[HN Gopher] Launch HN: CommodityAI (YC W24) - Shipment managemen...
___________________________________________________________________
Launch HN: CommodityAI (YC W24) - Shipment management for commodity
traders
Hi HN! We're Daniel, Philip, and Kyle, cofounders of CommodityAI
(https://commodityai.io). We make commodity shipments easy to
manage. Commodities are the center of global trade. The food you
eat and the coffee you drink all start with agricultural producers.
A trader organizes a shipment of it to a receiver near you, who
then manufactures and distributes it to your local supermarket or
coffee shop. That sounds simple enough, but any shipment involving
commodities involves around 200 processes, at least 50 documents,
and extensive coordination through emails and phone calls. The
amount of paper and processes is staggering. Errors can be
catastrophically costly, ending up with lawsuits over millions of
dollars. Commodity trading companies allocate 40% of their
workforce to logistics and operations. It's a relationship-focused
business. These aren't trading firms with Bloomberg terminals doing
spot trading with algorithms all day. They are trading actual
physical commodities. Many of these deals are agreed on via text
message. Now imagine you're managing thousands of shipments a
month. Nearly every trader we've talked to can recall a time in the
past 6 months where a document or process error resulted in the
firm losing hundreds of thousands of dollars. This led us to
explore how we could tailor AI, automation, and collaboration tools
to automate the handling of shipment workflows for more efficient
operations and accurate documentation. Our team has the background
to solve this problem--an ex-sugar trader with 8 years of
experience and two engineers with a combined 14 years of experience
from logistics and software companies. There are a few layers to
solving the pain points of these customers: (1) Data
Preprocessing: we give them a way to clean up and organize their
existing data, which is usually a mess. We help them process
documents, classify them, extract relevant fields, organize them by
shipments, and search for them. (2) We use AI and automation to
leverage OCR, Multi-Modal Transformers, and indexing methods across
150+ different commodity and logistics document types to automate
business processes. (3) Actions and Workflows: Once the data is
processed, we enable users to take actions or kickoff workflows
with them, such as entering data into ERPs, using AI agents to
compose emails, update freight matrices, track their shipments and
predict delays, and more. All of these things are things the user
does manually currently and are prone to error, so we automate them
to save time and money. An ops team currently spends 60% of their
day on these tasks. With our initial product, we are already saving
them 2 hours a day. Our customers so far include one of the
largest U.S. agricultural traders, who use our platform to process
over a million documents and automate internal workflows, like
creating vessel nominations and updating shipment statuses. They
also use us to find critical documents quickly. Importers and
exporters randomly get audited to make sure the commodity they
declared matches what was actually delivered (quality, weight,
polarization). These audits can happen years after the shipment,
and searching for the relevant documents is a needle/haystack
situation. Sometimes they have been unable to find them, and you
can imagine how well that goes over with the auditors. We make this
easy. We can't give HN readers direct access to the product
because it can only be used if you have commodity data and that
data is sensitive. But here's a video showing how it works:
https://www.youtube.com/watch?v=slOGliSR6IM#t=26s There probably
aren't too many commodity traders checking HN every day, but we
figure some of you have worked as traders or know traders from
previous work experience, and we'd love to hear whatever you think
about this!
Author : KyleManuel
Score : 45 points
Date : 2024-03-26 14:12 UTC (8 hours ago)
| kaltsturm wrote:
| Congrats and keep up the good work! Can you describe your tech
| stack and challenges you faced in your first 6 months - we love
| to learn from your experiences.
| KyleManuel wrote:
| Hey there!
|
| We have worked extensively with many of the latest LLM models
| and tried many different techniques to solve the various
| problems for our customers.
|
| We use a combination of tools from Google and OpenAI to
| leverage LLMs. We use React and Heroku to support the web
| application.
|
| The biggest challenge so far has been getting good field
| extraction and the ability to accurately search millions of
| documents using NLP. We've iterated our tools several times and
| finally reached a good solution that adequately solves the
| biggest needs of our customers.
|
| We've built a RAG and iterated several times in order to try
| and get the best configurations for our customers.
| 1oooqooq wrote:
| how clients react when you show them those companies' privacy
| policy?
| PreInternet01 wrote:
| Well, my only experience with commodity trading is that blocking
| any emails that come from a previously-unknown sender and contain
| the substring 'commodit' cuts back spam significantly...
|
| But, good luck with that, I guess?
| danielmarkbruce wrote:
| What commodities are you focused on? Like, softs, energy, metals
| etc
| KyleManuel wrote:
| Our first few customers are agriculture based. Sugar, Coffee,
| Cocoa and more things like this that have similar distribution
| supply chains.
|
| We believe we have the tools and expertise to expand to energy
| and metal in the future. They are much different, but we
| believe with the tools we have developed, they can be
| repurposed quickly to support a wide range of commodities.
| danielmarkbruce wrote:
| Interesting, and makes sense. Good luck with it.
| Svdc wrote:
| I was doing AI implementations for a company in Oil and Refined
| products. And I can tell you that AI can massively improve the
| processes and and streamline their operations. In fact, it can
| easily transform any data entry to structure data without the
| needless sw development of API integrations and gateways.
