A B2B return gets lost in emails. And by the time it resurfaces, the batch has already been sold three times over.

The problem

The return comes in through one channel and goes out through three — all manual.

In industrial B2B, a return rarely comes in through a well-defined system. The typical sequence:

1. Scattered origin — one customer complains by email to the sales rep, another opens an incident in the retailer portal, another phones the account manager. Every channel, its own format.
2. Internal ping-pong — the sales rep forwards it to quality, quality asks the warehouse to recover the product, the warehouse stacks it in a "pending" area, quality investigates whenever it can.
3. Losing the batch — by the time somebody identifies the specific batch behind the return, that same batch has already been sold to other customers. If the root cause warranted it, the scope of the problem has already multiplied.
4. A decision out of time — the rework/scrap/recall decision is taken too late, without a consolidated dossier, and with the constant feeling of firefighting.

The problem is not quality — quality does what it can with what reaches it. The problem is that there is no system closing the loop: one that normalizes the intake, categorizes the cause, connects it to the batch, calculates the scope and prepares the dossier for the decision. Meanwhile, SLA clauses with B2B customers get breached, recalls become more expensive and the brand erodes.

How it fits into the IRIS system

iLEAN does not replace quality — it takes the dumb work away and gives the judgement back.

IRIS (Industrial Reality Intelligence Systems) is the category; iLEAN is the system. Reverse logistics is the textbook case of "multichannel capture + agent + glue between systems" — exactly what iLEAN is designed to do. The information about the return exists (emails, photos, portals, calls), the traceability system exists, the cause catalogue exists — but nobody has the time to cross the three of them, return by return, in real time.

Connect normalizes the multichannel intake. Tracer connects each return to its batch. Agent categorizes, calculates scope and prepares the dossier. Writer drafts the reply to the customer. The person signs — always.

The iLEAN pieces applied to B2B reverse logistics:

• iLEAN Tracer — the traceability subsystem the whole operation is built on. Every product sold has its footprint (batch, route, customer, conditions). When a return comes in, Tracer is the source that lets you reconnect the physical product with its full history, even when the label is only partly readable.
• iLEAN Agent (quality / returns management) — an assistant that categorizes the cause by cross-referencing email, photos, complaint and internal catalogue; connects it to the batch via Tracer; calculates the scope (other customers holding the same batch, other products from the same process); prepares the dossier with an estimated cost per option (rework, scrap, partial recall, full recall). It proposes; the manager signs.
• iLEAN Writer — drafts the reply to the B2B customer through the same channel the return came in by (formal email, portal message, call script), respecting tone and SLA. The reply does not go out without a human signature — Connect transports, Agent decides, the person signs. The iLEAN golden rule.
• iLEAN Connect — the multichannel capture that normalizes a wildly inconsistent intake (email, portal, WhatsApp, transcribed call). Three levels of integration with the B2B customer's system depending on what each account allows, without asking you to change anything in your portal.

See the full IRIS architecture →

Before and after

Manual reverse logistics vs. the loop closed with iLEAN

Aspect	Manual handling of the return	With iLEAN Tracer + Agent + Writer
Return intake	Multi-channel, multi-format, multi-bounce	Normalized by Connect at second zero
Cause categorization	Manual, whenever quality has the time	Proposed by the agent with its confidence level
Connection to the batch	"We'd have to look into it" — days	Via Tracer, in minutes, with OCR if there is a photo
Scope of the problem	Discovered when the next return arrives	Calculated as soon as the cause is categorized
Dossier for the decision	Assembled by hand once the committee is convened	Generated automatically, ready for management to sign
Reply to the B2B customer	Drafted manually, SLA at risk	Drafted by Writer, person signs, SLA met

Impact estimate

Impact estimate for your plant — to be validated with your numbers.

This block is an estimate to be validated with the actual data of your plant. It works as an order of magnitude for the committee; we refine it during the diagnostic.

• B2B plant with a multi-customer portfolio, returns arriving through 3-6 different channels, currently handled between sales, quality and the warehouse with no single system to close the loop.
• A 3-5 day immersion with the mixed team. Pareto: the 5-10 cause types that concentrate 80% of returns, the 3-5 customers that concentrate 70% of the volume.
• First agent (categorization + connection to batch) running within a few weeks. Writer integrated in the following cycle.
• Average return closing time down ≥ 40% in the first measurable cycle.
• Measurable improvement in SLA compliance with B2B customers — it depends on each account's exact SLA, but the hard lever is replying in hours, not in days.
• Indicative payback between 4 and 9 months. The hard levers: lower cost per return handled, prevention of bigger recalls through early detection of the affected batch, recovery of product that admits rework and used to be written off as scrap.

The underlying figure — B2B quality demands another level

In automotive, the demanding standard sits in the order of 25 PPM (parts per million) ^[1]. In other B2B sectors the requirement varies, but the direction is the same: the industrial customer does not accept the defect ratio the end consumer tolerates, and contractual clauses are starting to include ever stricter SLAs for responding to returns. Closing the loop is no longer an improvement — it is commercial survival in large B2B accounts.

And the CAIO's doubt — does the AI invent the cause?

"What if the agent categorizes a cause wrongly and drives us into an unnecessary recall or, worse, into skipping one we should have run?" — hallucination is a problem of free generation, not of anchored tasks. Cause categorization is an anchored task: read the email, compare against the catalogue, propose with a confidence level. On anchored tasks, the best models brought the error rate below 1.5% ^[2]. And even so, the final decision (rework/scrap/recall) is never automatic: the agent proposes it and the quality manager or senior management signs it, depending on severity. The three safety rings are there for exactly this.

^[1] Symestic — quality standard in automotive.

^[2] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.