David Raab Article Archive

Service Bureau Systems

Donnelley Marketing Database Link
Metromail MetroBase for Windows
Direct Marketing Technology Direct Select
Database America DBAbase
Creative Automation Database Query System
Catalog Marketing Services CMS MarkeTrack
MBS Multimode
May & Speh PowerStation
Acxiom MarketGuide
David M. Raab
DM News
January, 1995

These columns have always avoided software that service bureaus make available only to their own clients. The main reason is that choosing a service bureau involves so many factors that it would be misleading to treat database software in isolation. The large number of service bureaus also makes it difficult to cover the topic fairly.

Those problems haven’t gone away. But there are enough interesting products at service bureaus today to make it worth writing about them anyhow–at least to illustrate the range of possibilities. Still, this column comes with a large Warning: Software Alone Does Not Make One Service Bureau Better Than Another.

Disclaimers aside, service bureau systems are particularly interesting. Because they evolve from the requests of many clients and internal staff, these systems have a sophisticated operational perspective sometimes missing from software developed primarily for marketing analysis. For example, key coding, random test samples and seed-name insertion are virtually standard on service bureau products, but have just started to appear on stand-alone marketing databases. On the other hand, service bureau systems tend to have weak reporting and analysis tools, and rarely allow users to change the underlying data.

Equally intriguing is how quickly these systems have developed. Until a few years ago, most service bureaus limited their user-access tools to “quick count” systems designed to tell the number of names available for list rental selects. These systems typically stored precalculated counts for common selection criteria such as state, gender, and recency. Quantities for complex selections were often estimated by scanning a small sample of the actual records or by mathematically combining the results of basic selects. These systems generally had simple, character-based interfaces linked to mainframe computers. Service bureaus often used a different tool to create the actual selection jobs. The selection systems were even harder to operate, since they were designed for highly trained in-house staff.

Today, both types of systems have often been replaced by a Windows-based product that lets users define complex queries with a point-and-shoot graphical interface, view the resulting counts in a two- or three-dimensional matrix, and create sophisticated multi-cell mail plans.

These systems tend to look similar on the surface, but underlying technologies vary greatly. They may run on a stand-alone PC, a central server (PC or Unix), or (most often) on the service bureau mainframe. They may work directly on the main database files, work on a separate system and pass a list of the selected records to the main system for fulfillment, or pass the selection logic to the main system to be executed again. Portions may be batch of interactive. They are often built on a proprietary database–sometimes quite–but others use standard database systems such as Oracle or Adabase. Many are intended primarily to give counts and select names, but some offer calculation, data manipulation and reporting capabilities comparable to the best stand-alone marketing database software.

Here is an attempt to sort out some of the differences. Because of the number of systems, this list will be presented over several columns. Readers are invited to contact me (610-565-8188) with suggestions of other systems to include.

Donnelley Marketing (708-495-1211) has recently introduced Database Link, a complete marketing database system built on a very fast, proprietary engine that Donnelley acquired when it purchased FDC last year. The engine supports multiple tables within the same database and allows full data manipulation including facilities to change data, add records, add fields and add tables without a full file rebuild. The system apparently uses an inverted structure, caching and other techniques (Donnelley won’t discuss the specifics). Query performance can reach 120 million or more records per minute, although speed is reduced as the number of fields and query complexity are increased. Databases are built on the Donnelley mainframe and then transferred to a DEC Alpha server running the VMS operating system. A Unix server is scheduled for 1995.

Users access the system from a Windows PC. They can define queries with a point-and-shoot interface or by typing commands directly in a proprietary query language. Queries are very powerful: they can incorporate multiple fields, include calculations, and do comparisons across and within fields. Users can create roll-up tables (such as states combined into regions) and then use the table categories in queries or reports. Users can also store and reuse complex formulas. By mid-January, the system is scheduled to allow users to create a new field and populate it with calculation results without an external programmer.

The system’s internal reporting capabilities include distributions (which show counts for all values in a single field) and two- and three-way cross tabs. The cross tabs show record counts, as well as totals, averages, and other simple statistics on one or two selected fields. Users cannot group values for reporting within the cross tab, although they can achieve the same result by creating roll-up tables and then reporting on those. Reports can be saved in Excel or Lotus formats, but not transferred directly into live spreadsheets. Users can run reports on-line or submit as many as 500 to run in batch. Users can continue to work in the system while reports or queries execute in the background.

