With the explosion of custom GPTs and AI-powered tools, it’s tempting for businesses to jump on the bandwagon as quickly as possible to say they’ve joined the future. But at Flock, we know that true marketing transformation isn’t about expanding tech stacks for the sake thereof, it’s about purposeful innovation bent on driving mission and vision forward. Building a custom AI model for your company that adds substantial and measurable value requires asking the right questions and executing the best possible strategy. In this post, we’ll unpack the strategy and five essential questions your marketing team need to answer before launching your next (or first!) model.
History of the Custom Model
In November 2023, OpenAI released a ChatGPT update allowing users with a Pro, Business or Enterprise membership to be able to create their own custom LLM models for personal use or to publish to the GPT store. This update saw a surge of customized models spanning over many uses, today the GPT store houses up to 3 million models. As great as this is for innovation and collective sharing of tech, there is now a model fatigue, causing weaker models to be at face-value indistinguishable from useful models due to the sheer quantity and overwhelming market share. Other big players in the AI game have since followed suit, with Google Vertex, Anthropic, Microsoft Co-Pilot and Minstral AI offering similar services where users can build their own LLM models, chatbots or agents, however, ChatGPT offers the highest degree of customization at this current point in time.
Many marketers today may ask themselves the same questions, such as when should they build a custom bot vs using a currently existing model (such as a GPT 4.5 or Gemini 2.5) or even outsource bot creation to external agencies and companies. These decisions can typically be answered against 3 key decision-making questions:
- Do we have the resources to build our own custom bot? (appropriate subscriptions to technology, in-house expertise and company infrastructure)
- Is it more cost effective to build our own model vs outsourcing or using a generic model? (generic model will be cheapest, but least efficient due to lack of company specific data training, outsourcing will likely be most expensive, but also the most reliable/safest option if you are outsourcing to reputable/knowledgeable companies/agencies)
- When evaluating the needs of my brand, what model do I believe will help me boost my brand the most effectively? (Can our marketing team prompt generic AI models well enough to receive high quality output, would we be able to brief an agency on the needs of our brand to build a bot in enough detail for a good model?)
Custom bots are the most hyper-targeted method to build out effective brand campaigns within your company, as you have full control of the data the bot is trained on, and the type of output you will receive. Let’s investigate the process of determining the build structure of your model:
Anyone with a GPT Pro (or subscription to another service) membership can create their own model, but to set your business apart from the rest and build a truly efficient model which lessens your pain points and bolsters your strengths, there are considerations you should review before embarking on this AI powered journey.
Refining your objective – 5 Critical Discovery Questions to build a bot to boost your brand
Here are the 5 key questions you should be asking your team when structuring a new custom AI brand brain within your marketing operations:
- What is the purpose of this model? – Determine specifically how this model will solve a problem, change your marketing processes, Is it for innovation, better briefing, testing, idea gen, etc?
- What data will this model be trained on? – An AI model is only as good as the data it is trained on, ensure your data is clean, fit for intention, free of any harmful bias and high value.
- Who will use this model? – Who in your organization will be using this model and what is their level of expertise on using GPT models? What are their biggest pain points they want this model to solve? What security measures are in place to ensure the data and output is secure and stays within the boundaries of your companies’ digital landscape?
- What is the ideal output of this model? – Consider what the model is presenting to the user, is it a one, two, three step output? Is the content text heavy or presented differently (PPT, Excel, tables etc)
- How will this model be evaluated? – Ensure you have baselines in place to evaluate your model to ensure it is still fit for purpose, minimizing hallucinations and providing high quality output.
Bonus question! Change Management: Change management plays an important part when implementing new ways of working across a team, or organization. You’ll need to plan ahead for training sessions, team check-ins and evaluations to ensure the use of a custom model within the marketing team is as seamless as possible.
Too many people and businesses rush out models without determining these questions to stay ahead of the curve or to boast to their clients and competitors that they are “AI integrated”. If your custom interface isn’t serving your team well, what’s the point? I would love to have a fancy car, but would it get me to Manhattan faster than the Subway? Probably not.
Instead, take the time to invest into your company or team’s reasons for having a custom AI model and ensure the performance magnifies your capabilities.
Once your marketing team has collectively answered these questions in detail, it’s time to move forward and put the model into development. At this stage, it is crucial you document everything you are doing, vigorously test your model, and ensure all training material is stored somewhere your team can easily access and in an organized manner (with name parameters!). This allows for better transparency to those who will have questions about the model as well as from an evaluation perspective.
Once you are happy with the first draft of the model, release it to a dedicated testing team who have been briefed on the above 5 points. This permits a closed feedback loop and fast solutions for potential issues. Once your model has reached final approval, you’re ready to go!
Now you have a custom model that is trained, tested and definitively serves a purpose within your organization’s marketing operations that will boost or improve a function directly and act as the brand brain of your marketing team. Time to tell your clients and the world all about it!
Flock has built a detailed reference guide called Illuminating AI that shows user cases, case histories, and a practical guide on how to implement AI (of all sorts). If you’d like us to present this information to you, please fill out the form below.
Contact us