From Novelty to Necessity: Getting Started with ChatGPT Workflows

By George Irish

“We tend to overestimate the effect of a new technology in the short run and underestimate its effect in the long run.”

This insight, by technology futurist Roy Amara during the microcomputer revolution of the late 20th century, sheds some light on where we find ourselves in these early days of the AI ‘revolution’.

It’s been a year since the release of ChatGPT. There is wide public interest in AI, and general agreement that we’re on the edge of an important new technology. But ChatGPT is still more of a novelty than a real tool. It’s cool and impressive, but it hasn’t yet had a big impact on work lives.

It seems we’re still “in the short run”, waiting for AI to deliver on its highly-hyped promise to transform our jobs. But that may start changing as we head into 2024. This year, the rapid pace of AI ‘breakthrough’ announcements will continue, but at the same time, we should also see practical AI applications enter into our day-to-day work.

Office software giants like Microsoft and Google are already introducing AI into their familiar office programs. Copilot for Office 365, Gemini for Google Workspace, and other AI assistants will launch with expensive marketing campaigns and claims to boost our job productivity. But don’t expect your daily work routines to change overnight. AI office assistants are still very new and will be aimed at enterprise commercial clients. They likely won’t fit so neatly into nonprofit operations, particularly in specialized areas like fundraising.

If you’re ready to start working with AI right now (and you should be), you don’t have to wait for AI assistants to land on your organization from the software giants. You can already start building practical AI workflows using the most available and accessible AI tool right now: ChatGPT.

Building AI workflows with ChatGPT

For most nonprofits, the best starting point for working with AI is ChatGPT.

1. It’s low cost: Either free or $20/month for a single ’Plus’ user account.
2. It’s powerful: With a paid ChatGPT Plus account, you get access to GPT-4, the most powerful AI model in the consumer market.
3. It’s adaptable: ChatGPT can perform a wide variety of tasks including text analysis and generation, reading images, text recognition, advanced data calculation, and even programming.
4. It’s easy to use: It has a simple interface that understands commands written in plain English – no programming required.
5. You’re probably already using it: You or your colleagues may already be experimenting with ChatGPT (officially, or unofficially). That’s a great starting point!

ChatGPT is a powerful platform for advanced AI workflows and can become a common interface for exploring many different use cases.

 

 

 

 

Know your ChatGPTs

If you’ve tried out ChatGPT, you’ve likely used the free version. But for building AI workflows, the better option is to upgrade to ChatGPT Plus ($20/month), which gives you access to the more powerful GPT-4 model with its advanced abilities including document upload, web browsing, data analytics, and image generation.

Best tasks for ChatGPT workflows

ChatGPT has a remarkable ability to process texts – even large blocks of text with many pages – and manipulate words and phrases based on their mathematical associations. It’s like bringing the power of Excel spreadsheets to your written documents.

Use cases for ChatGPT can be grouped into three task categories:

1. Document Processing

The best tasks for harnessing ChatGPT’s capabilities are for working with text documents. ChatGPT can analyze and process multiple documents in just a fraction of the time it would take an expert human analyst.

Here are some top knowledge processing tasks that ChatGPT excels at:

Consolidating:
“Combine these monthly reports into a single annual/quarterly report, following the same content structure”

Summarizing:
“Review these reports from our project partners and summarize the top findings on program impact”

Extracting:
“In these survey results, find all the references to positive donor experiences of our matching gift program”

Cross-referencing:
“Review these project reports and match the outcomes to key objectives in our strategic plan.”

2. Data Analysis and Reporting

If you have a paid ChatGPT Plus account, then you can use the Advanced Data Analytics toolset in GPT-4. It’s like having a highly capable data analyst always ready to take on complex data intelligence tasks:

Fundraising:
“Analyze this online donation data and create a summary report showing total gift income, average monthly gift, top 3 daily income totals, and top 3 single gifts.”

Marketing:
“Review these monthly ad spend reports and identify the channels that are providing the lowest lead acquisition cost”

Forecasting:
“Review this data on our past 5 years of face-to-face fundraising results to identify trends, then generate a prediction for the next 2 years.”

Visualizations:
“Create a chart showing our daily spend on Facebook ads compared with daily website visits and donations.”

3. Content Generation

ChatGPT is also a powerful tool for creating text, though this requires some additional caution. Any content generated by ChatGPT should be reviewed by a human editor to ensure accuracy and conformity with your organization’s communications guidelines.

Fundraising appeals:
“Create a draft fundraising email for a fundraising campaign incorporating details from our latest press release.”

Newsletters:
“Use this outline of stories and links for our latest newsletter, and generate a full newsletter draft, following our communications style and content guidelines.”