For more complex reporting and analysis, users can access their data through the Pilot Lightship system. Pilot, one of the leading products in its field, provides drill-down data analysis, sophisticated report design, integrated graphics and the ability to distribute predefined views of the data. Donnelly has built interfaces so that Pilot can read the Database Link files directly–no separate database is required.

Database Link also provides specialized functions for promotion planning and list selection. Users can create a promotion selection spreadsheet that includes the queries to select up to 2,000 cells, and then assign keycodes and output formats to each cell. By mid-January, the system should be able to do Nth, random and stratified sample splits as well. Users can instruct the system to tag selected records, create an output file, and maintain a selection history table.

The system will give counts interactively on the promotion spreadsheet, but the actual selections are run by Donnelley production staff. Users transmit them an electronic copy of the spreadsheet; they then review it and submit the selection job (still against the same Database Link file). This is intended as a quality control procedure, since the spreadsheet itself contains the actual selection parameters. Donnelley may eventually allow sophisticated users to submit selection jobs directly.

The selection process runs at about one million records per minute. An entire spreadsheet is processed in a single run, and Database Link will look at the entire sheet to determine the most efficient way to execute it.

There are currently six live installations of Database Link. The largest has about 35 million records with 33 gigabytes of data. Donnelley is targeting the system at databases with more than 2 million records. Clients are given a fixed annual price based on factors including the size of the database, number of users, update frequency, complexity and on-line support level. A simple system with one million records might cost $100,000 to $150,000 per year.

Metromail (708-620-3196) provides clients with MetroBase for Windows, a Windows tool linked to Adabase files on the Metromail mainframe. Adabase is a standard database engine that is frequently used for high-speed query processing. Adabase can be set up with multiple relational tables, but it also can store an array of multiple values (say, several purchase transactions) in a single variable. This is the approach used by MetroBase. Complex queries generally run in under one minute on files with several million records, and frequency distributions appear instantly because Adabase stores the required statistics when indexes are created. The largest MetroBase installation is about 56 million records.

MetroBase offers elegant implementation of a limited set of functions. The system has a point-and-shoot query interface with several nice touches: users can assign their own names to fields and coded value8s; the system estimates the time to run a query before it is actually submitted (and lets the user choose between on-line and batch execution); and each element in a query generates its own count, so users can see the number of records remaining as each restriction is added. However, queries are limited to simple logical operators and do not allow user-defined functions, subqueries or complex relationships among query elements.

Selections can be based on queries or two alternate approaches. The first assumes that a file has already been scored outside of the system. It allows the user to build a financial model based on the response rate associated with each score level, and incorporating user-supplied assumptions about cost and revenue. (Several sets of formulas are available, tailored to different industries.) The user can then graphically move through a gains chart of the ranked records, and see the estimated return rate, profit, return on investment, etc. as the cut-off point changes. The user can save a particular cut-off point, which the system will then translate into an actual selection.

The second technique–which Metromail reports is actually most popular–involves creating a segmentation report, displaying it on the screen, and then highlighting the cells in the report that represent the records to be selected. Again, the system will then automatically create the actual selection statements to find the specified records.

Multiple selections can be combined in a single promotion plan. Users specify how duplicates are to be distributed (in the highest-priority cell only, randomly across cells, or appearing in all cells where they occur). Cells can be further split by Nth selections, and each group can be assigned a keycode, accounting code, seed list, title slug, output format, medium, and shipping instructions. Users can even input the weight and dimensions of the mailing pieces to be used, so that the output can be directly fed into Metromail’s postal sort system and, ultimately, to the lettershop. The selections are automatically executed once they are submitted.

Reporting in MetroBase is done with a flexible and easy-to-use report designer that allows up to nine dimensions: three page breaks, three row breaks and three column breaks. A dimension is usually a single variable (for example, state); once this is selected, the system will automatically report on all unique values of the variable, or let the user combine values into groups. A dimension can also be an existing query statement, such as customers who received a specific promotion or meet a complex set of conditions. In this case, the dimension generates a single page, row or column.

Once the dimensions are defined, users can specify up to ten values to appear in each cell of the report. These typically include a count of the qualified records, but can also report on other fields and include simple formulas such as average revenue per order. The format of the report appears in a preview screen as it is created; the user can change the structure and see the new format immediately.

Although report design is interactive, the actual processing is done as a batch process on the Metromail mainframe. Jobs can be returned in as little as fifteen minutes. Or, up to twenty reports can be run together in a process that might take one or two hours. Report batches can also be saved and rerun monthly or at other regular intervals. The user can do other work in MetroBase while reports are processing.

