SLAs, CSR fatigue (defined by utilization relative to peers), and average speed of answer in the center.

The analytics workbench component performs a thorough analysis of the historical customer data, and identifies correlations between key CSR and customer attributes that define success. Additionally, flexible modeling tools are employed allowing attributes to be quickly added or removed, facilitating response to a rapidly changing business environment, such as the addition of new products).

Applying RAMP’s sophisticated analytics to routing means a customer is more likely to be assisted by a CSR that understands their situation. Better understanding generates improved rapport and increases the likelihood of a successful outcome. Demographics, psychographics and historical performance models may be employed to find CSRs with ‘things in common’ with the customer. Overall customer satisfaction noticeably increases.

Successful outcomes also increase CSR confidence levels as they are more likely to get calls from customers they can deal with positively. As CSR performance metrics improve, their days become more predictible and less stressful, and ultimately help to lower center attrition. Natural learning curves mean that CSR’s are more likely to increase their performance over time.

After implementing RAMP at Assurant for a large Financial Service client in a Retention Center, the number of saved calls increased by 49% and the amount of saved fees increased by 119% within the first year. In a separate case study also at Assurant, Sales fees increased by 37% and conversions by 29% with the added bonus that agent attrition dropped 25%.

A significant portion of the benefits are realized as soon as the application is turned on. Basically a call that would have gone to an agent that was not optimally equipped to handle that call with now go to a different, and better matched agent; that change is instantaneous, Beyond that, there is ongoing tuning to ensure that the attributes being used are weighted correctly and maybe even to change the attributes themselves; this tuning would occur as needed. Of course, the system must also be updated as the business environment changes

The data points that the solution uses for the customers and the agents are described as their “attributes” The complete spectrum of attributes is being 'harvested as we go' with the goal of spanning all industries and functional areas (Retention, Collection, Sales etc) that the application is likely to encounter. This list, as well as our experience with similar clients will be leveraged as we perform the data analysis for each client and build their unique data model. The modeling itself then gives us the best correlation between customer and agent attributes to be used in the real time matching. We have a set of metrics including KPIs that we believe demonstrate the performance of the application against stated objectives.

The data analytics will typically build data models using 3 or 4 customer attirbutes and 3 or 4 agent attirbutes. This data will usually come from the billing system, the CRM system and any other generic data sources. The customer attirbutes might include Recurring Revenue, Payment plan, Billing History, Credit and Balance scores / levels, Demographics and Value Segmentation. The agent attributes might include Success Rates, Persistency of success, Revenue increase by customer and Demographics (if available).

The Cognos reporting package is the default reporting solution. This is integrated with the DB2 data repository and can provide a variety of reports that include information on the Match Quality across the agents by customer segment. The match quality is a measure of how close the agent that was selected was to the optimum agent, the bullseye, as defined by the affinity score. This is a good measure of RAMP success but is also affected by the resource levels and call volumes. The real measure of business success will come through reports that show changes in the target KPIs. A series of financial reports will show the changes in revenue which is often the metrics being tracked. Cognos also has the capability of showing individual agent perfromance as a report

RAMP can operate in a virtual, cross center environment. Larger pools of agents definitely improve match quality and therefore the business results. However, we have seen impact with as few as 20 or 30 agents. Just as important to RAMP's success is how busy the center is. It is more difficult to make optimal decisions in a center that is constantly overloaded, where calls are constantly in queue.

RAMP is scalable across multiple (virtual and physical) contact centers. IBM has certified a single instance of RAMP using a single CPU being capable of handling volume equivalent to several thousand agents. RAMP is horizontally scalable as well. Multiple instances of RAMP can inter-operate seamlessly provided they are handling separate traffic and agents.

If customer satisfaction is determined to be an attribute that is modeled, this data will be used. The customer satisfaction measure which must be linked to a particular agent, will be pulled in through the data flow that is established and dependent upon where that data resides (usually an ETL pull or push).

The RAMP matching process is a fast, real time decisioning that happens deep in the Analytics Runtime Engine. Purpose built tools that expose the scoring as the matching occurs are available and show RAMP's matching abilities both in real-time and "in slow motion".

RAMP works in parallel with Skill Based Routing; elevating it to the next level. When RAMP is operational, the attributes / models that describe each of the agents also uniquely define the affinity score for every potential caller. This offers a much richer level of granularity and better insight into what will determine a successful interaction. The caller is routed to a specific agent rather than a group of agents where the final connection is simply determined by which agent has been idle the longest. RAMP seamlessley integrates into existing skills based routing environments from major Call Center providers such as Aspect, Cisco, Genesys and Nortel/Avaya.

