Nobody wants a parts list. They want the thing the parts do.

01 — The problem

The bottleneck is the spec, not the purchase

A custom PC is one of the few purchases where the customer is expected to design the product before they can buy it. That expectation is where the sale stalls.

Ask someone in Bangalore how they bought their last desktop and you get a story about time, not money. Four evenings of YouTube build guides. A PCPartPicker list abandoned halfway because a compatibility warning appeared and they didn't know if it mattered. Two Reddit threads with contradictory advice. Then WhatsApp messages to three shops on SP Road, each of whom replied with a different parts list at a different price, none of which could be compared to the others.

Meanwhile, the shop is stuck on the other half of the same problem. An inquiry lands as "budget 1.5L gaming pc suggest". Before anyone can quote, someone has to ask: at what resolution, what refresh rate, do you already own a monitor, is this only gaming, do you need it this week. That is four messages and often a day of latency — and the customer is asking two other shops the same questions in parallel.

Stated as one sentence

What this project deliberately does not solve

Keeping the problem narrow was the first design decision. The Store already handles catalogue and pre-builts. This project covers only the stretch between intent and a quotable inquiry.

• No cart, no payment, no order tracking.
• No live inventory or real-time pricing feed.
• No automated compatibility validation engine — a deliberate call, defended in section 17.
• No attempt to remove the human from the sale. The human is the product; the flow exists to make their first call a good one.

02 — Why it costs money

A slow decision is a lost one

Custom PC buying in India is a comparison market with no switching cost. A buyer who is undecided is simultaneously in conversation with two or three other builders. Every hour a quote is delayed is an hour a competitor can answer first — and the shop that answers first with a specific configuration tends to win, because being specific reads as competence.

So the business case isn't "make a nicer website." It's narrower than that:

Cost of a vague lead

Four clarifying messages, roughly a day of round-trip latency, and a real chance the customer has already committed elsewhere by the time a number arrives.

Cost of an abandoned config

The buyer doesn't come back to the site. They go back to YouTube, which is not where a sale happens.

Cost of the wrong build

A returns conversation, a reputation hit in a market that runs on word of mouth and Instagram, and an assembler's day.

What good looks like

An inquiry arrives already containing purpose, budget and preferences — enough to quote on the first call, not the fifth message.

03 — Research approach

Talking to both ends of the same conversation

The problem lives in a handoff, so research had to cover both sides of it: people trying to buy, and people trying to sell. Everything was qualitative and small-sample by design — the goal was to find the shape of the decision, not to measure it.

Methods

The teardown told us what not to build

Of the seven configurators reviewed, five opened with a component choice — usually processor or motherboard. Two opened with a pre-built category. None opened by asking what the machine was for. Four hid the price until the final step. Every part-picker-style tool surfaced compatibility errors as red warnings without explaining consequence, which in the interviews turned out to be the single most common point of abandonment.

That gap was the opening. A salesperson at the counter never starts with the motherboard. They start with "what will you use it for, and what's your budget?" — and then they do the translation themselves. The product had to do what the salesperson does, in the same order.

04 — What people said

Nine conversations, one recurring sentence

Different words, same complaint: the tools available made them feel like they were being tested on knowledge they'd come to acquire.

05 — Findings

Five things that changed the design

1. People arrive with a job, not a parts list

Every single buyer described their need in terms of an activity before they described any hardware — and they did it unprompted, in the first thirty seconds. The parts vocabulary only appeared later, and only for people who had already done research. Purpose is the natural first question because it's the one people can already answer.

2. Expertise is per-component, not per-person

This one reframed the whole product. I went in expecting to design two modes — one for beginners, one for experts. The interviews killed that idea. The design engineer knew processors cold and guessed at GPUs. The 3D artist knew GPUs cold and had no opinion on RAM timings. Confidence was uneven within each person.