Once a report is complete, the user can view it on the screen or save it as an Excel, Lotus or ASCII file. The reporting system does not allow users to browse individual records, although Metromail can create custom browse screens tailored to a client’s file structure. The system does not allow calculations to be directly posted onto the database itself (except during the file building process).

There are over fifty MetroBase files in existence. About twenty are accessed directly by Metromail clients, while the rest are used internally by Metromail. Monthly fees are fixed for each client, based on the estimated database size, update frequency and activity level. Costs start at about $100,000 per year, and might be about $250,000 for a client with three to four million records, monthly updates and a typical activity level.

Direct Marketing Technology (708-517-5600) has developed Direct Select, a query and counting product that allows users to load and analyze up to 500 data fields on a Windows PC. The system is built on a proprietary index system that returns counts and other information very quickly, but does not hold actual names and addresses. Response on a five million record database can range from a few seconds for data that has been placed into ranges, to two minutes for a record-by-record file scan. The system currently supports two data levels (such as customer and transaction), which Direct Tech is working to expand to four. Existing files hold over ten million customer records and up to fifty million transactions–generating PC databases of four to eight gigabytes. The files are built on the Direct Tech mainframe, and then passed to clients on tape.

Direct Select provides a point-and-shoot query interface, newly redesigned for ease of use. The interface generates a proprietary query language that can either drive the PC-based engine or Direct Tech’s mainframe-based file system. Direct Tech has purposely kept the front-end and back-end independent, so it can eventually use the same interface to access multiple databases including standard relational systems.

The system provides a powerful ad hoc report writer that lets the user select up to four fields for segmentation: it will create separate cells for all combinations of all values in these fields, or let the user combine some values into one cell. For each cell, the user can see counts or sums of as many as twenty-four fields from the database. The results appear automatically as Excel spreadsheets or can be saved as ASCII files. Users are also typically provided with fifteen to twenty prebuilt report templates, which create Excel spreadsheets as well.

Direct Select allows users to create mail plans with up to 39 segments. Each segment can be defined by an existing query statement, and users can specify whether duplication is allowed between segments. When a segmentation is complete, users can execute it to see counts. But actual selection and label generation is done by Direct Tech staff on their mainframe system, where they handle keycodes, test splits and other operational requirements. Direct Tech is working to add these capabilities to the desktop system, and would then allow users to transmit the mail plan electronically to the mainframe system.

The company is also working to place names and addresses directly on the desktop system and to allow it to store key codes and other calculations on the PC. These capabilities are expected some time in 1995.

Pricing of Direct Select is based on the number of users, plus cost per thousand charges for file creation. Annual fees and updates combined for several million customers with monthly updates would probably be under $20,000. The system is currently used by about 45 clients.

Database America (800-223-7777) has long offered mainframe-based systems for customers who need a heavy-duty marketing database. More recently, the firm has developed a PC-based tool called DBAbase that is aimed primarily at mail preparation and simple reporting.

DBAbase uses a proprietary, read-only database engine that allows users to access data from multiple tables. Simple counts (on a single variable for the entire file) are returned instantly because they are calculated when the database is loaded. Otherwise, speed ranges from two to five million records per minute for a multi-table query on a file with several million records. The largest existing database has nearly six million customer records plus additional records in transaction and promotion history tables, and the system has been tested with up to thirty million records.

System files are built on the Database America mainframe and then loaded onto the user’s PC. Each new copy of the file also contains an updated data dictionary, which allows the PC system to adjust automatically to any changes in the record layout.

The system provides a Windows-based point-and-shoot query interface. It uses a powerful proprietary query language that can select on groups of records from subsidiary tables (such as all customers with three or more orders in the past six months), which is quite difficult in standard SQL. When a user selects a coded field, the system will show all the available values for that field, plus counts for each value–making it easy to know what to select. Queries can be named and saved for reuse, and can also be stored as templates to help build new queries.

Once a query is executed, the qualified records can be evaluated in a two-dimensional cross tab that allows the user to pick a row and column field and then see counts or percentages for each unique combination of row and column values. Results can be viewed as a table or a frequency distribution chart, or moved into an Excel spreadsheet or PC SAS at the push of a button. Users also have the particularly convenient ability to highlight any group of cells within the table, and have the system automatically construct a new query that includes the highlighted records only. This enables users to fine-tune counts and selections without writing new queries directly.

