By David M. Raab
Let’s say you’ve built some really good statistical modeling software. Your product is well regarded and modestly successful, but most potential users have invested years learning the current industry standard product, SAS, and see no compelling reason to switch. What do you do?
Impact! (Unica Technologies, 781/259-5900, www.unica-usa.com) represents one vendor’s solution: Build a related application in a more fluid industry where integration with your core modeling technology provides substantial added value. Then hope, somehow, to make your product stand out.
Unica chose the campaign management industry – one that is certainly more fragmented than statistical software, although it does have some well-established leaders. So success is far from assured.
Still, there is hope. Impact is an unusually capable campaign manager, offering just about every function on most users’ wish lists. Combined with the integrated modeling and a reasonable price, it is worth a look by most potential buyers.
The functionality of Impact starts with data access. The system uses metadata to link to an existing marketing database running on a standard relational database engine. This database is maintained outside of Impact and can have any structure as long as records are related by a standard customer ID.
The metadata hides some of the complexity of the underlying database, stores field descriptions and lets different users interact with views that are tailored to their particular needs. Screens automatically adjust to include new data elements as the metadata is changed. Impact has its own data structures for internally generated information such as campaign details, query logs and derived variables. Promotion history is stored in the main marketing database and can be fed by processes defined with Impact campaigns.
Lists are created with a point-and-shoot interface that provides standard SQL capabilities plus an advanced macro language to support complex user-defined transformations and calculations. Users can browse the selected records, generate statistical profiles or analyze them in cross-tab reports.
Once a list is defined, Impact creates a file with the customer IDs of the selected records and any additional attributes the user has specified. Further analysis and campaign segmentation is done on this file, providing better performance than running queries against the main database. The system uses bitmaps to flag records assigned to different campaign cells, further improving the speed of reporting and analysis.
Campaigns are set up using a flow chart with different icons for standard processes, including list creation, mergers or exclusions involving multiple lists, splitting lists with query statements or sampling, executing scoring models, defining outputs and executing queries to define response. Segments can be nested to split the file any number of times, and different segments can be merged into a single group. Sampling icons can distribute records with random, Nth, sequential or ranked selects and can specify the number or percentage of records in each sample. A built-in calculator helps determine the statistically valid sample size to use for each segment. Alternately, users can assign existing segments to existing campaigns on a dynamic campaign matching screen that shows existing segments and campaigns on side-by-side lists.
Scoring icons can call on existing models, either created with Unica’s own tools or an external product like SAS and apply them to records within a particular segment. Users pay extra for the Unica’s Model 1 software, which automatically creates and compares regression, decision tree and neural net models.
Advanced users can apply Unica’s Data Mining Engine to build and analyze models for themselves. Models can be stored along with the data preparation steps used to prepare their input. These steps will be executed automatically when records are scored within a campaign. But the system will not automatically compare the data used to build a model with the records being scored in a campaign, raising the danger that a model may be applied to an inappropriate set of data.
Optimization icons can use multiple models to find the most profitable mail plan by picking names from multiple lists or picking the best offer for each name. Snapshot icons can extract specified data elements into files or database tables, append records to an existing table, or accumulate values in an existing field such as a lifetime purchase amount. Contact icons control the treatment applied to each segment, including the offer assigned, extract format, tracking and response definition processes, and logging of promotion history.
Execution of each icon can be scheduled independently or a sequence of icons can be scheduled as a group. Users can also specify trigger conditions associated with an icon or assign a time interval between one icon and the next. The system provides standard reports on model results and on profitability at the campaign, segment, offer, and other levels. A calendar lists the promotions scheduled into the future. Users can also create and save their own cross-tab, profile and columnar reports using moderately powerful report writing tools. Reports can be output in spreadsheet or HTML formats. More advanced reports could be created with any standard SQL-based report writer going against the underlying database. This tool would not have access to the Impact metadata, however. An entire campaign can be saved as a template, to use as the basis for future campaigns.
Impact runs on Unix servers and Windows 95/98 or NT workstations. The system can connect directly with Oracle, SQL Server, DB2, Sybase and Informix and to other standard relational databases through ODBC. System security works at the individual or workgroup level and can limit users to specific system functions, data elements, reports or campaigns.
Pricing is based primarily on the number of customers in the underlying marketing database, and to some extent, on the number of users on the system. Costs begin at $95,000 for under 250,000 customers and reaches $750,000 for 15 million customers. Modeling software costs up to $82,500 depending on the functions purchased. Annual maintenance is 18 percent of the initial software license.
Unica was founded 1992 as a provider of data mining software and services. Impact was released in July 1999 and has one live installation plus four beta sites.