3. Budget is spoken first and shown last

In the WhatsApp log review, budget appeared in the customer's opening message in the large majority of threads. In the competitive teardown, most configurators revealed price only at the end. That inversion is the source of a lot of wasted effort: people configure a machine they can't afford and then start over, or abandon.

4. Compatibility warnings punish curiosity

Part-picker tools are built for people who already know the rules. For everyone else, a red compatibility warning is a failure state with no recovery path — you're told you're wrong without being told what to do instead. Several first-time buyers described this as the exact moment they gave up and went back to watching videos.

5. Nobody wanted to check out

Not one buyer said they would have paid for a ₹1,20,000 custom machine through a web form without speaking to a person first. They wanted a human to sanity-check the build and confirm the parts were actually in stock. This was the most consequential finding, because it changes what the product is for.

Which means success isn't conversion to purchase. Success is a submitted inquiry that contains enough for the shop to quote on the first call — and a call that happens fast.

6. On the shop's side: leads die in the scroll

Inquiries arrived across WhatsApp, Instagram DMs, phone calls and walk-ins, and lived nowhere durable. There was no shared list, no status, no way to know whether a lead from Tuesday had been followed up. The staff weren't disorganised; they had no object to be organised about. That is the entire reason the admin console exists.

06 — Personas

Four people the product has to satisfy at once

Three buyers spanning the confidence range, plus the person on the other side of the counter. The tension between the first and the third — hand-holding versus getting out of the way — is the central design problem.

07 — Jobs & how-might-we

The jobs, written the way people said them

Adithya

When I want a PC for the games I play, help me get a build I can trust and explain, so I don't feel stupid spending my family's money.

Meera

When my work spans CAD, simulation and model training, help me get one machine that respects all three, so I'm not silently optimised for gaming.

Nikhil

When I already know my parts, help me get a price and stock answer fast, so I can compare shops and move on.

Karthik

When inquiries come in through the day, help me see who to call and what to quote, so nothing sits unanswered overnight.

Which became four questions

• How might we let someone describe a job instead of a parts list — and still produce something specific enough to quote?
• How might we support a beginner on one decision and an expert on the next, inside a single flow?
• How might we make budget a frame for the decision rather than a verdict at the end of it?
• How might we hand the counter an inquiry that can be quoted on the first call?

08 — Feature set

What earned its way in

The tempting features were the impressive ones: a compatibility engine, a 3D case preview, live pricing. Each was cut or deferred because it didn't serve the one problem. What survived is small, and each piece maps back to a finding.

The list, with the finding that justifies each

09 — Information architecture

Where the builder sits in the site

Global Computers already had a content-heavy site — store, location pages, build guides, a blog. The configurator had to be reachable from all of it without becoming another content page. It lives at the top of the home page as the primary call to action, and as a direct link the staff can paste into a WhatsApp reply.

10 — User flow

Eight decisions, two of which can be delegated

The critical path is drawn in red. Every grey branch is an exit the design has to survive — someone who selects nothing, someone who doesn't know, someone who wants no graphics card at all.

11 — Journey map

Where confidence drops, and what we put there

Mapped against Adithya, the least confident of the three buyer personas, because he fails first. The trough sits in research — days before the product is ever opened, and the reason the product opens with a question instead of a catalogue.

12 — Sketches

Arguing with paper first

Three questions had to be settled before anything went into a design tool: how many steps, where the price lives, and whether the beginner and the expert get different screens. Paper settled all three in an afternoon.

13 — Low fidelity

Structure, before anything is allowed to look like anything

Greyscale, no brand, no photography. The only questions at this stage were reading order, what stays pinned while you scroll, and whether the counter's triage fits above the fold.

14 — High fidelity

Where the two rooms appeared

The most consequential visual decision came late and was almost an accident. The purpose step, styled dark like the rest of the marketing site, looked superb. The configuration step, styled the same way, was exhausting — dark panels, many controls, dense text, an hour's worth of reading. So the product splits: dark where you're being persuaded, light where you're working.