Once all queries are in place, the user can construct a multi-cell mail plan. Each query is called up independently and placed in a row on the plan; the current system allows 499 rows, although this could be easily be expanded. The user can determine whether to allow duplicates across queries or permit each record to be selected just once; can use a specific row as a suppression file; can limit the total quantity in a row; and can split a row into panels by percentages or Nth selects.

Each row or panel can be assigned a key code or the user can segment a panel into different key codes using data in the underlying records. This is done from within the mail plan screen by setting up a small table of key codes and the associated field values. The system can also append a database value (such as a segment score), package code or decoy file to each row.

Once the plan is ready, the system will provide a deduplicated count for each row and provide graph of counts by key code. For the plan as a whole, the user can define printed instructions including output type, postal sorting, ship-to address, shipping method and comments.

Because the PC database does not include names and addresses, it produces a file with the ID’s of the selected records, their key codes and other information. This file is then sent back to Database America where it is matched against the mainframe database to generate the actual labels or tapes.

Each selection can also produce an extract file with the record ID’s and any other information in the database, for modeling or other purposes. In addition, the system stores the selection history of each record in a database on the PC.

Pricing of DBAbase depends on a customer’s other relationships with Database America. Clients already doing substantial volume with the firm would be provided with the system for a design fee of $5,000 to $50,000, based on complexity, plus monthly fees somewhere between $500 and $3,000. Clients without other extensive relationships might expect to pay an additional $50,000 for a software license plus the cost of processing for monthly updates.

The system was introduced internally in early 1994 and installed at client sites later that year. There are currently four client-based installations, plus more within Database America.

Creative Automation (708-449-2800) offers Database Query System, which provides counts and selects records based on attributes extracted from the user’s main database. The system uses a proprietary, inverted file database that runs on the Creative Automation mainframe. Like other inverted systems, the engine can handle data that belongs in multiple logical tables (customers, transactions, promotion history, etc.). Attributes can have yes/no values or hold multiple entries such as codes or numbers, but do not include names and addresses. A system with 500 attributes on five million customer records would take under one gigabyte of storage. The largest system holds about 40 million customer records, although most are between 500,000 and five million customers. The company says databases over 100 million records are possible.

Clients can keep multiple generations of the database on line for comparative analysis. Combining the databases in a single report would currently take custom programming, although Creative Automation is working on reports that will do this automatically for migration and regression analysis.

Performance depends on the number and type of attributes included in a query. A simple query on a yes/no field is returned in ten seconds on an twenty million name file, but complex queries involving multi-value fields take significantly longer. Although files are built outside of the system, queries can permanently append a new attribute (such as a key code) to a record.

The system currently uses a character-based interface to the mainframe. Users write query statements in a proprietary language that can handle complex relationships such as selecting customers based on specific combinations of transactions. While not fully point-and-shoot, the system does provide lists of relationships, attributes and allowable values per attribute, and allows the user to include these in a query without retyping them. A true point-and-shoot Windows version is expected for release by mid-1995.

The Windows version is also expected to provide on-line reporting. Currently, clients can get a handful of standard reports with update statistics and list rental usage summaries. They can also define three-way cross tabs, but these take one to three days to complete because Creative Automation staff must place the values in the desired format and load them onto an Excel or Lotus spreadsheet. The basic cross tab gives a separate cell for each unique combination of the selected variables, although Creative Automation can combine rows or columns with custom programming when it is preparing the report. Each cell can hold either a record count or the sum, average or standard deviation of a selected attribute. Name and address files, or exports of other data, come off the separate master database maintained by Creative Automation.

The current system does offer a full set of mail planning functions. A single query can define a multi-cell plan and automatically eliminates duplicates among cells. In each cell, users can do A/B splits, reduce quantities by Nth selections and assign key codes. The resulting selections produce a list of records with key codes, which Creative Automation uses to create the actual name and address file from the master database. This is generally an overnight process.

The system also keeps a history of all selections, which is stored like any other attribute. This can be used as a segmentation variable or to omit prior selections in list rental fulfillment. For clients with multi-source prospecting databases, the system can keep all records separate and still output a net name file. It can also produce royalty reports that allocate revenues among list owners.

Cost to run the system depends largely on update and output volume. Minimum fees ($10,000 for initial development plus $5,000 per update) are set to discourage users with fewer than several hundred thousand customer names. A client with three to five million names, monthly updates and about sixty million annual output names per year might pay about $100,000 per year. There is no charge to run queries, and custom reports are built for fees based on programming cost.

