Automation: “The technique of making a process run automatically” – implemented using platforms like make.com, present in CRM’s and independently created using unique scripts (for a specific task).
The birth of prompting
The most widely known and used form of ‘AI’ are Large Language Models, which are driven by neural networks that map the data and the instructions they have received to the most probable and desired outcome. This is what you see on ChatGPT, Gemini, or CoPilot when you ask for an image, some code, or a solution to a problem. What we can implement in automations is of similar matter.
We have used this at Public Sector Media to map locations and company descriptions to group clients and prospects based upon region and industry sector – this is huge for teams like ours in terms of geographic targeting and highly specific, coordinated mass emails i.e. Police in the West Midlands. Doing this manually is not an intelligent task worth a human’s precious time, and it has saved us tens or hundreds of hours providing intelligent filtering – to improve the quality of the data that we have to work upon As a media company, this is an invaluable asset to our day to day operations, and the value our sales team can extract from our CRM.
LLM solutions extend our capabilities to solve complex, novel problems that are situationally specific. This could be applied in situations ranging from email categorization of types of customers (i.e. CEO, director) to connect team inboxes and decrease reply times or drawing from a range of sources to create high quality customer reports, reading and synthesising whole webpages about companies and organisations in a matter of seconds directed into your CRM about companies and organisations. Recently we used LLM technology to read emails and deciding if they were OOO, or BOUNCED, which then link onto other snowballing operations. This example highlights that if there is a repeating task which requires human intelligence it can likely be aided by an LLM.
As LLM’s improve, the reasoning capabilities of the AI models does too. As we move into GPT 5.2, the results are prime in terms of how often the LLM’s consistency to achieve a desirable outcome from your data and prompt (less hallucination). This makes now the golden time to work with AI and prompting, because it is something that will just keep improving. Without changing the basis of your automation – the exact same data and prompts, new models will do the same task with less costs, and a higher degree of information quality on the output. That is like improving automations without lifting a finger!
AI solutions are not a superpower, but they are getting closer
AI prompts need to be crafted well, and in the examples shown above, should not replace developers in teams. Humans and AI both have their own unique skills – as mentioned, AI excels in operation speed for categorization, whereas humans have the unique gift of creating unique, powerful workflows.
With the arising of AI agents, we can extend the reach of human based workflows. In isolation, human workflows are great for performing a specific task and doing it well: catching errors and managing edge cases. AI agents can decide in what order those workflows are implanted and with what data. By this I mean chaining an email response automation with an email scheduler automation – this is much more beneficial than creating them together, because it enables modularity, which is great for testing and development. However, the real magic is that AI agents can decide their own inputs: when an AI can do something like intelligently respond to a customer based upon the last three responses that they made, that is the superpower of where we are heading, enabling small teams to do more.
Contact PSM, knowing that we have plenty of time for you, and your ideas.