Why can’t services be launched in a day? Using a custom GPT to make a ‘real’ service and test business assumptions.
A line of inquiry that’s alive for me right now is returning to something that has niggled for quite a long time. Over many years of taking service concepts and building early versions, I’ve always felt friction from the time and effort required to get something into that first customer’s hands.
I think we’d all agree that the gold standard for understanding if your service is going to work is when that first customer puts their hand in their pocket and gives over some of their hard-earned cash in return for the value you offer. Until you reach that point, it is speculation and hope.
The rounds of low-fidelity work, prototyping, and research needed to build enough context and confidence in a problem are still overly abstracted from the final experience that the first customer will encounter. Recently, this abstraction has felt increasingly absurd given the fleet of digital tools, services, and low-code offerings available to us, which allow us to create working services in hours rather than weeks and months.
Opportunity presents it’s self
Towards the start of 2023, I stumbled on a potential service. I had engaged in some home renovation and had to deal with a builder disappearing halfway through with my money. The whole thing was annoying, to say the least.
Later I pondered the power dynamics of hiring a builder to work on your home. It’s high stakes, very stressful and most homeowners are underprepared to manage it. I conducted some desk research and quickly understood the scale of the home renovation industry in the UK, which experienced a boom after the pandemic. The statistics and stories illustrating how these projects can go awry are abundant. Home renovation projects have a high potential for leading to money disputes between homeowners and builders, and these disputes can range from disagreements over the quality of work to misunderstandings of cost estimations created before work begins.
I had a bunch of ideas around removing money from the equation by implementing a third party, potentially taking advantage of an escrow service, to possibly level the playing field between the two parties. Additionally, educating the homeowner on the pitfalls of entering this kind of relationship is crucial. Another fundamental, old-school approach involves helping to create a contract between the two parties that not only sets expectations upfront but also manages the situation if things break down along the way.
What if there was a service that could guide you through creating a contract between you as the homeowner and the builder you’re looking to engage with? This service wouldn’t just be full of legal jargon; it would actually help you think about what was happening, prepare you to do more background work to set up the relationship correctly and lead you to a space where you had an agreement between the two parties, putting everything on a much more solid footing ahead.
Arbitrary creation goal
So there’s no need to set such a limited timeframe to create this prototype service, but why not? If my original concern was about the time and stages needed to build confidence in a business direction, then flipping that and doing something very rapidly with the tools available today seems like a good test. So there we are, create a usable first version prototype in a day.
I’ll dive into what was created and for what purpose below. The overarching goal here is to have as real a service as possible, to be able to put it into people’s hands, and learn first-hand if they would use it, how they would engage with it, whether they would pay for it, and if it would then go on to add any value to their lives.
Bringing the service to life
There’s an old saying, “If you want to go fast, go alone. If you want to go far, go as a group.” As a creative, this has always made sense historically. However, today we’re presented with a slightly different opportunity where we can go solo and go far. But there’s a fleet of new smart tools that turn the solo creative into a virtual team. With advisors in the form of large language models and creatives in the form of stable diffusion, coupled with top-notch off-the-shelf design tools like Figma, all these elements come together to allow you to move unimpeded at really fast speeds.
Storyboard
This approach might have been a bit self-indulgent, but I wanted to bring the service to life for myself and, potentially, to communicate its development to others over time. So, I created a future version storyboard of how I see the service fitting into a life. I’m working quickly to define the major phases and then jumping into using MidJourney, with the character reference feature, to maintain a single character throughout the cartoon-like storyboard creation.
Branding and advertising
It felt important to bring another part of the service to life with a very rough brand and some mock-up advertising. This becomes valuable as a visual asset to bring the idea to life, but it’s also useful as collateral during testing to make it feel considered. Especially since this service operates in a legal setting, the idea that a consistent and believable brand could build trust is crucial. Again, using MidJourney, with the character reference feature, to maintain a single character.
Service experience prototype
The final stage, and the most involved part of this process, was putting together a custom GPT using OpenAI. The custom GPT I’ve created works to guide the customer through the contract creation process while also incorporating learning opportunities along the way. It pops up and teaches them something relevant during the process. At the close, it generates a Word file contract with all the necessary boilerplate content that the person can download. As a final step, it directs them to a Google Form where they can provide feedback on the experience, offering a chance to gather some insights.
I implemented this feedback mechanism on the off chance that I allow people to use this service in a completely unguided way, such as over the internet, where they could simply follow the prompts. However, I assume that the most valuable testing would occur in a controlled manner. Therefore, I would be in the room, or at least on a call with them as they use the service, to learn from their experience.