The system was introduced in 1993 and now has about twenty installations.

Catalog Marketing Services (612-636-6265) looked carefully for a desktop product to offer its clients, but found they would not pay the $50,000 to $100,000 required for most existing systems. Instead, the firm chose to build its own Windows-based system, primarily using the FoxBase database and related FoxFire reporting and query tool. CMS added custom functions for list selection and analysis. It then priced the resulting product, CMS MarkeTrack, at under $10,000 for software and installation. (Processing for updates and outputs is additional.)

MarkeTrack files are built on the CMS mainframe and then copied onto the user’s PC. The system uses separate tables for customers, transactions, products, offers and promotions. Current installations have from 300,000 to 1.5 million accounts and run on the FoxPro engine, but CMS is currently building a 3 million account, 30 million transaction system that will use Sybase on a multi-processor server. The FoxPro front-end can be linked to Sybase without modification.

FoxPro provides very fast performance on simple queries against indexed fields, returning counts virtually instantaneously. The system slows down on tasks involving multiple tables or calculations, but CMS uses its own enhancements and design techniques to complete most processes in a few minutes. For example, a cross tab on a two million name file takes about five minutes.

Most MarkeTrack work is done within the confines of the report writer. Queries, list selections and data extracts are all created as part of a report, although users can get a simple count without actually producing report output. The point-and-shoot query interface creates standard SQL code for simple access to fields in multiple tables, but cannot handle complex tasks such as selections based on summary calculations within the query. Sophisticated users can get around these limits by creating user-defined functions in FoxPro. A skilled user could append a new table to the database, but not add new fields to an existing table.

The system includes some standard reports, reports custom-built for each client, and user-defined reports. The user-defined reports can be record listings, group summaries, and two-way cross tabs. The summary and cross tab reports can use Recency, Frequency and Monetary Value codes that are placed on the records when the database is built. In addition, CMS has created a “dynamic range” function that lets users create ad hoc definitions for these variables. These definitions can be changed at will, although only one set can be active at a time.

Reports can show any field in the database and can show the sum, average, minimum or maximum for individual fields. But the cross tab is limited to a single value per cell, and more complex calculations would require special programming. Output can be viewed on the screen, printed, graphed or exported as spreadsheet, mail merge or database files.

List selection in MarkeTrack is also done through reporting, via a CMS-built option that lets users apply key codes to each summary group or cross tab cell. Users can later reduce the quantity through random, Nth or proportionate sampling, although this can only be applied to an entire selection. To adjust individual cells, users must create them in separate reports, modify them as needed, and then merge them into a single file. This file, which contains key codes and record ID, is then exported to the CMS mainframe to produce name and address labels. Names and addresses are not usually stored in the MarkeTrack database itself.

Copies of the selection files also remain on the system, where they can be used to suppress prior usage from future mailings and to track where each key code was used. Because merge/purge and postal processing may prevent some selected names from actually being mailed, promotion history is typically built outside of the system, by importing the actual mailing tapes.

The system can also create a “frozen” set of master records with details on the customers selected for a particular promotion, to use in future response analysis. The system has functions to match records against transactions with appropriate source codes and date ranges.

The key codes themselves are stored in a system-maintained database that tracks whether they have already been used. The system can automatically create new key codes, but users must manually apply key codes to each cell in a report.

The cost of MarkeTrack includes a $5,000 software fee and database design charges that can be under $5,000 for a simple installation. Processing costs would depend on volume and what other services were already being performed by CMS: if the company already had a database in place, there would be little or no incremental cost to create the MarkeTrack extract. For a company that needed the database built, full services including list cleaning and consolidation processing could reach $2.00 per thousand, or about $25,000 per year for one million customers with monthly updates.

MarkeTrack was released in 1994 and currently has four installations.

MBS/Multimode (516-673-5600) offers a marketing database tailored to its large base of retail and direct marketing clients. Running on the Adabase relational database and residing on the MBS mainframe, the product gives clients access to complete customer records, including names, addresses and full transaction detail. The largest installation has about fifty million transactions on five million customers, with over five gigabytes of data. (MBS is in the process of renaming the system, so it has no formal name as this is written.)

