The Perception–
Reality Gap
A full-stack UX audit of MetalCloud — the operating system for India's metal foundries.
Renewals looked healthy. Floor-level usage had quietly gone dark. Over ten weeks I audited three flagship modules — triangulating 9 months of usage data across 13 foundries, 121 session recordings, 40+ shop-floor validation calls and in-depth JTBD interviews — to answer one question: are customers paying for value, or for the appearance of it?
The terrain
MetalCloud digitises the foundry floor — from the chemistry of a molten heat, to the discipline of the melt itself, to the quality certificate that leaves with the casting. Three modules carry that promise:
Reads spectrometer chemistry instantly and pushes addition / dilution guidance to the melter's WhatsApp — so the crew knows what to add without walking to the lab.
Live heat tracking and melting-SOP enforcement. Every step of the melt is meant to be logged in real time, giving managers a transparent, traceable record.
Automated Material Test Certificate generation and QA traceability — audit-ready proof of what a casting is made of, without the manual paperwork.
The question I was handed
I framed the entire audit around three questions — one for each module, asked the same way every time. If a module couldn't answer them cleanly, its healthy renewal number was hiding something.
Not the pain on the pitch deck — the specific, real problem a melter or quality head feels on a live heat.
A tool that solves a rare, low-stakes annoyance will never earn a daily habit, however elegant.
Owners buy. Operators use. The audit had to separate the two, because they renew for very different reasons.
How I designed the study
I built a two-track investigation so neither the numbers nor the narrative could go unchecked — then anchored both to hypotheses I wrote up front, so I was testing beliefs rather than fishing for a story.
Behaviour, at scale, from the product itself.
The humans who live those numbers daily.
For every module I wrote explicit Problem, Solution and Implementation hypotheses, each with a measurable validation signal — e.g. for SpectroPro: foundries lose ≥3% per heat to slow manual dilution; a tagged spectrometer would cut alloy cost 1–2% at 95% suggestion accuracy. That turned a vague "is it working?" into something I could actually prove or disprove.
Write problem / solution beliefs per module.
Pick roles that reveal use vs. non-use.
Build case-based JTBD questionnaires.
Interview, record, collate the raw data.
Build the Perception-vs-Truth table.
The instrument I built
To separate what people say from what they do, I ran every interview through a case-based Jobs-to-be-Done instrument — the same card, filled from one real event, for every role and every module.
Reconstruct a specific, recent event — never a hypothetical or an average.
Pull the actual reading, message or file — evidence over opinion.
Identical fields across roles and modules, so answers compare cleanly.
I filled one of these per module, per role — reconstructed from interview transcripts and call recordings, never from memory. That's what let me later line up the same job across a melter, a quality head and an owner, and watch the answers diverge.
Into the field
Opinion surveys tell you how people feel. I wanted to know what they did. So every interview walked one real, recent case end-to-end.
Each script pinned the interviewee to a single specific event — the last heat, the last melt plan, the last certificate — and asked: what triggered it, what did you open first, what did you reach for outside the product, and what would you do if this module vanished tomorrow? Facts and artefacts, not feelings.
Melters, melting heads, quality heads and metallurgists across foundries — the people whose hands are on the heat. To triangulate, I also debriefed the internal onboarding and sales-adoption leads, then held their claims up against the raw Amplitude and Clarity behaviour.
One customer's weekly reports "weren't landing." I traced it: they were being sent to a decision-maker's number he no longer used. Nobody had checked. A small thing — but it captured the whole problem. We were measuring delivery, never absorption.
The people behind the data
The study surfaced a hard split: the person who buys MetalCloud is almost never the person who uses it — and the two renew for opposite reasons. Five archetypes carried the whole pattern, synthesised from the JTBD interviews, 40+ validation calls and the readership deep-dives, then checked against the behavioural data.
Perception vs. reality
When I laid the promise beside the ground truth, the same gap opened under all three modules. What was sold, and what actually happened on the floor, were two different products.
Renewed for visibility, not for value.
Every internal lead I interviewed arrived at the same verdict independently — the product solved real problems, but its adoption model was misaligned with how foundry teams actually behave. The gap wasn't technical. It was behavioural.
The emotional journey
I mapped sentiment for the buyer and the operator across the same journey. The two lines start close — then split. That widening gap is the whole risk, and it is invisible on any usage dashboard.
