With the use of AI becoming ubiquitous in personal and work lives, understanding how to take advantage of it without it costing an arm and a leg can be beneficial to ensuring that you are maximising the value that you are getting out of it. Using AI can save yourself, or your organisation, a lot of time and effort but used inefficiently can lead to value loss.

💡 Note, we are going to use the term "AI" in this article to refer generally to LLMs, GenAI and other models.

How AI can help your organisation

AI can be used to navigate around tasks that usually would require a large number of man-hours to either complete, or develop a system to complete the task. This can reduce it down to a few minutes with the help of AI. For example:

• Speeding up code development.
• Automating customer service.
• Data analytics.
• Generating marketing copy.
• Product innovation.

All of the use cases above can be used with the currently available tools, such as ChatGPT, Gemini, Copilot and Le Chat.

Understanding the cost of AI

When using commercial products (such as ChatGPT, Gemini, Copilot and Le Chat) to power your AI tasks it's important to understand how (most) products charge for the use of their services. If you are using the standard web or app interface to interact with the AI, your usage is often restricted based on the subscription tier that you have subscribed to.

If you are using an API to access the AI, you'll be charged based on "tokens". "Tokens" are used to measure the data that you are sending to the API and are used to charge users. OpenAI states on their website:

You can think of tokens as pieces of words, where 1,000 tokens is about 750 words.

As of the time of writing, they charge $2.50/1m input tokens. Your input would include your prompt, along with all of the data that you are sending over. For example, the sentence, "Axiom.ai is great!" is about 5 tokens - this can be language dependent, too. You can learn more from OpenAi on their definition of a token in their What are tokens and how to count them (opens new window) documentation.

Prompt engineering

Prompt engineering is the practice of designing and optimising inputs (prompts) to guide AI models to generate desired outputs efficiently and effectively. It's a critical skill for working with LLMs and other generative AI systems. Learning how to craft prompts that not only get the answer that you need, but does so without wasting resources.

Consider the following prompts, both used in ChatGPT using an Axiom.ai automation export as an attachment:

"Can you describe this file?"

vs.

"Can you count the number of steps in the automation, returning the value as a number only."

The first prompt will give us a full rundown of the file, including information on the keys that are contained within the JSON, the steps and details about them. It'll also give you a description of what the automation does. While this would be helpful in certain situations, you may find that you only need a subset of this data - the second step would give a much shorter response, responding simply with "4". Simply defining the output that you wish can be an effective way of ensuring that the prompt is more efficient.

Alternatives

While using the commerical off the shelf products to fulfil your AI needs, there are alternatives if you are looking to maximise the value of AI to your team, lowering costs but sustaining the helpfulness of it.

Local processing

If you are concerned about maximising value for your team, you may need to reconsider which parts of a project really need AI. For example, if you are processing a large amount of data and then performing analysis on it using AI, you may wish to break down the project into smaller parts, such as:

• Data collection
• Data cleaning
• Data analysis

By doing this, you can then determine which parts are more cost efficient to be done by AI and which may be more cost efficient to be done by the team. An approach such as this may work:

• Data collection: this could be handled by browser automation, such as Axiom.ai, or simply data that has been acquired from a data broker.
• Data cleaning: for larger projects, writing a Python script may be more cost efficient than AI, this also gives you much more control over the data that you are going to be inputting in the analysis step.
• Data analysis: this is where AI excels, you've already cleaned your data so it will be easy to feed into the AI to gain accurate insights.

Often, processing data or manipulating data through a local script can be more beneficial than doing it through AI as the scripts are often not overly complex, libraries such as pandas (opens new window) can help make this a simple task within Python, alternatives are available for R. If you do not have access to engineers, it may be worth a once off conversation with an AI to develop a script that could be used for this task.

Tools like Axiom.ai (opens new window) can also help in manipulating data and preparing it for processing by an AI, see our Manipulate Data steps to learn more.

Local LLM

Running AI/LLMs locally can be another alternative for your organisation - models, such as Llama 3.2 (opens new window), can be run locally on your PC to allow access without having to go through the Internet. This has security benefits as your data does not leave your device, however, it's also worth noting that these do not get updated with live data as it does not have access to the Internet.

Using an AI locally completely cuts the costs of accessing an commercial AI - replacing these with the cost of hosting the AI. Your IT department may be able to set up a server running this to use within your organisation, or you may run this on a single workstation. There are many existing tools out there that can be used to enable this, such as Langchain (opens new window), Ollama (opens new window) and LM Studio (opens new window).

Wrapping up

Breaking down your projects and considering what parts actually need to be done using AI is one of the most efficient ways of maximising value with you and your teams use of AI. Weigh the cost of doing a task manually (or using a script) against the cost of doing it with AI - you may be surprised to find out the differences! Running a local LLM is also definitely an option - and a lot more accessible than most people realise, if you have hardware that is powerful enough to run it.

We are still a long way from AI replacing most jobs, but it is still worth noting that there are areas where AI excels - specifically around data analysis. Even internally, we will use this to get insights on data or help with research on blog articles such as this one, though we rarely rely on the generative nature of these tools.