Are current AI tools useful?
This is the start of my personal business journey with AI.
This is Tim - the original founder of PrimaTrade and, as some of you will know, founder of some other companies as well. Today I am the Chief Product Officer at PrimaTrade. I am also not a "technical" person when it comes to software development.
We don't normally include personal posts on our website - but this one nicely links to our business.
I spent a bank holiday weekend here in the UK playing intensively and for the first time with what we laymen might call "AI". Let's ignore the implications for work-life balance at this point.
I wanted to understand whether the current breed of AI tools could be useful for us in our business - read on to find out why I felt this might be my job.
How to start the AI journey ...
Specifically for me that means the platforms offered at chatGPT.com and claude.ai - the AI tools developed by OpenAI and Anthropic. Of course, there's a lot more that the boffins talk about - but this is the consumer-facing frontline.
I supported my endeavours with minimum paid subscriptions to both of them.
To see if current AI tools might be useful - just start
Until I understood what these systems can and cannot do - it was rather frustrating.
I read plenty online, but generally found that to be a bit overwhelming. Moreover, there seemed to be a lot of braggadocio providing me with endless lists of prompts to use but not much visible evidence of results.
The irony of adding to the blogs about "working with AI" is also not lost on me - but my goal is actually, as a non-technical person, to deliver something in time that works and that you will be able to check out - a visible and tangible outcome.
Outside very large companies with big budgets - I've not come across actual products made by AI that I might use day-to-day - unless I wanted to fake up some photos of Donald Trump doing something he shouldn't.
I thought, how hard can it be? So I cleared some time away from real work and just got stuck in.
There are some lessons - the most important lessons were the last two!
One at time: Only ask the AI to do one thing at a time. So that is one task per message that you send. It will tend to ignore multiple instructions.
Limits: There is a size and complexity limit for each task. If the task is involved, the AI might well time-out or only give you half an answer.
Staging: If you are asking the AI to make something for you, it is best if the AI can "stage" it for itself.
That means it can try to check its own work for errors before handing it back to you.
If the AI says that it cannot stage or run what it is producing, do take note! It means that you will be testing what it delivers and you will be going back and forth a lot; the delivered output (called an "artifact") will have bugs (more bugs, in fact), just like when humans develop software.
Non-linear: AIs do not develop artifacts in the same way that humans do.
We generally create something, and then gradually improve it.
AIs tend to react to the latest instruction and try to deliver it - potentially destroying earlier work.
So our human process is naturally linear (A to B, add C, add D, reach E). The AI version of this process is A to B to C and then back to A plus (losing B and C en route).
Not production quality: Unless you are creating something quite small scale and limited in scope (which I intend to do!), it will not be useful without the support of other humans who have superior abilities to your own. They will iron out the kinks, make artifacts safe, stable and secure, and then work out how to re-write them (most likely) or integrate them (possibly?) into production environments .
But your efforts can pay off.
Why should I, a non-technical person, play with AI?
There are some simple reasons why this is worthwhile.
In our business, we are not going to hire AI specialists into full-time roles or bring in expensive consultants to advise us before we are fully confident that this effort will be worthwhile. A recent MIT study found that 95% of AI projects fail. So we need to have a plan and we need to know what we want to do before we invest any serious efforts.
So we need to investigate:
We can ask a capable, but perhaps not so business-aware, developer to dig into what's possible, or
We can have a business-aware but not technical person do the same.
And here's the point: the new generation of the AI tools are actually not that hard to use, even for a non-technical person. Route 2(b) - finding out what's possible as a business-aware but not technical person - that's the way to go.
Try out AI yourselves - that's what this post is all about.
And be careful - there's another MIT study out there which says that AI is making people stupid (click here for that one!).
What can I learn then (as a non-technical person)?
There's a lot of hype out there. Ignore it.
The key point is to start. This is what you face! Wow, scary.
I tried four different tasks.
These are all relevant to matters I worry about every day in my role at the company, and they are very relevant to our business as a whole.
None of these tasks relates to "personal" productivity (eg: my own calendar, my email inbox etc). I'm not ready for that.
Marketing and messaging: We have a comprehensive website with over 100 posts including videos and over quarter of million words. The posts are all about what we do and why it matters to clients.
Can the AI distil key messages and create short and concise marketing artifacts to improve our communication?
Security and confidentiality: the data that we handle is confidential and we take security very seriously.
What strategies can we use to help us take advantage of third party services (eg: embedding AI services via API that deliver real value to clients) without compromising on security and confidentiality - and where does AI add value?
Build me a website chat bot: We offer disruptive but very valuable solutions to our clients - and it would be great to have something smart on our site that handles visitor questions using both our own data and the power of the internet.
