Case 17: Triada, Ltd.
In early 1997, Joe Bugajski, founder and CEO of Triada, Ltd., was trying to decide on the best approach to raise additional financing. Triada is a cutting edge software company specializing in database management tolls for large commercial environments. After ten challenging and difficult years, Triada is poised for explosive growth. To drive this growth, Bugajski must build a Marketing & Sales organization, while continuing to add enhancements to Triada's products. Until now, most of Triada's funding has been internally generated or sourced from several small private offerings, but Bugajski felt a $5 million cash infusion was necessary to speed Triada's growth. A big concern for Bugajski was how an Ann Arbor, Michigan based start up could raise this quantity of money.
"Most people don't understand what they've got in their database. Our challenge is to make our product so simple anyone can understand it (their database)." This is the goal of Triada's products.
Established in 1988, Triada develops, markets and distributes a suite of proprietary software products based upon its patented core technology, Ngram and its proprietary language NAM. These software products are used to transform standard commercial databases into intelligent databases and provide a set of tools for formulating queries, data mining and knowledge discovery. The Ngram transformation dramatically reduces the storage volume of the database without losing its integrity. As a result, these operations can be performed quicker, cheaper, and at a greater level of complexity than permitted using the standard relational model. Triada believes its products deliver a 50-fold price/performance improvement over competing data analysis solutions.
Introduction to Database Systems
A database is an organized collection of related data accessible by possibly many concurrent users. Databases allow the users to organize, update, sort, and report on the information stored within them. Databases are generally separated into application areas. For example, one database may contain Human Resources data; another may contain sales data; another may contain accounting data; and so on. Databases are managed by a Database Management System (DBMS), which is a set of programs that enables the user to store, modify, and extract information from a database. The DBMS is responsible for accessing data; inserting, updating, and deleting data; security; supporting batch and on-line programs; optimizing performance; maximizing availability; acting as an interface to other systems programs; and supporting user interface packages.
There are three traditional types of database management systems: hierarchical, relational, and network. These terms refer to the way a DBMS organizes information internally which can affect how quickly and flexibly you can extract information. Currently, the most popular structure is the relational structure (RDBMS). In a relational database system, data is stored in tables consisting of one or more rows, each table containing the same set of columns. Each table is given a name and called a relation. There can be many relations in one database.
The most common method for extracting information from a RDBMS is called standard query language ("SQL"). SQL is a very structured language allowing for "and", "or", "not", "greater than", "less than", "equal to" and other Boolean operations.
Problems with Relational Databases
The relational model, upon which almost all modern databases are built, has two fundamental problems. First, the transformation of raw data into a "query-friendly" relational model results in a database, which requires far greater storage volume than the original raw data, oftentimes 10 times greater. This results in a relational database too large to store in main memory, and often must be stored in secondary memory, such as on large disks or tapes. This causes performance to be constrained by the rate at which information can be shuttled back and forth between the storage facility and main memory for processing. This task takes time; the greater the complexity of the query, the greater the amount of time it takes to complete. To attack this problem, several companies have developed indexing systems. Indexing systems use pointers within the database to "optimize" the relational database with respect to a predefined set of anticipated queries. Unfortunately, in the event an operator is interested in formulating an ad hoc query, one that was not anticipated but was perhaps stimulated by an interactive session with the database, the query can not benefit from the existing indexing system. To accommodate such a query in the future, the system must be re-indexed. As a result, the relational model requires greater storage disk space, greater main memory, and therefore greater expense for any given level of performance. The second fundamental problem of the relational model is SQL itself. Because SQL is based upon Boolean logic, it is difficult, if not impossible, to formulate more complex queries of a statistical nature, which is the basis for data mining and knowledge discovery.
In 1984, Joe Bugajski was driving down the road, listening to National Public Radio (NPR), when he heard a report on genetic engineering followed by one on Rubik's cube. These reports triggered something in his brain and suddenly neurons and synapses were firing. Bugajski said to himself:
"Aha! Here's the approach we have to take to solve this problem."
