Explore GPT applications with the GUM model^©

Summary: Explore GPT applications for your organization with our GPT Use Matrix^©. 

ChatGPT uses an extremely versatile language model (GPT) that can be used in many contexts and for a wide variety of tasks.

This ability to understand and generate convincing text on any subject enables GPT to be applied in a wide variety of fields: customer service, content writing, report generation, task automation, systems integration and more.

All these potential applications make it particularly difficult to map out GPT use cases for an organization. With so many possibilities, it's not easy to know where and how GPT can be effectively deployed!

That's why, at MARYLINK, we've developed a model that can be used to represent the use of GPT in different organizational contexts: the GPT Use Matrix^© (GUM^©).

It allows you to visualize GPT's use cases without restricting it to any particular functional scope (HR, sales, innovation, etc.). Which makes it extremely versatile! Like the MARYLINK solution in short... but let's get back to our subject.

Dimensions of the GUM model^©

GUM is based on two dimensions:

Dimension 1: target audience

This dimension represents a typology of stakeholders with whom GPT could interact.

• Internal individuals (employees, management of an organization...)
  GPT can be used to help an organization's employees or management perform certain tasks, such as searching for information or helping to write documents.
• External individuals (customers, experts, partners...)
  GPT can be used to communicate with people outside the organization, such as customers or other stakeholders. For example, GPT can be used as a customer service agent to answer customer queries.
• Technical systems (CRM, ERP...)
  GPT can also be used as an interface between different technical systems to improve their integration and communication, by interacting with other AI or business software (CRM, ERP...).

Dimension 2: Degree of specialization of GPT

GPT can be parameterized to suit your organizational context. This second dimension therefore represents the degree of specificity of the GPT model used.

• Standard GPT
  Corresponds to the standard GPT model. It is used in a wide variety of contexts and is not specifically trained for any one task or domain. It can answer a variety of questions and perform many tasks, just like ChatGPT.
• Framed GPT
  This is a version of GPT that has been adapted for specific tasks or contexts. For example, GPT can be framed to understand the technical vocabulary of an industry, to accentuate expertise in a field, or to restrict its scope to certain topics only.
• GPT extended
  This is a version in which GPT has been fed with data specific to an organization, enabling it to access, exploit and restore part of that organization's knowledge. GPT Extended can be used, for example, to create customer support ChatBots, when fed with the history of after-sales service questions and answers, or to guide experts, when fed with technical documents.

Graphical representation

There are 9 intersections, or 9 fields of application.

Application example

Here's an example of how a cosmetics (let's call it CosmetiX) could use GPT :

Interaction with an external audience

1/ External + standard GPT

CosmetiX customers can query a standard GPT to find out which colors go best with which skin or clothing color. In fact, GPT has a wealth of knowledge about color harmony as standard (because it's relatively well-known, public knowledge).

2/ External + framed GPT

CosmetiX can provide a chatbot to its customers so that GPT responds with a tone that is faithful to the brand's positioning (e.g. using luxury vocabulary, or youth vocabulary, depending on the target...).

3/ External + extended GPT

CosmetiX can provide its customers with a chatbot that knows the composition of the brand's products. It can, for example, provide information on the presence of allergens in a given product.

Interactions with an internal audience

4/ Internal + standard GPT

CosmetiX employees can use a standard GPT to summarize, simplify, clarify or translate the various information shared by other collaborators.

5/ Internal + framed GPT

CosmetiX employees can use a framed GPT to think about certain topics while getting suggestions relevant to their context, without having to specify. For example, they can ask "What are the major trends?" and a framed GPT will give them the trends in the cosmetics industry.

6/ Internal + extended GPT

An employee can ask an extended TPM to quickly exploit the information contained in one or more meeting minutes or in an internal report, by asking, for example, "what are the 3 main areas for development according to the latest COMEX report?". An extended GPT can be aware of the people and expertise of employees, facilitating the discovery of experts within the organization. This is a form of augmented organization chart.

Interactions with other technical systems

GPT can be used to bridge the gap between several applications (business or third-party if they have certain mechanisms, such as APIs):

7/ Other system(s) + standard GPT

A general purpose GPT can be used to format data from another system. For example, it can translate messages from another system (e.g., an alert system designed in English could have its alerts easily translated into French), or format automated reports (e.g., make computer logs easier to read for the uninitiated).

8/ Other system(s) + framed GPT

A framed GPT can be used to humanize another system. Framing allows you to assign a role and specify a context to the AI, which will use it to modify messages received by another system, embellishing them with comments and texts that humanize it. If we take up the idea of making IT logs more accessible, where a standard GPT can format, the framed GPT can write comments and introductions so that the report gives the impression of having been written by a member of the company.

9/ Other system(s) + extended GPT

An extended GPT can retrieve information from other systems and cross-reference it with knowledge or databases. It can also reinject this reworked data into another system. For example, it can retrieve data from an ERP system (the number of products in stock at any given time), and with an internal tool that specifies how much CO² is emitted by the storage of a given type of product, to deduce the CO² cost of inventory. The extended GPT can transmit this amount to another system. For example, to a database that feeds reports.

Conclusion

Towards fully autonomous organizations?

Communication between technical systems (cases 8 and 9) arouses the most fantasies and fears, particularly as regards productivity and job destruction.

Indeed, the automation and blockchain integration of such elaborate AI systems can have a significant impact on existing jobs.

While these concerns are often amplified by pessimistic scenarios of the great replacement of man by machine, the reality will certainly be more nuanced. While some tasks may be automated, new job opportunities may also emerge. But that's another debate. For now, let's just emphasize GPT's usefulness for collective intelligence.

Targeted collective use cases are more rewarding!

The GUM model explores how individuals and GPT can interact to create innovative solutions that maximize added value for organizations and their ecosystem members.

We believe in an approach that fosters human-machine collaboration and enables individuals to use the capabilities of AI systems to collectively accomplish complex tasks, improve efficiency and develop new skills together.

We're convinced that the added value of generative AI is first and foremost to enhance the capabilities of collectives, rather than to replace them!