Yes. In the outbound world we are reversing the concept so that instead of mapping best fit agent to waiting caller, we are mapping best fit prospect to agent as they become available. The affinity scoring model still holds true.

Yes. RAMP is a relatively "headless" application, in that there are limited components that export a user interface. However, these tools are all developed in Java, which is intrinisically multi-lingual.Consequently, porting RAMP to another single byte or double byte language is a relatively simple process.

Yes - to date RAMP is available in Cisco and Genesys environments. Other environments (e.g. Aspect, Nortel, Siements, etc) require a "Translator" to be built.

RAMP is typically priced per seat, although we are willing to enter into discussions on performance based pricing

The Engagement would start with a professional services team of consultants working on the data analytics effort to build data models. The client team would not need to be heavily engaged until the deployment phase (including UAT).

The optimal population for RAMP is 100 + Agents, so functionally speaking, we can still provide significant gain and lift for smaller agent pools of activity.

RAMP is live at Assurant. We manage multiple contact centers in an outsourced relationship for many large clients, mainly in the Financial and Insurance sectors. RAMP is routing calls to 100+ agents that are performing retention for financial insurance products.

RAMP really has no negative impact to Contact Center agents. Since the agent scoring is performance data based, and does not involve agent profiles, surveys or interviews, the agent population isn't impacted - or even involved - in the creation of the Agent score. The RAMP approach gathers and analyzes historical call result data, as it pertains to an individual customer. Then, through a series of iterative data mining techniques, a performance scoring model is created - based on the performance of each agent for each type of call - or customer action. Once the scoring approach is built, it then runs completely behind the scenes in real time for each call, but is totally transparent to the agent. Further, the RAMP premise is really all about customer call to agent success. In addition to the RAMP's revenue increase value proposition, an underlying benefit is that more agents experience more successful customer interactions and therefore have greater job satisfaction and attritition is measuably reduced. Since the RAMP scoring and matching analytics combination results in higher value customer decisions and outcomes, both customer and agent experience that positive outcome - happier customers, better results, more of the time.

Centers that handle Collections, Sales and Retention as the sweet spot. The main reason is that in these settings we have some really clear data about the transaction. For instance in a retention center we have very finite information on whether the agent saved the caller. If they did was it on the same product or a different product, what was the value of the save and how long did the save then last. This information, as opposed to a customer care center where it might be a more ambiguous 1 to 5 rating of the agents performance, allows us to build very good data models. Also the fact that they are revenue generating also adds to the business case. In general,100 agents is a minimum number, this is only because this will very likely make a solid business case financially. In fact, as soon as RAMP is making an informed decision between even 2 agents success rates will increase.

The data points that the solution uses for the customers and the agents are described as their “attributes” The complete spectrum of attributes is being 'harvested as we go' with the goal of spanning all industries and functional areas (Retention, Collection, Sales etc) that the application is likely to encounter. This list, as well as our experience with similar clients will be leveraged as we perform the data analysis for each client and build their unique data model. The modeling itself then gives us the best correlation between customer and agent attributes to be used in the real time matching. We have a set of metrics including KPIs that we believe demonstrate the performance of the application against stated objectives.

The data analytics will typically build data models using 3 or 4 customer attributes and 3 or 4 agent attributes. This data will usually come from the billing system, the CRM system and any other generic data sources. The customer attributes might include Recurring Revenue, Payment plan, Billing History, Credit and Balance scores / levels, Demographics and Value Segmentation. The agent attributes might include Success Rates, Persistency of success, Revenue increase by customer and Demographics (if available).

The Analytics Workbench which is the tool where all the initial data analytics work is performed, provides environments to create and tune agent / customer models. What-if scenarios can be created and tested in these environments away from the production environment.

Yes, the result of every interaction is interpreted and pulled back into the data used for subsequent interactions. We learn from the outcome of every interaction; future scores reflect that latest knowledge. It is even possible to reflect the results that are occurring in the current shift into the remaining matches. The latter is done in case an agent is having a particularly good, or maybe bad day.

RAMP can operate in a virtual, cross center environment. Larger pools of agents definitely improve match quality and therefore the business results. However, we have seen impact with as few as 20 or 30 agents. Just as important to RAMP's success is how busy the center is. It is more difficult to make optimal decisions in a center that is constantly overloaded, where calls are constantly in queue.

Yes, RAMP has its own database. There is a DB2 instance that captures all of the RAMP metrics and transactional data as RAMP runs. In order to achieve the necessary throughput there is also a high performance in-memory database (Solid) that contains the latest agent and call state metrics. Depending on the environment at each customer there may also be a Data Warehouse for Reporting purposes

There is a very specific skillset required to set up the initial analytical models; these skills will be provided by the RAMP SME team. The ongoing operation can be performed by the existing operational resources with very little overhead and very little training

If customer satisfaction is determined to be an attribute that is modeled, this data will be used. The customer satisfaction measure which must be linked to a particular agent, will be pulled in through the data flow that is established and dependent upon where that data resides (usually an ETL pull or push).