The "problem" to be solved was figuring out how the human brain functions. Since he was 11-years old, Joe had promised himself that some day he would.
That evening as Joe arrived at home, he decided it was time to quit his job at Ford and fulfill that promise. Wen he quit, Bugajksi, just hitting his mid-30s, oversaw a staff of 40 in five locations and had a budget of $40 million a year.
"(In 1985,) I holed up in my basement and started working on mathematics. I wondered if I'd lost all touch with reality. The work was so fundamental in its mathematics and its physics, I'd say to my wife, 'I don't really have a clue where this is going. I just pray every day that I can figure out something to do with it.' My goal was to model the brain. If I could model the most powerful tool on the planet, I'd be able to figure out an economic application for it."
After successfully devising a computer code that mimicked brain storage, Bugajski was granted a U.S. patent on software that dramatically increased a computer's ability to compress and store data. In some cases Bugajski's process provided a 200-fold efficiency improvement over current methods. This was the first major step in resolving a growing problem with many large databases.
As far back as 1987, the innovations in hardware were allowing companies to store larger and larger quantities of data. Figuring out how to efficiently store the data and then effectively utilize the data was a growing unsolved problem. System departments for large corporations are swamped with enormous quantities of information that take up millions of dollars worth of tape or discs and thousands of square feet. Companies that have terabytes of information to store – a terabyte is a million megabytes – typically needed 10,000 square feet of floor space and spend up to $5 million on storage. Also, once these massive amounts of data have been stored, it becomes cumbersome to keep track of what has been stored, to identify where it is located and to upload it back into the system.
The first phase of the evolution of Triada focused on the development of marketable systems to compress the disc or tape space needed to store data. This system was originally envisioned to contain both a software and a hardware component designed not only to save money in storage space and equipment, but also to make the data more accessible. The system Bugajski was developing would sell for $2 million and would reduce the storage space to 100 square feet and improve data retrieval time 20-fold. There were further benefits as well. Operators, workstations or even PCs would be able to handle data that before could only be manipulated on costly supercomputers or mainframes.
In 1987, having solved his original task of deciphering the brain's storage mechanism, Bugajski considered the next steps:
"Ford taught me to build things, and that's what I want to do. I love building products . . . I had taken the mathematics as far as they could go. Building the machine was the next step."
Joe together with his two partners, Bob and John Weins, contributed a total of $14,000. The money was used to build and file a patent for "the machine," a prototype circuit simulating neural architecture. In early 1988, Triada, Ltd. was incorporated in Michigan. By mid-1989, Triada raised another $135,000 to fund continued product development.
But, by March of 1990, optimism turned to gloom. Bob Weins, who had been handling the business end of things; doing the books, paying the bills, writing the business plans, and raising the capital, had to quit.
"Bob had gotten to the end of his rope. His resources were tapped out and he had to get a job or his wife was going to make him live somewhere else."
Additionally, Bugajski's savings were nearly depleted, except for a last resort fund of $15,000. With his wife's encouragement, he used that money to pay enough bills to get Triada through the summer. He borrowed on his credit card and sold the family car. He almost quit, but Bugajksi was certain he was within 60 days of the technological breakthrough. A breakthrough that would lead to the practical data compression he needed to make the business take off. Chuck Henderson, director of the Center for High Tech where Triada leased office space, waived his rent for two months:
"Joe was right. If he makes this breakthrough, the rest is history."
In 1990, Joe was introduced to Mike Raymond, an attorney specializing in start-up high-tech firms. "When I first met him, Triada was little more than drawings on the back of a napkin," says Raymond.
"Mike was the first guy I ran into in the professional arena – lawyers, accountants, senior business people– whose first words were not 'You cannot start a business in Michigan', but rather 'To start a business you have to put the capital together and get it rolling.'"
Triada's prospects brightened as Raymond raised $200,000 in seed financing in 1991.