METHOD — sentiment coded from interview transcripts at six shared journey moments, per persona. The curve is directional — a synthesis of what people said and did, not a measured instrument.
Report delivery keeps the owner reassured from pitch to renewal — the optics never break, so the sentiment never drops.
Confusion at the first report hardens into apathy in daily use, and quiet failure at the exact moment the tool should matter most.
The renewal paradox
If the floor barely used SpectroPro, why did owners keep paying for it? I called through a 40+ user sample to find out. The answer had almost nothing to do with usage.
METHOD — each user's call transcript was coded on three behaviours: reads the WhatsApp reading, trusts it, acts on it. Filled = consistent, half-tone = conditional, empty = absent. Four users shown as the representative spread of the 40+ sample.
A TV screen and WhatsApp pings create a constant sense of activity — even when no one acts on the data.
Once installed it runs passively. No daily effort means no friction, no complaints, no reason to cancel.
Chemistry on a screen signals modernity to clients and auditors — a reputation asset, not a tool.
No operator says “this is useless.” They quietly ignore it — and silence reads as success upstairs.
Against the cost of machinery, renewal is trivial — easier to keep than to reopen the decision.
The same alerts reach owners, who assume everyone downstream is acting on them. No tracking, no gap noticed.
The customization myth
Customers had one recurring theory of their own low usage: "build my custom request, and we'll finally use it." DigiMelt's own execution data said otherwise.
Of 17,364 heat plans created across 13 foundries over nine months, only 6,932 were actually executed — 40%. Foundries that got their custom features built didn't climb; one high-volume foundry sat near 14% execution regardless.
The blocker was never the missing feature. It was readability — even literate melting heads struggled to interpret the reports — compounded by complexity and a missing feedback loop: nobody ever learned whether reading a report changed an outcome, so the habit never formed. The handful of power users who did read the graphs proved the value was real — just locked behind friction.
What it all meant
Across every module, the same six forces kept surfacing. Together they describe a product with a behaviour gap, not a feature gap.
Usage falls off sharply after install. The value promised at purchase is never realised on the floor.
Owners buy the vision; operators experience the tools as extra work, not support.
The product fits the workflow in theory, but nothing anchors it into the operator's daily rhythm.
Users never see whether acting on a reading helped — so the behaviour is never reinforced.
Buyers assume teams will adopt automatically. Without a push, they simply don't.
A report on WhatsApp feels modern, but modern delivery masks near-zero comprehension.
Where I pointed the roadmap
Diagnosis is only half the job. I translated the findings into a sequenced set of moves — reframing what "success" even meant, then fixing the floor before adding to the ceiling.
A renewal built on optics is a false positive — the health metric had to become "did an operator act on this today?" not "is it switched on?"
Kill the dead-click and rage-click hotspots and the grade-duplication errors surfaced in Clarity. Add clear loading and disabled states — trust erodes the moment the tool stalls.
Make readings self-explaining, cut cognitive load in the reports, and add guided navigation so users stop getting lost between modules.
Close the feedback loop — show operators that acting on a reading changed the outcome. Reinforcement, in-flow nudges and visible wins are what turn a delivered report into a daily behaviour.
What changed — and what I'd caveat
An audit succeeds when it changes decisions, not slides. Here's what moved, stated honestly, without inflating a research project into a revenue miracle.
Module health shifted from installs-and-renewals to floor-level active usage — the false-positive metric the whole audit exposed was retired as the yardstick.
Stability and report readability moved ahead of new custom builds; customization requests were reframed as signals to validate, not promises to fulfil.
The long-standing "change the format and they'll use it" theory became an instrumented experiment — the requested change shipped with usage tracked before and after, instead of shipped on faith.
The Perception-vs-Reality synthesis became the shared evidence base for invest / fix / hold conversations across the three modules.
Depth interviews were few by design; confidence comes from triangulation across three independent data sources, not from n.
Mitigated — not eliminated — by the case-based method: one real event, reconstructed with its artefacts, never opinions or averages.
The emotional journey is coded synthesis, not a measured instrument. It maps the shape of the gap, not its magnitude.
Not every module was equally tracked — the audit itself surfaced event-tracking gaps, which capped how far some behavioural claims could go.
A product can pass every dashboard and still be failing on the floor. This audit built the apparatus to see that — and gave the team permission to act on it.
The researcher's real job is to find the gap between what people say, what they buy, and what they actually do — and to name it plainly, even when the numbers look comfortable.