15 — Final UI

The shipped product

Six screens for the buyer, one for the counter. Read in order, they are the argument: ask what it's for, ask what it costs, let people delegate what they don't know, and never hide the total.

16 — Admin console

One screen the counter can work from

The buyer-side product is worthless if the inquiry lands somewhere nobody looks. The desk was designed against a single test: can Karthik decide who to call next, and what to say, without opening anything?

What production is already telling us

Three observations from the live desk. The sample is four inquiries — far too small to be evidence of anything — so these are flagged as things to watch, not findings.

17 — Decisions defended

Five calls worth arguing about

Two steps, not one page and not eleven

One page would have reproduced the tax-return form that was crossed out on the first sketch sheet. A long wizard would have multiplied the exit points. Two steps maps to how the conversation actually goes at the counter: establish the purpose, then build inside it. The split also creates a clean commitment point — once you've answered step 1, you've invested something, and step 2 opens with your answer already reflected back at you.

No compatibility validation engine

This is the decision most likely to be challenged, and the reasoning is finding 4. A validator is a policeman: it tells you that you're wrong after you've already chosen, with no route to being right. It also only helps people who understand the rule being enforced. The alternative is to curate — offer models within a platform that are already sane together, keep the questions at the level of capacity and purpose, and let a human engineer resolve the rest. The trade-off is real: someone can still submit an unbalanced combination. But it arrives as a conversation rather than a dead end, and the assembler was going to review it anyway.

Dark for the showroom, light for the workbench

Consistency was the obvious argument for keeping the configurator dark, and the prototypes killed it. Dark works for a hero with one decision and a photograph. It does not work for eight stacked control groups that demand sustained reading. The mode change also does useful signalling work: crossing from dark into light marks the transition from being sold to, to working. Risk acknowledged — it can read as two different products. The persistent header, the purpose chips and the shared red hold it together.

"Help me choose" instead of a beginner mode

A mode toggle asks people to classify themselves, and finding 2 says they'd classify themselves wrong, because expertise is uneven within a single person. Per-component delegation lets Meera be an expert about processors and a beginner about graphics in the same session, with no ceremony. It also produces better sales data than a mode switch ever could: every delegation is a labelled point of low confidence.

The goal is an inquiry, not an order

Finding 5 reframed the success metric. Adding a cart would have been the conventional move, and it would have optimised for something nobody in the research wanted to do. Designing for a qualified inquiry meant the summary panel's job is comprehension rather than conversion, the button tells the truth about what happens next, and the real finish line is on the other side of the handoff — a first call that opens with a recommendation.

18 — How we'll know

Targets, not results

19 — Limits & next

What I'd hold this work to

Honest limitations

• Small, qualitative sample. Nine conversations and a WhatsApp log review shape the design well and prove nothing statistically. Everything here should be read as directional.
• Bangalore-shaped. The WhatsApp-first contact model, the rupee presets, the same-day assembly promise and the two-store footer are tuned to one city. Some of it wouldn't survive being copied elsewhere.
• No accessibility audit yet. The purpose cards rely on photographs and a colour-plus-check selected state; the check mark carries it, but contrast on the dark surface and full keyboard operation of the card grid need a proper pass against WCAG AA.
• Mobile is untested in the wild. The layouts stack, but the pinned summary rail is the component most at risk on a small screen, and it's the one carrying the most weight.
• No result claimed. Uninstrumented at launch. The targets in section 18 are a plan, not an outcome.

Next, in order

1. Instrument everything in section 18 and establish a real baseline before touching the design again.
2. Save and compare builds. Already evidenced by duplicate submissions in production; removes the workaround people invented.
3. Give the status field a job — an overdue list driven by time in New, so triage has a consequence.
4. Return the quote on WhatsApp as a structured message, closing the loop on the channel the customer chose.
5. Accessibility and mobile pass, starting with the summary rail's behaviour on a phone.
6. Feed the delegation map back into content. If most people delegate graphics, the guide pages should explain graphics — the configurator becomes a research instrument for the rest of the site.