|
| My first point as a contribution is, traders are lazy and
| operators are control freaks. In a way traders only care about
| doing deals so any trade entry in a system us prone to errors and
| misalignment, such as the ETRM system's reference data. And
| Operators are control freaks in a way that if they find our that
| your AI missed a single entry they will mistrust it forever. They
| rather spend more time copy pasting that to trust your AI. But
| any automation is welcome they still want to have control (which
| some in contrast will argue that that control is a nuisance and a
| waste of time).
|
| Oh well, though customers is what they are.
|
| I am really happy to know that there are others trying to solve
| the same problems in a world of ineffectiveness.
| KyleManuel wrote:
| Yes, we have noticed traders just want to make more deals, and
| operators require their information to be accurate.
|
| We are trying to develop super sleek and intuitive tools built
| to help traders update their ERP systems using LLMs and
| generative AI to better format their data.
|
| We hope to prove that operators can actually get more accurate
| data into their systems faster because we provide tools for
| data entry that make it a breeze to get accurate data.
|
| Additionally, we're trying to help our customers expand their
| businesses as well. As operations are generally their
| bottleneck, traders may not expand into business sectors
| because they are turned off by the overhead required to manage
| the operations.
| danielmarkbruce wrote:
| Out of interest, why not start a commodity trading house instead?
| KyleManuel wrote:
| We have had this thought was well. It's tough to go into a
| vertical with software because you end up with tools that make
| it easier to get an edge, and then why not just trade yourself?
|
| Generally, we just enjoy building software and it suits our
| personalities better. But we do believe that it is a good
| benchmark for our product to think, "could we use our tools to
| have a better edge in the market if we were a trade house."
|
| Great question!
| danielmarkbruce wrote:
| Right, fair enough. It does seem like some of the very big
| trading houses (glencore, trafi etc) could be built from
| scratch with a modern stack.
| KyleManuel wrote:
| Yes, this is true, they often do invest internally to
| create their own tools.
|
| We believe however that we offer better tools at a cheaper
| price and can get it into their hands now!
| Svdc wrote:
| for what i have seen in the demo it looks very very good.
| And I agree with a teak here and there this sw can be
| adopted by a range of trading houses of any kind of
| commodity, including the big ones with refineries. The
| difficulty in building systems like these, even if AI
| driven, is the myriad of business rules and the different
| specifics of each commodity trading.
| all2 wrote:
| > (1) Data Preprocessing: we give them a way to clean up and
| organize their existing data, which is usually a mess. We help
| them process documents, classify them, extract relevant fields,
| organize them by shipments, and search for them.
|
| To this point, I work with folks who are experts at accessing,
| organizing, tagging, and triaging data. If you'd like to set up a
| call, I'd be more than willing to facilitate.
| alexpetralia wrote:
| My opinion is that this is not really a tech problem, this is a
| business problem. Will the traders at Cargill, Vitol, Exxon,
| Trafigura, Glencore buy it? Do they want to take the time to
| "learn" another tool? Does it "just work", without any "let me
| get back to my product people"? Also, of course, every commodity
| market is completely different and has a whole web of intricate
| details in logistics, regulation, product, people, etc.
|
| I'd guess that only an established (or former) trader already
| using the tool with a deep network of contacts could convince
| other traders to use it. And then of course there is the whole
| other problem of enterprise sales and penetrating big financial
| companies (legal, compliance, procurement, infosec, etc.). So
| maybe you start with small traders? But my understanding of
| commodities trading was that is kind of like high frequency
| trading - there are only large, very established players.
| btown wrote:
| From what I'm reading above, the pitch is more "I can keep
| using my existing systems, but use this tool as a layer on top
| to help me with my data entry and anomaly detection, and our
| clients/external parties don't need to know that we have these
| superpowers."
|
| It's a great way to break into a market, but there's also less
| lock-in and forced virality relative to a system that you
| encourage suppliers/partners/counterparties to use. Which can
| be compensated for by word of mouth and ease of onboarding! But
| this can be a double-edged sword, particularly when investors
| are evaluating whether you're (planning to be) necessary or
| expendable to the industry.
| KyleManuel wrote:
| Yes this is our approach. We noticed that all our customers
| spend a significant amount of resources getting their tech
| infrastructure configured. We want to "supercharge" their
| existing solutions, not rip and replace.
| KyleManuel wrote:
| Yes you have some really good points. There are a lot of
| business problems to overcome here!
|
| Our CEO Philip is exactly that, a commodities trader with 8
| years experience and a wide network for us to tap into.
|
| As for every commodity market being different with its own
| intricate details, you are right. We are taking an approach to
| start in agriculture (sugar, coffee, cocoa) because these have
| similar supply chains. However, we are building configurable
| tools from the foundation to support more complex
| customizations to allow companies to easily tailor the
| solutions to their needs.
|
| We have also had some calls recently with commodity producers
| who have shown a lot of interest in the platform so far,
| because they many of them manage their own logistics teams
| internally, which opens up an additional market outside of just
| the big trading players.
|
| I really appreciate your feedback, thank you.
| secondcoming wrote:
| Just curious as to how well AI and OCR can handle locale issues
| in documents? E.g. is 1.000 is 1000 in German. There was also the
| famous Xerox scan bug.
| Bostonian wrote:
| Could your software be useful to institutional commodity futures
| traders who do not participate in physical markets?
___________________________________________________________________
(page generated 2024-03-26 23:00 UTC)