Quality Control

Once a product is developed, the intended flavor profile must be maintained so that repeat purchases are made by targeted consumers. This can be especially challenging with coffee, since it naturally varies year to year and throughout the year. If the flavor profile has been developed in enough detail, the most important aspects of flavor that a consumer will expect to find can be maintained with reasonable consistency.

Besides having the preferred and expected flavors, a large part of quality control is to ensure that defective or poor flavors do not occur. Poor handling at any point in the process, from harvest to roast to packaging, can cause certain types of defective flavors to result. Good quality control programs have in place the following parameters.

  • Well-defined and achievable specifications

  • A statistically valid sampling plan that accurately reflects the overall quality and size of production

  • Action to be taken if standards are not met

  • A record keeping system of results so that trends and other useful information can be evaluated

Specifications

Coffee specifications should be written in measurable terms. While this is more difficult with sensory specifications, intended levels of various flavor attributes, what should definitely be perceived in the coffee, should be defined and compared with standards. Typical coffee specifications include:

  • Green coffee: physical specifications (defect inventory, total moisture, water activity, bean size distribution, density) and sensory aspects (free of taints and faults, specific flavor attributes). If a blend of different coffees is used, specifications for each component should be written and substitute components may be specified.

  • Roasted coffee: physical specifications (total moisture, level of roast), production specifications (time of roast at different stages as well as final roast level), and sensory specifications, including those flavor attributes to be developed as the result of roast

  • Ground coffee: grind particle size distribution.

  • Freshness: water activity of coffee before packaging and percentage of allowable oxygen in package; amount of allowable time between production and packaging

Required level of sampling

To obtain an idea of what the consumer is most likely to experience, a statistically valid level of sampling for testing must be in place. A single test of a single roast does not necessarily reflect the overall ongoing quality of the product. A random sampling of green coffee that is received should be compared to the initial purchase sample. If a quality control program is being initialized, it is a good idea to over-test at the beginning to see where the most variation occurs. Gradually, a more efficient system can be developed.

Action to be taken if specifications are not met

Inevitably, something goes wrong in the process. The reputation of the roaster will suffer if poor or inadequate product goes out, but some actions can be taken to minimize the damage to the bottom line.

  • Isolate the offending cause. Is it just one batch, or just one bag of green coffee? Is it a systematic occurrence (poorly maintained roaster machinery that is no longer controllable, for example) or a one time occurrence?

  • Use in another product. If a coffee is over-roasted, for example, it may be possible to blend it in with another darker roasted product, if the flavor requirements of the blend allow such a mix.

  • Donation to charity. Some coffee may not be what was intended, but still be consumable.

Record keeping

If a result is used only once (to determine whether or not specifications have been met), the time and expense of testing is not being efficiently used. Regular evaluation of testing data can reveal useful information and trends that can be used to streamline production (including the testing program) and determine where resources should be applied.

A relational database is most useful for storing results. This allows the user to design queries over specific time periods and relate different tests (physical and sensorial) on a particular batch or product to one another. A time consuming aspect of this is data entry, but direct electronic entry of some results is possible in some cases. Even sensorial testing can be streamlined by having tasters enter their results directly on a computer touch screen and downloading results.

Some trends reveal themselves only through evaluation of large amounts of verifiable data. Data mining techniques search for pre-defined patterns and relationships. Depending on the technique employed, data can be evaluated in terms of time (perhaps showing seasonal trends), how different production aspects are affecting the product, or even for predicting a likely outcome given certain parameters.

More specific discussions of these aspects can be found in the white paper link below (Designing Quality Control Programs for Coffee Products).