In 1993, Bugajski business strategy still envisioned licensing agreements for the software first, followed by hardware sales as he eventually ported his software with upcoming generations of parallel processors. All of this, however, was contingent upon raising $2 millions of funding for a parallel processor or supercomputer. Triada needed this supercomputer to do the necessary testing to work out the bugs and provide proof that the data compression worked at the 100 gigabyte or terabyte scale, where it became most useful to upper-end users.
"It's a critical time. There's this very critical window of opportunity, where he'll either take off or collapse," said Roy Scruggs of Staffing Innovations in Atlanta, which leases programmers and other systems to major U.S. corporations. "I'm doing all I can to promote it and assure the investors that this technology works. . . This is one of the most impressive technologies I've seen, and I've been in the game for 35 years." Scruggs is now Triada's Director of Applications Engineering.
Raising $2 million proved to be a daunting task for Triada. In 1992, Joe was introduced to a former vice president at Goldman, Sachs & Co. named Gerald Timmis who now led his own investment firm, GC Timmis & Company in Michigan. Mr. Timmis, who had studied mathematics, was intrigued by Triada's technology. He offered to help Triada sort through their financing options. But, even with a very qualified investment banker on the team, venture capital still wasn't an option. Triada had no revenues, no concrete "product" existed, and the firm was located in Michigan, a desert for high tech venture capital. A private placement to angel investors was the most realistic source of funding.
"There's an incredible lack of venture capital in the state. On a per-capita basis, it's one of the lowest in the nation. There is virtually no institutional venture capital in Michigan," said Timmis. "Venture capital is not easy for Joe, for us, or for the investors. It's a long, hard road, but what Joe has is the basis for a billion-dollar company."
One of Triada's problems in raising the much-needed capital was the daunting complexity of Bugajski's technology. "It's so revolutionary, there's a steep learning curve just to get your foot in the door, much less convince people to put their money down," said Raymond. Another problem with raising the $2 million needed was that the window of opportunity for the software Bugajski was developing was closing. Data storage technology was rapidly evolving, bringing the cost of storage down exponentially. By 1993, the opportunity to capitalize on the compression capabilities of Bugajski's algorithms had gone and the company was not able to raise the entire $2 million they had envisioned, instead only receiving $225,000 from angel investors.
The Evolution of Triada
The setbacks Triada experienced from its inception through 1993, however, were a blessing in disguise. The lessons Bugajski and his team learned from these experiences laid the groundwork for the evolution of its products. Triada's value added, they realized, was not the hardware, but the software. According to Mr. Timmis, this was the first of three "epiphanies" in the evolution of Triada. In 1993 the company stripped off the hardware component of its product and concentrated on software.
The second epiphany occurred in 1994 when Triada recognized that their product could be queried like a database. In other words, Triada's product could be queried using the industry standard SQL database language. Mr. Scruggs acknowledged that the technology should not be viewed as a storage product, but rather as an "active database." This revelation changed the focus of Triada's product concept and allowed Triada to focus its product development around a database not around a storage product. What at first seemed an incidental sideline to Bugajski's innovation became ultimately more important than increasing the efficiency by which data was stored. Not only is the data compressed and stored, but it is also serendipitously stored in intelligent patterns that make complex data much easier to understand and query than traditional databases. This resulted in a greatly enhanced query speed, especially for ad hoc, complex queries where speed could be increased by 5 to 100 times.
However, the real benefit of Triada's technology, the third epiphany, is its ability to understand the information content of data. This capability is referred to as knowledge discovery. This includes the capability to read data, understand the content, and feed this information back to the analyst at a level superior to that delivered by data mining tools.
Generally speaking, companies could use Triada's products to produce a competitive edge in four areas:
The culmination of these three epiphanies was to focus Triada on using software solution to address the problems with relational databases. Triada's product provides three practical improvements over existing RDBMS technologies.
Amdahl Corporation, a $2 billion subsidiary of Fujitsu, tested Triada's product running on a more economical Sun Solaris 2.51 workstation versus SQL queries of a RDBMS running on a powerful, multiprocessor, parallel supercomputer. On average, Triada's products answered the queries, which were complex and identical, in 1/7th the time of the latest RDBMS application. Application versions of Triada's product have yielded similar, remarkable results in the areas of:
SQL Accelerator Speeding query solving to a data warehouse for users who must analyze its contents.