Users access the system through character-based screens with modem or direct connections to the MBS mainframe. (A Windows-based version with a Unix server is expected in the third quarter of 1995.) Queries are written on fill-in-the-blank screens that are custom-built to match the record layout of each client. A single screen can include fields from multiple tables and screens can be linked to define exclusions separately from selections. The systems generally contain presummarized data to speed processing and simplify selections. The query logic itself is limited to simple and/or relationships among fields on a screen.

Response time depends on the nature of the activity. A quick count on an indexed field is generally returned in seconds, and can include basic qualifiers such as date or store. A more complex query is submitted for batch processing and might be returned in an hour. Mailing labels are generally produced in batch jobs that run overnight.

In addition to the main query screens, most installations include a mail planning facility that can define as many as twenty segments on a single screen. These screens still use the fill-in-the-blanks approach, but limit users to the most common fields. Users can assign key codes on the screen and the system will automatically ensure that each name is selected only once. Some systems allow users to do random sampling within each segment, although other systems do not provide this ability because their users do sampling after the output tapes are created. The system builds its promotion history by importing actual mail tapes, rather than assuming that everyone selected is ultimately mailed.

Reports in the system are created for each client, and generally include customer listings, counts, profiles and cross tabs. There is no end-user report writer, although the profile and cross tab functions generally allow users to select the row and column variables from a list of existing fields and can use a query to filter the records included. MBS can custom-program more complex reports as required.

The first installation of the system was in 1990, and MBS now has about 14 versions in production. Initial design and set-up costs from $25,000 to $100,000, and users pay additional volume-based fees for updates and data storage. There is no direct charge for usage.

May & Speh (708-964-1501) offers PowerStation, a licensed version of the MegaPlex Fast Count Engine. This is an extremely flexible and powerful client/server system; the details are described in my review of Fast Count on March 14, 1994. May & Speh has two complete installations of PowerStation, with a third under way. The product is aimed at large databases–the existing May & Speh systems store 20 and 60 gigabytes of data respectively. Installation, including the Fast Count license and design fees, typically costs from $100,000 to $200,000, with data storage and update processing additional.

Note: the following material was never published because Acxiom was unable to provide firm pricing information. Prepared November, 1994.

Acxiom (501-336-1000) uses the name MarketGuide to describe its family of marketing database products. These access Acxiom’s proprietary Relationbase engine, a mainframe-based, multi-table product built to manage very large databases. The largest existing Relationbase system holds 65 million records with about 7,500 bytes per record–nearly 500 gigabytes of raw data–and systems are now being developed with over 100 million records. The MarketGuide tools themselves run on Windows PCs, linked to the Acxiom mainframe directly or by modem.

MarketGuide is primarily built around batch processing. The system has a sophisticated selection tool that allows users to define complex queries with either point-and-shoot or typed entries, handles multi-cell selections, can select individuals based on household characteristics, supports Nth and random samples, and lets the user specify key codes, seed lists, and output formats. These selections are accumulated and then submitted in one large batch job (usually nightly) that might include several thousand individual queries. Typically users submit a job once, check the counts the next day, and then submit again that night for the actual selection.

Acxiom reports that it might take three to four hours to complete the nightly batch job on a database with 30 million records of 1,000 bytes each. The company has recently reengineered the selection process to run on symmetric multi-processing systems, where it is expected to run four to five times faster.

MarketGuide clients who want immediate response can use a “QuickCount” database, a set of indexes built from Relationbase files. QuickCount will yield speeds as high as twenty million records per minute on simple queries, and closer to one million records per minute for MarketGuide reports. These can show counts and percentages for combinations of one, two or three selected variables. The reports always include all values for the selected variables, although users can limit the universe by using a standard query as a filter. A nice feature is that results appear automatically as an Excel spreadsheet.

More sophisticated reporting is generally done outside of the system, using flat-file or multi-table extracts prepared by MarketGuide and fed into third-party reporting and analysis products. Although MarketGuide does not allow calculations directly against the database, a scoring module can apply externally-developed algorithms to the files during the nightly batch runs.

Acxiom is working on a module to allow selects based on mapping information, due some time in early 1995. The company is considering a plan to open its MarketGuide front-end and its RelationBase and QuickCount back-ends to other products by making them ODBC-compliant. This would allow users to access an Oracle file via the MarketGuide tools, or to use a third-party query product against RelationBase data.

There are about 25 MarketGuide installations. Most clients are charged a fixed price based on the modules employed, database size and expected activity levels. The fees range upwards from $100,000, and are in addition to the charges to build and maintain the RelationBase files themselves.

* * *