One of the powerful elements of RAMP is that it uniquely analyzes, for each client, the agent and customer attributes that are key to driving successful interactions. There is no predefined attribute set that can be implemented into every installation regardless of the clients unique situation. We will enter the analysis period with a good hypothesis on what the answers is - based on the industry, similar client experience, agent processes etc - however we will complete the implementation with a model that is unique to each client and represents what has historically driven successful customer-agent interactions in their particular situation. Therefore the source of the agent attributes lies in the client's historical data including, most importantly, the agent performance to date. There are no agent questionnaires or personality testing since the models are not built on personal characteristics. As an illustration, for the retention function of a financial institution, we have seen attributes such as agents

The analytics team will need access to all the customer records that the client has. This customer data has to be analyzed alongside the agent performance and transactional data so that the team can build the analytical models. Similar to the agent data, there is no predefined set of attributes for the customer that the team always uses. The RAMP analytics team will do a thorough analysis of the customer data and how the attributes play a role in successful agent interactions. The team will translate this data into analytical models that ultimately will drive a measure of the affinity between an agent and a customer and predict the likelihood of a successful interaction. The success here is obviously very dependent upon the depth of data that the client keeps. If this is insufficient it may be necessary to ask the client to capture additional key data elements for a time period in order that the desired level of modeling can be achieved. As an illustration, for the retention function of a financial institution.

RAMP scoring provides normalized values for customers and for agents which are calculated daily and stored ready for use by the real time components at the time the call is received. Every day during operations as calls arrive at the center, the customer data for each caller is retrieved and a base affinity score is calculated for that customer for each of the agents that are logged in. These baseline affinity scores are driven by the models that the RAMP analytical workbench creates. These baseline scores are further refined by a set of operational parameters including how soon each agent will be finishing their current call (based on AHT), how busy that agent has been (Fatigue), how well the center is performing against it's answer targets (SLA) and how long a caller has been waiting. These are calculated at runtime and impact the routing decision real time. All these factors are used to score each agent against each call and the highest scored match is selected based on analysis of the resulting multidimen

There are a number of steps that need to happen before the asset can start to generate the affinity scores. An analysis of the historical data allows the Analytics team to build a customized model that is based on which attributes are contributing to success and the relationships between those attributes. This model is unique to that client and reflects what the success factors the client wants to impact and what the historical data reveals about how to drive to those factors. These models are then used to build a table of affinity scores based on the latest, and constantly refreshed, data. Each time the solver runs (multiple times per second) it builds a matrix of affinity scores for all the logged in agents and the calls in queue. This matrix is used to generate the individual requests to route a particular caller to a particular agent in real time.

Technical Architecture included in a number of items of available collateral. Contact the RAMP team at Assurant.

The RAMP Administration Workbench allows access to the data elements that are exposed for the client to control. This includes the setting of the agent / supervisor / manager hierarchies used for reporting; the control of the Service Level Agreement (SLA) targets which can be configured at the product level; the switch which controls whether the matching engine (ARE) uses the fatigue of the agent as a matching factor and the setting up of the relationships between products and the numbers being dialed (usually toll free). This control panel will also allow the RAMP administrators to turn RAMP on and off if this is required. A demo of the RAMP Administration Workbench can be provided. This is a web based tool; Authenticated users can access it from their own computers via the intranet or extranet.

RAMP has been designed to work with the majority of commercially available telephony providers / components. Each installation will require some client specific integration work based on pre-built adapters. With a common, widely used IVR, this should be minimal.

No specific CRM package is needed. The solution has been designed to be as invisible to the agent as possible and as minimally disruptive to the center as possible. The desktop will likely be unimpacted by the release and the solution will run alongside the existing telephony infrastructure providing flexibility and ready made redundancy. The asset will come on a preconfigured box with all the required software and will run on that box as a stand alone server. Only the monitoring software needs to run externally to the box.

RAMP has been designed to work with the majority of commercially available telephony providers / components. Each installation will require some client specific integration work based on pre-built adapters. With a common, widely used IVR, this should be minimal.

We do not require "screen pop" integration. RAMP provides it's own dashboarding/reporting regarding it's matching success/lift rates, but is also designed to integrate with existing desktop software as necessary.

We do also need access to customer and agent data - we first build scoring models to determine the optimal matching, and subsequently leverage these in real time to self-learn and continue to recommend the optimal match in real time.