Data Quality V Validating data and correcting errors to ensure analysis results are valid.
Warranty claims analysis Surfacing causal chain factors for manufacturers, shortening the time to understanding and fixing deep repair problems.
Market basket analysis Observing patterns indicating buyer purchasing behavior across a wide product base.
Internet usage analysis Analyzing patterns of Website usage revealing client behavior to realize greater service profitability.
Finally, Triada's product is very flexible and works seamlessly with existing data forms, such as RDBMS tables. Therefore, the Triada data engine can be used to overhaul an existing system, can be imbedded into products offered by software vendors who wish to improve the power of their product offerings, or can be used as the foundation for a completely new form of decision support database.
By 1996, Triada had some revenues from its associations with its select customer base. These customers were "test driving" the technology on select database applications. It was a win-win situation for Triada. Not only was the company able to generate revenues, but it was also able to develop its product further in a "real world", hands-on environment. In 1996, Triada was able to raise $1.25M in private placement with approximately 10 accredited investors. For this placement, Triada was valued at $10M.
The Relational Database Market
In 1996, the database market as a whole - including desktop, relational and mainframe databases - grew 15%, with revenues up to $5.7 billion from $4.9 billion in 1995, leading all other application software types. The database systems and services market, which includes the hardware and services sold in association with database software, is 4 to 5 times larger than the software market alone. The dominant players in the relational database market are Oracle, Informix, Sybase, IBM, and Microsoft. Some important segments are discussed below.
The data warehouse concept sprang from the growing competitive need to quickly analyze business information. A data warehouse stores current and historical data from disparate operational systems that business decision-makers need to condense in a single, consolidated system. This makes data readily accessible to the people who need it without interrupting on-line operational workloads. For example, a user of the Wal-Mart data warehouse could access and query a single database drawing information from Wal-Marts all across the country.
Data mining is what most organizations aspire to when they build a data warehouse. A data warehouse stores large quantities of data by specific categories so it can be easily retrieved, interpreted, and sorted by users. Those customers also want to sift through that data with mining tools. Data mining, through the use of specific algorithms or search engines, attempts to source out discernible patterns and trends in the data, infer rules from these patterns, and predict future values or behaviors. With these rules or functions, the user is then able to support, review and examine decisions.
Companies are beginning to realize that their most valuable asset is the information they possess in their databases. Competitiveness increasingly depends on the quality of the information obtained which in turn improves the quality of decision-making going forward. The ability to improve their knowledge of customers and markets will enable businesses to better target their products and services. The Gartner Group estimates that by the end of 1997, approximately 80% of the Global 2000 companies will have or will be planning a data warehouse strategy that likely will incorporate data mining.
Triada's Business Strategy
Triada's objective is to unify and dominate the fields of data warehousing, data mining and knowledge discovery. Their strategy for achieving this objective is to focus initially on a single vertical market, the market for manufacturing product quality ("MPQ"). Sales of Ngram and related consulting services into this market will be on a direct basis. Triada intends to demonstrate its capability of achieving its larger objective by dominating this market. They will depend upon value-added resellers ("VARs") and original equipment manufacturers ("OEMs") to penetrate other vertical markets such as financial services, insurance, healthcare, utilities and retail. The Company also intends to develop a complete data warehouse solution through its alliance with Amdahl (a subsidiary of Fujitsu).
Triada's Vision & Mission
"We envision the evolution of the computer from a digital file cabinet to a functioning agent of the organization – one that can see relationships, draw conclusions, and take actions."
"Our mission is to deploy Triada's technology for knowledge delivery, helping global 2000 companies solve information analysis problems with Ngram products and professional services."