Can the AI take our content plus what's available on the internet and create an automatically-updating, powerful and interactive tool that we can add easily to our website; that's to help our visitors get more out of our website and potentially encourage visitors to call or contact us?
Data analysis: We are a data business. We capture the fulfilment of purchase orders for clients together with the digitization of documents, the resulting shipments, payments, customs filings and financing. That's a huge amount of interesting data involving millions of records.
What analysis would AI conduct on the kind of data that we capture and how would AI present its results to time-poor clients so the results are simple and easy to understand - can it build me a prototype?
What happened?
Well - here I am sharing my journey so far.
Of the four tasks above:
the first two tasks are advisory (tell me useful stuff) and
the final two involve delivering something (make me useful stuff)
Most of us have already tried asking questions and seeing what comes back. If you have not done, start there.
You can find plenty of guidance about how to ask questions albeit there is too much of that out there and perhaps a lot of it is not backed up with evidence of results.
When it comes to specific advisory work that is useful to your business, you can provide guidance and materials for AI to process and work from. Generally the better (and shorter) your guidance about what you want, the better the result you might receive.
Marketing: It is a great test of the quality of your marketing if the AI can return a concise and sharp summary of your key messages from the marketing material that you provide to it. You can also ask it to do this on other websites which can be even more interesting! If you don't like what it says, you can always ask it for advice on how to improve.
Using AI with data: I learned three things here.
If you can anonymise the data that you send out to an external AI system, there are viable and inexpensive ways to embed third party AI services via API calls. That's a powerful point. If we can do that, it means we can already embed the power of third-party AI into our production environments without needing to hire specialists or build expensive in-house platforms.
When you have confidential information that you want the AI to process and you cannot anonymise it, implementing AI becomes complicated and expensive because you have to bring the LLM in-house or work with a partner that provides robust ring-fencing of your data (and that's not what the household-name AI providers normally do). There's a trade off between cost and benefit which has to be carefully considered.
It is important to identify where AI actually adds value:
PrimaTrade digitizes international trade paperwork (invoices, shipping documents, packing lists, inspection reports, certificates etc). This means we get unstructured, often language-based, data that is embedded in trade documents that needs to be cleaned and organised as part of digitizing each shipment we process. We already use 3rd party AI systems to help us tidy up data as it is created in our platform via connections to specialist AI systems that sort that out. Tick the box there.
As a result of this first step, nearly of the data in the PrimaTrade platform is already well-structured and already tidy. So AI can help us to design better algorithms that make sense of that structured data and better dashboards to present it - improving our designs. But day-to-day operational usage of AI is going to be limited apart from one or two areas.
But what about creating actual "artifacts"?
What a lot of fun, but patience is required.
Note this point - it is important: if you don't want your creations shared with the world-wide-web, you have to build your own environment locally on your laptop or PC to run the code that the AI generates.
So I asked ChatGPT to help me do that - and it took about 30 minutes of downloading and fiddling around to set up a fully-working local environment that can run the code that AI systems generate. This means downloading and setting up free tools that can run code (Python, React, etc). Honestly, it's not that hard. It's a bit like following a complicated cooking recipe in the kitchen.
WIth that in place, I could then ask the AI platforms to build me "artifcats" which I can then copy into my local environment on my PC and run.
With two AI platforms running, I also had a lot of fun taking artifacts created by one AI and getting the other to review and improve it. Claude works significantly faster than ChatGPT, but ChatGPT was quite good at polishing and improving the artifacts that Claude would make.
And then - amazingly - working prototypes emerged almost immediately.
Data analysis: one of my development colleagues provided me with an anonymised sample dataset (not too big) from our platform. Off I went. Keeping the data as a single "flat" file - I threw it into the AI with some background instructions - and I got basic working dashboards, interactive filters, alerts and analytics. Wow - a ton of new ideas.
Website chatbot: this is a great use case for AI - simpler than data analysis for sure. WIth some small effort, I made a basic chatbot trained on our website. Hmm - not yet something I can use in a live environment, but maybe I can invest some more time in it and come up with something that will work. It's a start.
What next?
Come back here and find out.
I will post some more on what happens
Can I make (without support from colleagues) a production quality and useful product with the AI tools? Can I deploy it so you can actually see it and try it out for real?
Can we find a way to link our confidential production environments (CRMs, CMSs, databases) safely and securely into public AI tools produced by the likes of Anthropic and OpenAI (Claude and ChatGPT)? This will give us quick and low-cost paths to embed AI capabilities into our production environments, enhancing the value we can deliver to our clients.
Do get in touch with us to talk about this post - or any other matters relating to trade finance, supply chain finance, AP automation or trade digitization.
Here's how:
Email me here (tim.nicolle@prima.trade)
Book a call with me (click here)