Donor Communications:
“Using this latest project report, write a draft of an update to donors that highlights recent achievements and includes one personal story from the report.”These use cases have all one thing in common, they ask ChatGPT to consult additional information sources – usually documents or Excel charts. Creating the bridge allowing you to connect ChatGPT to your documents and data files is the key building block to enabling AI workflows using ChatGPT.

It’s all about context

Like any real-world assistant, ChatGPT can only deliver a successful project if it has been given all the information (context) that it needs for the job. That means thinking beyond the simple short prompt that most of us are familiar with: “Name three Taylor Swift songs about break-ups” or “Tell me a joke about iguanas”.

ChatGPT is very good at receiving and following complex instructions. You can get the best results by building your prompt to include as much context as possible about your task.

• What role do you need ChatGPT to perform (a persona or job title)
• What you want it to do
• What it needs to complete the task
• Step-by-step process to follow
• How you want to receive the response

Here’s an example prompt for a summarization and categorization task:

“As a marketing advisor, analyze the survey results from our focus groups and create a report on the most common recommendations and concerns about our upcoming campaign.  I will provide the raw survey data in an Excel file. Identify the most frequently mentioned recommendations and concerns, and then summarize them into a report in order of most to least common. For each item, include a relevant quote directly from the survey data.”

It’s kind of like the step-by-step instructions to build an IKEA shelf. Everything needs to be included – including the Allen key.

But what’s not included in the prompt is the actual survey data. That’s in a separate Excel file, so before ChatGPT can get to work on this task, it needs to be sent this additional document.

How to send documents and data to ChatGPT

The available methods to send additional documents and data files to ChatGPT depend on your account type. The good news is that it’s generally done with familiar methods like ‘cut-and-paste’ or ‘select a file to upload’.

Method One:  Using ChatGPT Free version

Using a free ChatGPT account, the only way to add additional texts to a chat is to cut and paste blocks of text directly into the chat window. This limits the size of text or data that can be sent and is best for tasks needing only limited external documents or data.

Start with a prompt that tells ChatGPT about the task and how you will provide the additional documents.

“Summarize three documents into a single report. I will provide the text from the three documents, one at a time. After you have received the three documents, proceed with the summary.”

ChatGPT will respond with a confirmation such as: “Understood. Please provide the first document.”

You then cut and paste the additional documents into the chat window, one at a time.

ChatGPT has a limited ‘active’ memory, so if you have a lot of documents or long documents, then it might start forgetting the first parts. You may need to segment your tasks into separate pieces and then combine the results at the end.

Method Two:  Using GPT-4

If you have a paid ChatGPT Plus account ($20/month) then you can use the more powerful GPT-4 model which offers an “upload file” button.

 

Using GPT-4, you can prepare your prompt outlining the task and then upload the additional documents before you submit the request. The prompt can be simpler because the files will already be available.

“Summarize these documents into a single report.”

You can upload most common document formats (text, word, excel, PDF, PowerPoint) and GPT-4 will be able to read and analyze the contents. You can load about six documents (depending on their size) that the GPT-4 model will be able to use.

Each time you start a new chat you can re-load the same or different documents, so this approach is very adaptable to a variety of tasks.

Method Three:  Create a Custom GPT

For some common tasks, you’ll find yourself needing to upload the same documents every time you start a new chat – for instance, loading your strategic plan as a reference or providing a style guide before writing a fundraising appeal or newsletter. For these repetitive tasks, you can create a Custom GPT – a new feature that was added to ChatGPT in late 2023.

A Custom GPT allows you to create a library of reference documents and instructions that are automatically loaded each time you start a new chat. Then you only need to provide a prompt outlining the task and the Custom GPT will automatically add the reference documents.

 

“Use this outline to write a newsletter.”
(Your Custom GPT will automatically load and use the style and branding guides that are in its document library)

This is where we can really start to see the promise of ChatGPT-powered workflows, turning time-consuming processing tasks into simple point-click commands.

I’ve used the Custom GPT feature to build a Fundraising Appeal Evaluator with a library of best-practice fundraising guidelines, and a scoring matrix. It only needs a cut/paste or upload of text from a fundraising appeal to generate an evaluation of strengths, areas for improvement, DEI assessment, and an 8-point score based on fundraising best practices.

It’s free to use here:  https://bit.ly/appeal-evaluator  but you will need a ChatGPT Plus account (paid) to try it out.

We’re still very much in the early days of AI, but with new tools like Custom GPTs, we’re getting a glimpse at how our work routines could be significantly transformed by AI. A lot remains to be seen – and learned – as we move into our second year of AI in our workplaces. We’re probably still underestimating its impact in the long run.

 

George Irish is a veteran of strategy, coaching and consulting for AI-powered charity fundraising. He works with Amnesty International Canada and Greenpeace, among other organizations. He writes this column exclusively for each issue of Foundation Magazine.