AI in Vehicle Recycling: The Complete Guide for UK ATFs

Walk into most UK Authorised Treatment Facilities today and you will find the same setup: a whiteboard on the office wall tracking incoming vehicles, a stack of CoDs waiting to be posted, a pricing spreadsheet that one person understands, and a mobile phone that never stops ringing with collection requests.

This is not a criticism. It works. ATFs have run like this for decades, and the businesses that survive are the ones where experienced operators carry the complexity in their heads. The buyer who knows that a 2014 Ford Focus 1.6 diesel is worth £180 at today's scrap price but £240 if the catalytic converter is intact — that knowledge is hard won and genuinely valuable.

The problem is what happens when that person is on holiday, off sick, or leaves. The problem is what happens when the yard goes from processing 40 vehicles a week to 80. The problem is the regulatory burden growing every year — depollution records, quarterly returns, CoD tracking, hazardous waste documentation — all of which still rely on manual data entry at most sites.

The existing software landscape has not helped much. Most ATF management systems were built 10–15 years ago. They handle DVLA notifications and basic stock tracking, but they were never designed to price vehicles automatically, optimise collection routes, or provide real-time dashboards. They are database frontends, not intelligent systems.

Some larger operators — groups running 5+ sites — have invested in custom ERP systems. But the typical single-site ATF processing 30–100 vehicles per week has been priced out of that market. An enterprise ERP implementation costs six figures and takes months. For a business turning over £500k–£2m, that is not realistic.

This is the gap AI fills. Not because AI is a magic word, but because the cost of building intelligent software has dropped dramatically in the last two years. Tasks that required teams of developers and months of work — natural language processing, pattern recognition in pricing data, automated document generation — can now be built by small teams in weeks.

The question is no longer "can AI work in vehicle recycling?" It is "which problems should it solve first?"

I have worked at Reclamet Recycling for 20 years. For most of that time, pricing a scrap vehicle meant the same thing: a customer phones in, you ask for the registration, look up the vehicle on the DVLA site, estimate the kerb weight, check today's scrap price, work out the collection distance, and mentally adjust for the state of the car. An experienced buyer does this in 10–15 minutes. A new hire takes longer and gets it wrong more often.

The problems were obvious. Pricing was inconsistent — two different people quoting the same car on the same day could be £30–50 apart. Busy periods meant missed calls, and missed calls meant lost vehicles to competitors. There was no record of quotes that did not convert, so we had no data on why customers were going elsewhere. And when scrap prices moved — which they do weekly — updating the mental model across the team took days.

The pricing engine I built solves this. It pulls vehicle data from a registration lookup, applies weight-based pricing using live scrap metal rates, adds zone-based collection costs calculated from the customer's postcode, and applies margin rules set by the operator. The quote is generated in under 30 seconds.

The system uses a zone model — concentric rings radiating from the yard, each with a collection cost that reflects fuel, driver time, and vehicle wear. Postcodes are mapped to zones automatically. The operator sets the pricing rules: minimum offer, target margin per tonne, premium for intact catalytic converters, deductions for missing wheels or non-runners that need a flatbed.

The technical architecture is React with Supabase on the backend. Vehicle data comes from a registration lookup API. Metal prices are updated from published indices. The pricing logic runs server-side so the rules are not exposed to customers. Every quote is logged with full context — the vehicle, the price offered, the factors that influenced it, and whether the customer accepted.

What changed: quotes that took 10–15 minutes now take 30 seconds. Pricing is consistent regardless of who is on shift. The quote log revealed that 40% of lost customers were in postcodes where our collection cost was too high — something we could not see before. We adjusted zone boundaries and recovered some of that volume.

What I am honest about: the pricing engine is shipped and running. The broader platform — fleet management, full compliance automation, the driver PWA — is built but still being tested with a small number of users. It is not a finished enterprise product yet. It is a working system built by someone who has done the job, and it solves real problems I have lived with for two decades.

The reason the NikkSi cost is lower is not because the work is less thorough. It is because I am not learning your industry on your budget. When a generic agency takes on a vehicle recycling project, the first 4–6 weeks are spent understanding what an ATF is, how pricing works, what a CoD is, and why collection zones matter. I have already done those 20 years.

The in-house option makes sense for large groups processing thousands of vehicles per month across multiple sites. For a single-site or 2–3 site operation, the fixed cost of a full-time developer is hard to justify against the volume of work.

A practical approach for most operators: start with one module — typically the pricing engine because it has the fastest return on investment — and expand from there. A working pricing engine with a customer-facing portal can be live in 4–6 weeks for under £10,000. If it proves its value, add fleet management and compliance automation as separate phases.

The first step is a discovery call — 30 minutes, free, no obligation. The purpose is straightforward: I need to understand how your operation works today and where the pain is. Not every ATF needs the same thing. Some are drowning in compliance paperwork. Others are losing vehicles because they cannot quote fast enough. Some need fleet management before anything else.

If we decide to proceed, week 1 looks like this:

From there, weeks 2–4 are refinement: handling edge cases (commercial vehicles, flood-damaged cars, vehicles with no MOT history), building the customer-facing portal, connecting to your DVLA notifications, and testing with real quote requests.

The most important thing about week 1: you will know by the end of it whether this is going to work for your business. There are no six-month surprises.