NCCCS — Holbury; Case study
Starting a new business is hard, especially if you do not have any investment. Hiring staff becomes almost impossible and is a huge expense both in money and time. That said, Assist.wtf has completed a project for a working prototype of their first AI Employee.
The client is a local small business that has recently started trading. They specialise in sourcing MOT failures and old and unloved cars, which they then ‘part out’ and sell on eBay. The physicality of this business means the small team often finds it hard to make time to sit down and list every item for sale. For this team, it’s easier to pull engines than think up product descriptions.
NCCCS — Holbury’s logo
Initial investigation.
Assist.wtf was brought onboard to automate the online side of the business. The initial investigation resulted in the following parts that can be quickly automated for the client so that they can focus on what they enjoy doing.
The tracking sheet for individual parts
Part tracking using a simple spreadsheet — The client needs to provide the part name, some details about the vehicle, and a tracking number for their storage system.
Auto pricing — Why would anyone want to sit and search for similar prices for parts, when you can scrape eBay and do some basic maths instead?
Auto product name and description writing — Utilising LLMs, the AI comes up with SEO optimised names and descriptions for each part. It also includes a brief description of NCCCS — Holbury and what they are about, as well as a cheeky shout-out to Assist.wtf.
Item & Image matching — The team needs to take photos of each product and upload them to a Google Drive folder. They also need to rename the images to match the part number. There are plans to automate this using a multimodal LLM in the future, but it’s expensive and out of scope for this initial AI employee.
eBay API access management — Access tokens need to be created and recycled to connect to the API.
Integration with eBay API to create inventory items, update offers, and publish listings — eBay has quite a complicated API flow, which means this AI employee has a lot of parts to consider.
Imagine trying to hire someone to do this. That is an IT and an admin person as a minimum. IT for integrating with the API as well as fixing any computer-related issues, and admin for searching for parts, writing listings, and managing them on eBay. The online side of NCCCS — Holbury is quite complicated for a business that deals in individual car parts.
The full AI Employee flow visualised in n8n
The image above shows the complexity of the AI Employee. There are a lot of integrations with different products, services, API’s, LLMs, and Agents. Most companies, especially new start-ups, simply do not have access to this kind of IT understanding, which is why the Assist.wtf agency was created.
The build.
Let’s break down the build. First, the purple section for the eBay API was created. This allows manual interaction with the API to do all the things the AI Employee will need to do. It also allows the team to post listings that need to be fixed, but more on that a bit later.
Once the API was connected and tested, it was time to understand and create the flow that each eBay listing would require. The best way to do this is to take a single product from the list and manually post it. After speaking to the NCCCS — Holbury team, it was important that they had pricing for each product completed as quickly as possible, as they often get requests for parts, as well as needing to know the potential value of a car. So, this was implemented first.
Price section for each part at NCCCS — Holbury
The pricing section is pretty straightforward. The part name is used with some added keywords to search eBay, fetch the relevant HTML webpage, and then scrape the prices into an array. This is then used to populate the minimum, maximum, and median price fields in the sheet. For now, the AI employee sets the list price to be equal to the median price, as it is the most accurate and realistic for 90% of part listings. The list price for each part can also be manually entered if the AI employee can not find an accurate price point via web scraping.
After pricing comes writing the product title and description. This utilises two AI Agents, both connected to an Ollama LLM, running Llama3.1:8b. Using some very clever prompt and system engineering, as well as some code to format the output correctly, the AI employee ingests the product name, details about the car that it’s from and some select key words to produce the required outcome. On top of this, a further section is automating added adding some details about the business and their technology provider.
Part name, AI-generated description, and title for each product.
Moving on, the next stage is matching images and parts, creating upload links, and formatting them into an array that eBay’s API accepts. This part was a bit tricky to solve due to eBay’s documentation on how to structure the JSON body being pretty terrible. Eventually, through a lot of trial and error, the image section was completed, and we could move on to creating and publishing product listings.
eBay product listing for a Vauxhall Zafira rear light cluster
The biggest struggle with this AI Employee was structuring the data in a way that eBay’s API accepts. Like a lot of software companies that built APIs over the last 20 years, their documentation is pretty poor, there are quirks in the system, and their error reporting is questionable at best. That being said, having built the manual flow for the API first, this part was easier to hook into the main flow.
Once a listing is published live, the AI employee finishes up the flow by updating the sheet a final time to include whether the part has been posted correctly or not. It is basic error reporting, but it helps when you’re batch posting 40 parts at a go. This then allows a very simple overview dashboard within the sheet to be updated for the members of NCCCS — Holbury to quickly view how things are going.
Simple overview dashboard
The data
With the AI Employee deployed and tested, the NCCCS — Holbury team is happy, and a huge amount of time and money has been saved. let’s look at some stats.
Time to fill in the sheet with part names and the vehicle they came from for 40 individual parts — 5 minutes.
Time to take pictures of each of the 40 parts and upload them to a Google Drive folder — 15 minutes.
Total human time required before AI Employee takes over — 20 minutes.
Time for the AI Employee to run through all 40 parts and publish them to eBay — 48.823 seconds.
Let’s now look at the price for the build using Assist.wtf
Beta user access with Cloud hosting — £150 per month — Hosting + 10 hours dedicated development time included.
An additional prioritised 30 hours of development time to get this AI employee up and running within a week — £1,500
Total for build — £1,650
In other words, for less than the price of all the parts currently listed using the AI Employee, NCCCS — Holbury has a full-time admin that can operate 24/7 for a month and post as many parts as they want to eBay. They keep this functionality for £150 a month AND still get continued development.
The future
With the initial build deployed, the next step is to make it more usable to NCCCS — Holbury’s use case. What this means in practice is that we will be integrating the entire AI Employee into a WhatsApp bot. The team will be able to send the photos of their items to the bot and use a number of the manual flows to interact with the API if they need to update or correct individual listings. Further to this, in time, we’ll have order tracking and delivery automated, and weekly/monthly reporting available. Not only with this AI employee have their core responsibility of posting parts from the sheet, but they will also be interactive and respond to requests and commands when the team may need some manual intervention.
Example of a WhatsApp AI agent that could be used.
Final thoughts
This AI Employee project has been the best use case so far that Assist.wtf has found for this kind of technology. Not only was the team able to get the entire system built within a week, but they also did not replace any existing member of staff. The timing of this new business starting up means that they were able to automate from day 1 with no technical debt to unpick. This is exactly the kind of business opportunity we are actively looking for.
If you want to get involved, head to www.Assist.wtf for more details and to get in contact.
For locally sourced 2nd-hand car parts, check out NCCCS — Holbury’s eBay page here: https://www.ebay.co.uk/usr/ncccs-holbury
