Purchased Model Price List
|Model Type||Description||Calendar Duration||TMA Fee|
|Response Model||Customer Acquisition||6 weeks||$14,500|
|Retention Model||Customer Attrition||6 weeks||$14,500|
|Retention Model||Time Until Attrition||8 weeks||$17,500|
|Revenue Model||Revenue Dollar Value||8 weeks||$17,500|
|Cross Sell Model||Customer Upgrade Model||6 weeks||$14,500|
and Revenue Model
*If the model type you desire is not listed, please inquire by email or phone for an estimate.
- Estimates include data exploration and deriving new attributes.
- Estimates include one model scoring at the time of model delivery. Additional scoring may be conducted by the client, a data processing vendor, or performed by TMA upon request:
- <= 1,000,000 records: $2,500
- >1,000,000 and <=5,000,000 records: $5,000
- >5,000,000: Contact TMA to assess scoring job
- Client should request scoring 10 days in advance
- TMA will return scored file three business days after receipt of valid file.
- Additional categories for the same data set will require additional models at $9,500 each.
- The need for data cleansing and preprocessing will be assessed prior to modeling assignment. One of five categories will be assigned:
- Minor preprocessing: no extra charge
- Light preprocessing: $1500
- Moderate preprocessing: $4500
- Heavy preprocessing: $9500
- Rejected: Data resources are not presently adequate to reach business objectives
- A model validation exercise to determine whether the model needs to be refreshed is $2,500. A single file with a list of those solicited, the model scores and a flag noting who responded is needed to perform the validation.
- The model may be refreshed at a $5,000 discount from the standard model pricing only if the data dictionary (i.e. file columns) is unchanged.
- Unexpected data issues and extra requirements which could not be unearthed earlier in the project will be identified and written as a separate statement of work.
- Resources will be made readily available to TMA as required to obtain data, understand data codes and dictionaries, approve specification changes, and sign off on deliverables. TMA resource request delays lasting more than three business days will invoke a $500 fee per day, and may disproportionately extend the deliverable timeline.
- Criteria for including or excluding data from scoring will be correctly and fully specified in advance.
- Client will provide a development file via FTP or CD-ROM, comma-separated, with labels in the header row. A separate file in .txt format is preferred. For each file, approximately 5,000 responders and 10,000 non-responders should be provided.
- The client should provide a clear statement of the dependent variable, it’s representation, and desired or required level of precision. It is preferred that the client provide performance metrics and baseline levels of performance from existing models whenever possible.
- Estimates do not include ongoing support after report and delivery. “Wire support” which provides remote consultative support by phone or email may be purchased in 10-hour blocks at $3,250. Customer requests will be documented in a time sheet and effort drawn in 15-minute increments.
- The client is responsible for providing the same variables included in the model development data set at run time. The data will support the target functionality.
- The deliverable will be either coefficients for regression models, or a deployed model (in either PMML, R, C++, Java, Visual Basic or similar) with a report explaining the results, algorithm used, findings, etcetera.
Purchased Model Details
Customer Acquisition Model
Who is most likely to subscribe or respond? Response modeling drives efficient and effective campaign performance while revealing the factors that influence customers to respond. Response models identify individuals most likely to exhibit certain behaviors based on past purchases, demographics, lifestyle and attitudinal data. A score is generated for each prospect that identifies the most likely candidates to respond to a given promotion based on intelligence derived through data mining. Upon reviewing a lift chart of predicted gains by decile or demi-decile, the marketing manager may choose a threshold from which to conduct the campaign. Resources may then be focused on those who elicit the greatest potential for acquisition.
Customer Attrition Model
Which customers are about to leave? Since it is generally accepted that the cost of retaining an existing customer is far less than acquiring a new one, attrition modeling can deliver substantial value. Attrition models are similar in structure to acquisition models, with the objective (or dependent variable) shifted to focus on retention. The attrition model will score and rank customers based on their likelihood to churn, allowing for targeted retention and up-sell efforts.
Time Until Attrition Model
Who is most likely to churn at a given point in the future? Although the Time Until Attrition model sounds like a slight modification or extension to the Customer Attrition model, the model’s structure for forecasting time is quite different. The best way to develop a Customer Attrition model is to pick a point in time: “who will most likely churn six months from now?” The resulting model will produce a score for each customer reflecting the probability of attrition at the chosen time. Alternatively, the model may be designed to predict the actual time of attrition at the expense of accuracy.
Revenue Dollar Value Model
Who will perform just like your most profitable customers? There are numerous ways to view customer value, to include total revenue, profitability, longevity and purchasing frequency. The Revenue Dollar Value model provides the means to identify the most important aspects of customer value from which to segment your existing customers or selectively guide future prospecting. The resulting scores will support targeting the nature of customer interactions for various segments and shift resources and priority to those with the greatest return potential.
Customer Upgrade Model
What is each customer most likely to purchase next? It is typically less expensive and more effective to introduce additional products and services to existing customers than to acquire new ones. The Customer Upgrade Model can produce multiple scores for each customer that will be most receptive to various products based on their buying history and demographic characteristics. Interrelationships across multiple products may be identified and translated into actionable profiles. The Customer Upgrade Model can also provide market intelligence by revealing and ranking multiple influences that drive interest in particular products.
Two-Stage Acquisition & Revenue Model
How can you acquire new customers who will perform like your most profitable? This bundled option extends a package discount for purchasing two of the most complimentary and symbiotic predictive model combinations. Just as it does not make sense to acquire customers who are likely to be unprofitable, it is not desirable to spend marketing resources on potential customers who have a very low propensity to respond.
The two-stage Acquisition and Revenue Model maps the results of the two models into four quadrants: [low response / low dollar], [high response / low dollar], [low response / high dollar] and [high response / high dollar]. Although targeting the [high response / high dollar] segment is a given, choices may be made in extending the campaign to a certain depth into the [high response / low dollar] and [low response / high dollar] categories depending on campaign and business objectives.