At this point, you might have noticed that there isn’t a mechanism for charging the service participants any actual money to engage in the service. OpenAI has not yet implemented this feature into their custom GPTs. It sounds like it could be something on their roadmap. Additionally, they don’t have a way for embedding the custom GPT that you’ve made into another web page. I did consider adding a “if you found this valuable, please donate” button towards the end of the service and directing people out of the GPT. However, I’m not sure that this would be a particularly valuable test. Thus, I am reliant on people providing feedback through the form or verbally on the value they found in the service. I can also track who has used it and then follow up later to see if they implemented the contract and understand if they’ve seen any value. But ideally, there’ll be a paywall in place for this service at some point in the near future.
My view on price testing here, which I attempted to implement, would have been to take even the smallest amount from the customer, perhaps 50p or a pound. The aim was not to test the actual price but rather to gauge the friction involved when customers have to pull out their purses and input their card numbers. If this process felt like manageable friction and they perceived the contract as offering high enough value to warrant that action, that would be a sufficient indication of the value of the service being provided.
One interesting and very valuable side effect I found when creating the GPT is its flexibility within the service. It can really handle all the different types of questions, deviations, and unexpected queries that a tester using this service might present during the test phase. I find this aspect intriguing because it allows for qualitative testing to surface assumptions and avenues of thinking that might not be uncovered through other testing methods. The limited tests I’ve conducted with people at this stage have revealed all sorts of interesting thoughts that I hadn’t anticipated when building the prototype. However, the GPT managed to handle these effectively and get back on track very easily. Thus, the entire testing process wasn’t a washout, and the learning gleaned from it was phenomenally high value for me.
Want to know more?
I find exploring this space highly interesting, and I’m really happy to share in minute detail what I did and how I approached it. I think especially some of the techniques involved in creating the GPT for keeping it on track and structuring the prompt-based programming are particularly noteworthy. So, if this is something that you think you’d like to learn about or you’re interested in implementing it in a commercial sense and want to know more, hit me up. I’m very happy to chat about it.
Promoting used to drive the custom GPT
Your goal is to guide a user through a contract creation process. The end goal is to take their answers and compile a Word document following a set template.
*PROCESS STEPS*
1. You will greet the user "Hello, ready to get started with your contract?"
2. Using the list of questions and lessons below, step through them one at a time and capture the responses from the questions only.
-- Take time to evaluate the question answers you are given against the good and bad answer guidance for each question.
-- At the end of each lessons ask 'when ready to progress just let me know'
3. Once you have all the answers create a summary and ask the user if they are ready to create the Word document contract
-- If the user is ready: Create the Word document and supply it back to the user
-- If the user is not ready: Ask them for anything they would like to update. Once you have that loop back to step 4
4. Ask the user if they are happy to complete a feedback survey about their experiences and provide this url that will take them to a survey https://docs.google.com/forms/d/e/1FAIpQLSfPEPYkhA-eqzy7BxgK0-zPJdpQPYkdnpb1MxFf4nnkgTxIGw/viewform?usp=sf_link
*QUESTIONS AND LESSONS*
Question 1. Please provide your full name
- Good answer example: 'Rob Boyett'
- Bad answer example: 'I don't want to share my name' in this case let the user know that a contract without a name is not going to be any use for anyone
Question 2. Please provide your builders full name
- Good answer example: A full name with surname
- Bad answer example: One word answer
Lesson 1. It's always a good idea to ask your builder to show you proof of business registration and licensing.
Question 3. Please provide your builders business name
- Good answer example: A full business name with that type of company e.g. partnership or limited
- Bad answer example: Single word answer
Lesson 2. It's important to ask your builder to see their insurance documents. Check they are up to date and cover enough value if you need to reverse what they might have done.
Question 4. Please provide a short description of the building work being undertaken. We will go into more detail about the works in a later question.
- Good answer example: A paragraph of descriptive text
- Bad answer example: Short answer with little description. In this case, ask them if they would like to add more information
Question 5. Please provide the total budget assigned to the project
- Good answer: '£200k' a clear number and a currency
- Bad answer: 'not above 200 grand' in this case translate to a good answer and ask the user to confirm it is right
- Bad answer: 'I'm not sure' in this case let the user know that a contract without an assigned budget is not worth having, they should spend some time to come to a budget that will work for them
*TEMPLATE FOR WORD DOCUMENT*
This is a contract between a customer and a builder.
Customer: [Add the customer's full name]
Builder: [Add the builders full name]
Building company: [Add the builders company name]
Project description: [Add the project description]
Project budget: [Add the total budget assigned to the project]
Terms and condition
1. Boiler plate nonsense that would be added
2. More boilerplate
*TEMPLATE END*