Triada Database Management Value Chain
Setting up the database for querying
Companies collect a wide array of data and store them in a large transactional database. Oracle, Sybase, Informix, Microsoft and Red Brick Systems provide these types of relational database software and storage. The Triada value chain requires this data to be transformed into a reduced size intelligent database for Ngram queries. Because each set of databases is different, and often from multiple sources, the conversion software may need to be developed for multiple databases. Currently, these programs are written to specifications for each customer by system integrators.
The relational database frequently requires a large amount of expensive storage space. Conversion of this large database into an intelligent database dramatically reduces the size and storage costs of the data.
A query begins with the end-user communicating a query through the application-specific GUI. These commands are then interpreted by the middleware into query instructions. The NAM Application Programmer Interface converts the query instructions from the middleware into an executable Ngram command. The Ngram data engine software then executes queries with the results then being returned to the application GUI. Triada delivers greatly enhanced query speed, especially for ad hoc, complex queries where speed can increase by 10 to 100 times.
Triada has developed some application specific middleware and GUIs. There is an opportunity to license Triada's technology to companies who can develop application specific middleware and GUI for a variety of industries.
One way to establish a beachhead in the MQD market is to provide value-added analysis of warranty data. Triada estimates that in 1996, $40 billion in warranty claims were processed and $400 million was spent on software to analyze and process such claims. However, these estimates may understate the potential size of this market because manufacturers are only now looking for analytical tools, which can help them eliminate defects more quickly. It is estimated that warranty problems represent 5 to 10% of total sales for large durable goods manufacturers. The small decreases in defect elimination cycle time that Triada could make possible would lead to large cost savings for manufacturers. Triada selected this market for its initial entry for the following reasons:
Triada Product Scope:
Triada intends to develop a complete end-to-end data warehouse solution (See Exhibit 1 - Triada Database Management Value Chain). In 1995, Triada developed a working relationship with the Amdahl Corporation. Working with Amdahl allowed Triada to "test drive" its product concept, as well as provide a much-needed source of revenue. Triada is also in active discussions with several other potential strategic partners. They are investigating partners with best-in-class capabilities in areas outside the company's core focus including data warehouse design to build, legacy database interface user tools and a graphical user interface ("GUI").
An example of one of these potential partners are Sagent, which provides Data Warehouse middleware, software capable of loading data from disparate sources (including legacy databases) to Ngram TRANSFORM-DB, the database engine. That is, Sagent's software and Ngram TRANSFORM-DB could be used together to build the Data Warehouse. Sagent also provides a flexible interface that allows multiple users to graphically formulate queries of the data in the Data Mart and such queries need not be SQL. That is, the interface allows for NAM-based queries. Other potential partners are data mining tool providers with established customer bases but limited product functionality. Examples of such companies are Angoss and ISL, and SAS.
Triada's first step in developing a data warehouse is code named Ngram LIGHTNING which was developed jointly with Amdahl and is being released in August 1997. Ngram LIGHTNING is an Ngram database, which emulates an RDBMS, runs on a Windows NT platform and can be queried using SQL commands.
Customers and Applications
To date, Triada has provided software and services to customers in a variety of industries including automotive, financial services, hi-tech, and data management.
Ngram Database products can be used in many applications, including the following representative examples:
LAM (Large Automotive Manufacturer)
LAM built a multi-million dollar warranty data warehouse that includes mainframe and super server hardware as well as an RDBMS system. This warehouse receives information about thousands of claims daily from dealers and others around the country, which the company needs to analyze to provide guidelines to product engineering about process and design improvements. LAM was unable to run an ad hoc query involving two large tables. It spent millions more on even more powerful hardware and sophisticated software which allowed them to run the query in 45 minutes. Using Triada technology, LAM was able to run this same query on a PC in 90 seconds.
Amdahl was looking for a way to help its customers who complained that their hardware running RDBMS technology was not good enough to deal with queries involving two or more large tables. Amdahl tried 'best of breed' data warehousing solutions such as Red Brick, but these made no significant impact on the main problems they were facing. Their customers had problems with load times, query times, system costs (several $ millions), the need for a lot of high technology consulting, large memory and processing/storage requirements, as well as with the inability to update the database without a complete reload. Amdahl found that Triada's technology could deal with all these problems in a highly cost effective manner by building a hierarchically organized associative memory structure from the input data. This allowed a user to actually do the query and do it quickly without a lot of consultants, update the data without a reload, and use low cost hardware such as PCs as well as see important relationships in the data. As a result Amdahl is incorporating Triada technology in its proposed warehousing solution to its customers.
