Custom GPT - It's Not Technical, It's Tactical

Custom GPTs. It's not technical, it's tactical

Introduction

Two thirds of organisations are now using Large Language Models (LLMs) such as ChatGPT as part of their day-to-day work, at least at experimental level. According to McKinsey's State of AI 2025, more users will continue to move towards piloting, scaling and full-scale deployment.

Without robust controls like enterprise tenant management, clear usage policies, data residency safeguards etc much of this activity is effectively uncontrolled. Teams may be uploading commercially sensitive documents, intellectual property, or code to public LLMs without realising it, potentially creating Non-Disclosure Agreement breaches and compliance failures.

As organisations try to keep pace with this lightning-fast era what safeguards can be put in place now that allow teams to work productively with AI, without introducing unacceptable commercial or operational risk?

Here is an example of how this plays out every day in commercial and operational roles.

A Day in the Life of a Commercial Director 2025

Imagine you are the Commercial Director talking to a Bid Manager about a crucial contract that you are bidding for. The Bid Manager has been tasked with reviewing the tender documents and developing a commercial proposal.

You know the Red Flags, so you ask a specific question.

The Bid Manager quickly opens CoPilot and 'Searches' for a word relating to what you just asked about in the tender document.

No mention of it.

A visible sigh of relief followed by an emphatic: 'No. It doesn't say anything about that'

In your head, you make an expression like Gromit with the side-eyes.

Outwardly, you stay calm. What you actually feel like saying (loudly) is something like…. 'It's not the Wizard of Oz! you need to ACTUALLY read the document!'

But the reality is that in your Organisation there may be only one or two people who will read full documents. Most will turn to AI to summarise them.

Custom GPT Agents

We are living through a transformative era which is moving faster than any other before it. Organisations need to embrace change and move on.

Here is where Custom GPT Agents can be the tool that high performance thrives on but with built-in safety features. Just to be clear, by Custom GPTs, I mean internally governed agents trained on approved knowledge not public, open-ended chat tools.

Custom GPTs can become your Technical Specialist Agents who can:

- Think like you

- See risks & red flags like you do

- Output information in a format that you need it

- Act as Voice of the business and not only an Oracle

The Benefits of Custom Agents over General LLMs

General purpose LLMs can sound confident while being wrong. Confirmation Bias plus failure to deeply analyse information could lead to a spiral of failures and catastrophes because of poor risk management.

They also lack persistent organisational memory unless explicitly designed and governed so frustratingly, there is an element of reinventing the wheel each time you ask it something and you lose the original intent of saving time.

Custom-built agents provide inherent safeguards and organisational memory, making them fundamentally different from open, general-purpose LLMs. In practice, they not only deliver significant efficiency gains, but also reduce risk, minimise errors, and improve quality through consistent decision-making because they are trained to reflect your commercial judgement and standards.

What level of Knowledge is needed?

Firstly, I am not an AI expert. I am a lifelong learner. I learned how to build custom GPTs through an online course and the rest was trial and error using Chat GPT.

Years of front-line operations have given me a deep understanding of Risk, Commercial Judgement and how to structure outputs and workflows. That is why I understand how to build the Custom GPTs. It's not technical, it's tactical.

If you find that you are the most experienced and knowledgeable person in your team, custom GPT is an excellent way of getting all of that know-how out of your head, removing yourself as a bottleneck and freeing up your time to work on better things.

Strategic Oversight

Custom GPTs do require Strategic Oversight and Subject Matter Expert input to optimise them because they expose deeper issues like gaps in current systems and processes. Without in-house expertise, find a solutions partner who can help. There are also new tools being released which are designed to guide users through the process.

How to Build a Custom GPT

Think of a Custom GPT as an assistant rather than an Oracle. Assistants need to be trained. My general method of building GPTs is to use the 'Configure' option in Chat GPT (look for GPTs in the side bar, go to Explore GPTs and + Create). The two main places that need your expert input are 'Instructions' and 'Knowledge'.

Instructions give the Custom GPT its Purpose, Persona, Mandatory Behaviours, Process Steps, Outputs and is the place to build in guard rails to protect things like Intellectual Property, Prevent assumptive behaviour etc.

Knowledge is the place to upload safeguarding information like Red Flags, Commercial information that you want the Agent to consider before it outputs. You can also use Knowledge to store pre-determined information that you want the GPT to output based on Triggers.

Iteration & Experimentation

Custom GPTs need iteration. The iterative process can be done inside the GPT build window 'playground' which allows you to test it before deployment. This is the fun part because you get to see just how powerful it can be and it is the best way to learn.

Some examples (Use Cases)

Custom GPT Agents can be ideal for areas that require technical expertise, where there is a scarcity of skills, for tasks that consume time and are bottlenecks and also to assist processes where there are high risks such as Contract Management, Project Management, Technical Documents, Engineering Reviews, Risk Tracking, Validation Documentation, FMEA etc.

In Summary

Custom GPT agents enable teams to move quickly while embedding safeguards, judgement, and expertise into everyday workflows. When built and governed properly, they unleash advantage and knowledge that would otherwise remain in the hands of a few. Building them is not technically difficult, but it does require tactical savvy. Recognising that opportunity firmly belongs in the realm of leadership.

Imelda Kehoe (Lean Expert) 

https://www.linkedin.com/in/imelda-kehoe/