Automation leaders across the globe are tuned into the developing power and potential applications of generative AI. But trepidation around the unknowns of this rapidly evolving technology is stopping them from putting this tool to work for themselves, particularly regarding security and compliance. Does that sound like you, too? We’re here to address these concerns and offer guidance to approaching your first generative AI business use case with confidence and success. In this blog we’ll discuss:
- Approaching challenges with generative AI adoption
- Use case possibilities
- Preparing to build your first generative AI business use case
Adoption is the challenge
8 years ago an unfamiliar tech was growing in prominence. There was uncertainty around how it worked, whether it was safe and secure, and whether it threatened existing jobs. That tech was intelligent automation. Fast forward to present-day, we’re observing the same unfamiliarity and uncertainty with generative AI. When speaking to our MVPs about their questions and hesitations regarding generative AI, these were their top concerns:
- There is so much new information on this tech that wrapping your head around how to approach it strategically is overwhelming.
- Security, security, security!
- How will it fit into the existing landscape of an organization? Who will own it and who will own the backend architecture to be able to leverage it appropriately?
- No one has paved the way for generative AI in the organization yet and there is no established template to follow.
- Will upskilling/reskilling team members be required?
All great questions and valid concerns! To that we say—just as you initially worked to have new automation capabilities adopted into your organization, you too should follow the same steps to nurture AI adoption. Maybe it’s been a minute since you’ve been back at the start phase, so let’s review our strategy for adoption:
- Assess the landscape in your organization to understand skill sets internally
- Obtain an executive sponsor to champion your efforts
- Identify the executive or organizational vision and business goals regarding generative AI
- Make a business case for generative AI and clearly understand the business benefit recognized by using it, what you could do now that you couldn’t previously, and expected ROI with the initial use case
- Start a low-risk, low-complexity pilot
Set a strong foundation to bring generative AI into your automation program, then you can grow from there!
Automation + Gen AI use cases
When thinking about generative AI, there are a lot of standalone use cases with this tech. You can open up ChatGPT and do so many things right off the bat. But once you combine generative AI with intelligent automation—which itself is very powerful—you unlock a whole new level of capabilities. Understanding how the two are different as well as complementary is important! Let’s take a closer look.
Some standalone generative AI use cases are:
- Creating content
- Updating content
- Classifying content
- Extracting content from broader data sets
But together with automation + generative AI, some use case possibilities are:
- Interacting with business applications
- Workflow capabilities
- Interacting with different data sources
- Executing communications (e.g. sending Slack or Teams messages)
- Using APIs
- Distributing workload
- Using document automation
Intelligent automation amplifies inherent capabilities of generative AI, and the Automation Success Platform comes into play in terms of integrating Automation Co-Pilot directly into your business applications securely and at scale.
Building a Use Case
As you start to put together a first use case for generative AI and plan to present internally, we want to lend a hand by sharing an outline of key considerations you should have dialed in ahead of your conversations with stakeholders. This will help you be prepared to properly address concerns and frame solutions around generative AI.
- Define a clear objective
- Having clear goals will guide the implementation process.
- Ensure you have the executive buy-in from the start!
- Communicate and quantify benefits to your executive sponsor(s) or functional area(s) you’re serving.
- Understand cost implications
- Cost comes in a variety of forms: development, software/licensing, expertise/talent, data collection, maintenance/support, integration with existing systems, and cost of failure. Have a clear understanding of exactly what costs may be involved with integrating generative AI.
- Different AI providers have different cost structures and not all models are one-size-fits-all. Bear in mind that if you design automations in a way that are configurable, you can make things like prompting configurable, which will make your generative AI use cases easier & cheaper to maintain. Also, using generative AI to help more quickly resolve automation incidents should reduce the overall cost of supporting your automation program.
(We recommend referring to Bot Games Season 4 Challenge #3 on resilience!)
- Address and mitigate security concerns
- Security is the biggest concern most organizations face. The quicker you have a secure environment for interacting with generative AI models, everyone will start to relax about experimenting with it and the learning can begin.
- Do your research before selecting a model. Be certain you know exactly where data is going to live and where it is coming from, as well as if data will be used for re-training. With Google Vertex AI and Microsoft Azure AI, data is not used for re-training.
- Don’t dive in with use cases that use sensitive data from the beginning! Save those for once you’ve gotten your feet wet and better understand implications of using different models.
- Forecast training and enablement
- Almost anyone can go to ChatGPT and start querying for benefit in the work they do everyday. But that isn’t the same as when an automation is trying to make a request of these large language models (LLMs) and interpret those results. Think about how you need to train people to write prompts such that the results can be returned in a structured format, such as an XML file or JSON, and then parsed by an automation.
- New Skill Booster in Pathfinder Academy: Prompt Engineering for Automation Developers
- Understand the impact on user experience. Solicit feedback from users and stakeholders to iteratively improve the models and address any usability concerns.
- Consider how generative AI will integrate with existing systems and processes. Ensure implementation can scale effectively to handle increased demand and align with your IT infrastructure and architecture.
- Almost anyone can go to ChatGPT and start querying for benefit in the work they do everyday. But that isn’t the same as when an automation is trying to make a request of these large language models (LLMs) and interpret those results. Think about how you need to train people to write prompts such that the results can be returned in a structured format, such as an XML file or JSON, and then parsed by an automation.
- Consider the model
- Be aware of different options with models, but at the same time don‘t get into analysis paralysis. We recommend treating the generative AI capability as a pilot. Narrow down the prompt engineering or performance aspect of your selected model and then when that is optimized, treat that generative AI piece as one component in your overall automation.
In a nutshell, remember that the generative AI adoption journey will be a mirror of your intelligent automation adoption journey. You simply need to show your organization why they need to adopt generative AI and clearly explain how you are mitigating risks. Do that and you will surely win people over just as you did with automation.