IT asset disposal business · Platform architecture, AI & automation
IT asset disposal touches a warehouse bench, a corporate client’s compliance team and a wholesale buyer. They need completely different interfaces and exactly the same truth underneath.
Live platform counts, August 2026. AI evaluations since June 2024.
The situation
The business processes IT assets at volume: collection, data destruction, testing, grading, pricing and resale. Each of those steps was a judgement made by a person, and the ones that fell through the logic fell through silently.
The harder problem was that the business has three audiences who share nothing except the asset record. A technician at a bench needs hardware access and a fast keyboard-driven flow. A corporate client needs collections, quotes, documents and an audit trail they can show an auditor. A wholesale buyer needs stock and prices. Build those as three systems and they disagree within a month.
My role
I shaped the process, built the software, and sat in the pitch.
This engagement didn’t start with a specification. It started with a workflow that runs from collection through destruction, testing, grading and pricing to resale, and the question of which parts of it were judgements a person was making by hand, and which of those a system could make more consistently than a person could.
Answering that’s the commercial half of the job. Building the result is the technical half. Both were mine, which is why the automation landed where the money was, not where the demo was easiest.
I also worked on the pitch. When a prospect asks how the chain of custody holds up under challenge, the person who designed it can answer in the room, and that’s a different conversation from a salesperson promising to check.
This one didn’t end at handover. The platform runs, and I’m still working with them on what it does next. That’s the arrangement the retainer describes: they could stop any month, and the reason they haven’t is that the decisions kept coming.
The outcome
It stopped being an operational detail and started being something the business sells on. It has since won work with some of the largest insolvency practices in the UK, where the ability to evidence what happened to every device is the thing being bought. I was in those pitches and the chain of custody came up in them. I can’t show you the version of events where the platform didn’t exist, so take the sequence as fact and the causation as my read of it.
Deliberately no efficiency percentage here. The platform counts above are measured; a time-saving figure wouldn’t be.
What changed
Two technicians could look at the same laptop and write down different things, and nothing downstream could tell. Every device now resolves to a defined grade with stated reasons behind it. Over 248,000 tests have gone through it.
In asset disposal the certificate is the product, and theirs depended at one point on somebody reading a number off a drive and typing it in. It doesn’t any more. That’s the difference between a document and evidence, and it’s what a buyer at this level is paying for.
What a corporate client could see and what had happened to their assets used to be two accounts of the same events. They are one account now. Most of the pain in this industry lives in that gap.
Grading is machine-assisted and configured for repeatability rather than cleverness, because a grading system that disagrees with itself is worse than not having one.
The business reports from a layer built for reporting, and the cost of every AI call is logged against a budget, so spend is a line somebody can read and not an invoice that arrives.
Some hardware couldn’t be audited by the main application for reasons that belong to its operating system. Those devices reach the same records as everything else now.
Decisions worth explaining
The bench app, the client portal and the wholesale side are deliberately different products over one asset record. The alternative is a shared interface that suits nobody, or three databases that disagree. That’s the failure mode this industry lands in.
Cosmetic grading feeds pricing. A model that’s creative about a grade is creative about a price, so settings are chosen for repeatability and the output feeds a rule instead of being the decision.
Devices that couldn’t be priced were falling out of the logic with no error raised. They’re now sorted by the reason they failed and queued for a person, which turns an invisible loss into a visible task.
Destruction records, test results and audit reports come out of the same flow that does the work. Evidence reconstructed afterwards from memory is the thing that fails an audit.
Stack. C# / .NET and Python, MongoDB, Azure storage and hosted vision models, across desktop, browser and web clients.
“We brought Ashley in to find where the business could run better, optimise it and build what it needed. He delivered far more than we asked for. Processing that took a technician most of a morning now takes minutes. The system he built has helped us win pitches with large corporate clients. And I no longer spend every day in the weeds: I run the business.”
Work with me
Tell me the problem you’re trying to solve.