FEM (Farm Equipment Manufacturer)
FEM had a warranty database for its tractors and other heavy equipment, but it had not been able to establish the reasons for its product failure from this database using existing analysis systems. Triada's technology allowed FEM to establish a causal chain in their warranty data that led to the identification of a faulty manufacturing process. Triada's technology also demonstrated to FEM that by using Triada's technology and data transformation, its storage requirement fell by a factor of 16.
BSC (Banking Services Company)
BSC had designed a multimillion dollar data warehouse, with conventional mainframe and RDBMS technology, to analyze information supplied by its banking and credit card company customers on the entire US population (250 million people - 2.9 billion records) in order to provide them with information on credit risk and market prospects. This design anticipated turn-around time on a query to be approximately 2 weeks. Triada has demonstrated to BSC that it can query such a warehouse in under an hour with Triada software using an NT workstation, as well as allow BSC to see previously unseen relationships in the data, while shrinking the database by a factor of 10-16.
Intellectual Property, Patents and Other Proprietary Rights
Triada currently holds three U.S. patents with seven additional U.S. and foreign patents pending. These patents cover the hardware aspects of Triada's proprietary data-compression technology. Triada believes that these three issued patents provide them with broad protection of its proprietary fundamental technology, both in terms of apparatus and processes of use.
Sybase develops client/server and Internet software and services for the online transaction processing, data warehousing, mass deployment, and online electronic commerce markets. Software products include middleware for data access and connectivity, as well as database servers for online transaction processing, mixed workload and Internet applications. It also offers data warehouse servers, middleware, and development tools. 1996 revenues were $1 billion.
Red Brick Systems develops relational database products and services for warehouse applications. Products built for query processing rather than online transaction processing. Their major product is a relational database designed for data warehouse, data mart, data mining, online analytical processing, and database marketing application. 1996 revenues were $36M.
CrossZ Software is a medium size software company specializing in fractal data compression. Their software can be used to create fractal views of large databases in a highly compressed format. Queries against the compressed databases can be run on a PC instead of a mainframe or a supercomputer. Cross/Z has an established customer base among Fortune 500 companies which use fractal compression software for their decision support systems.
Arbor/Essbase Software develops client/server software for online analytical processing. Its major product is an online analytical processing (OLAP) server optimized for business planning, analysis and management reporting applications. Server supports multi-user read-and-write access, large-scale data capacity, analytical, analytical calculations, and flexible data navigations and it enables delivery of OLAP applications directly from operational systems or as data within an overall data warehousing architecture. 1996 revenues were $47.4M.
Oracle develops information management software. Products include a relational database engine and a universal server. In March 1997 it announced a beta release of its next generation object-relational business platform for high-speed transactions, decision support and network computing applications. 1996 revenues were $5.2 billion.
By 1997, Triada was on the move. Things were going well and a marketable product was in development with an anticipated delivery date of early 1998. Exhibit 2 shows the projected forecasts for Triada through the year 2000. While the future looked promising for Triada many questions remained unanswered. What was the right way to grow the company further? With so many potential applications and industries to exploit, what would be the best way to ensure that Ngram becomes the de facto industry standard? Should Triada license its technology to competitors? What type of strategic alliances would be most beneficial and with whom? Which elements of the value chain should Triada deliver? Should they provide an entire data warehousing solution or would that restrict the company's ability to focus on their core product? How would the company be able to attract and retain both the technical and managerial talent to properly grow this company in the Midwest when most of the required talent pool was in Silicon Valley? All of these issues were on Joe Bugajski's mind as he looked forward and decided which road to follow in order to continue growing and developing his vision, Triada, Lt.
Back to the Menu