01 — What These Two Rounds Actually Are

Assessment #4 (Communication) and the Final Interview · Accenture India campus, 2027 batch Compiled 2026-09-17 from 40 local claims and 16 online sources. Read this before anything else.

The short version

Assessment #4 is a ~30-minute AI-graded English test that can eliminate you, delivered in Chrome on a separate date. Accenture publishes exactly what it scores. It is half listening, and it has a written email section that every prep site on the internet says does not exist.

The interview is one person, about 30 minutes, the last round. They decide on your overall performance across the whole process. Four of the questions asked are known verbatim.

The single biggest change from the predecessor handoffs: Accenture's own campus support page answers most of what three previous runs left open. It was never opened. It is the most authoritative source in this project and should be re-read the week of the test.

Part 1 — Assessment #4: Communication

Confirmed mechanics

What it scores — the official parameter list

This is the rubric. Everything in the communication library is built against it.

Two things worth noticing.

Two of the seven verbal parameters are comprehension, not production. Active Listening and Spoken English Understanding are scored separately from how you sound. A meaningful part of this round is simply hearing accurately — which is trainable, and which most candidates never practise.

The written group has no parameter for length or speed. It is Etiquette (register, greeting, sign-off, politeness), Content (did you actually address the task), and Grammar. Nothing else.

The sections you will probably see

Ranked by how good the evidence is. Prepare in this order.

How many sections there actually are is unknown. The "seven sections" in the handoff is a union of two partial accounts; one candidate said "about 5–6", the other listed 3, then 4. Prepare the union.

The email section — the best-evidenced and most coachable part

REV-001 named the two topics: "Error reporting" and "Pitch based".

Vendor documentation for this class of test (Pearson Versant, used as a format model only) specifies an email task that matches the reported shape almost exactly — 2 items, 9 minutes, at least 100 words, a short workplace scenario naming the recipient, and three themes you must address. Responses under 30 words or off-topic score the minimum. The reported "two topics" and the vendor's "2 items" converging independently is the strongest structural signal in this library.

This is where the marks are. It is the one section with a published rubric, a known genre pair, a predictable skeleton, and enough time to think.

Timing

~10 seconds per short question (REV-001, hedged) and 5–10 minutes for the email — where REV-001 explicitly said he could not remember accurately. The vendor figure is 9 minutes. Practise the email to a 5-minute floor so a shorter allowance cannot hurt you.

The failure mode that should worry you most

Not a low score — a withheld one.

"When any noise from the surrounding is audible, system might prevent you from receiving a score." — Accenture "…the speaking score may not be available" due to "very soft voice, whispering, or background sounds" — vendor documentation

Two independent sources say the same thing: bad audio does not cost you marks, it costs you the score. Wired headset, quiet room, nobody else present, speak up.

On accent — the question the predecessors left open

A regional Indian accent is not itself penalised. Do not try to sound American.

Accenture's parameter list names Pronunciation and Articulation — never accent. The speech recognisers in this product class are explicitly optimised for non-native speech and calibrated against non-native distributions.

But be precise about the limit: pronunciation is scored, and scoring works by checking whether expected words are recognised in the expected sequence. An accent heavy enough to make words unrecognisable, or a mumbled or rushed delivery, will cost you. The mechanism is intelligibility, not preference.

What is actually measured: response latency · speaking rate · position and length of pauses · word stress · segmental pronunciation. So: start promptly, hold an even pace, pause at clause boundaries rather than mid-phrase, and finish your words.

Part 2 — The Final Interview

Confirmed mechanics

The questions actually asked

Four are known close to verbatim from a selected candidate:

1. Walk through your projects — expect depth on at least one.
2. "Imagine you live in a society and you want to build a project that helps people living in it."
3. What certifications or courses have you completed?
4. "What tech are you most comfortable with, and can you switch between tech stacks within a month?"

Plus standard HR and cultural fit. And from a second selected candidate (Oct 2025): databases and operating systems came up and they credit it for their selection.

How technical is it, really?

REV-001: only 2–3 candidates per campus got genuinely technical questions, "not in depth" — but also "depends on the interviewer, be ready for both." The Oct 2025 account credits DBMS and OS knowledge.

Read it as: probably not deep, possibly present, and the variance is the panelist. Do not skip CS fundamentals, but do not build the prep around them either. The prep is your projects.

Does a weak coding round follow you in?

Not directly answerable. But the final decision is made "based on the overall performance", and REV-001 noted "each round had separate evaluation criteria." The honest reading: prior rounds almost certainly feed the final decision, but there is no evidence the panelist reads your code. Do not raise it. Be ready to discuss your approach if asked.

Part 3 — The honest uncertainty list

Everything below is genuinely unknown. Nothing in this library should imply otherwise.

Assessment #4

1. The cut-off value. Confirmed to exist; never published.
2. How many sections, and how many questions in each.
3. Sample Accenture speaking topics. None exist anywhere online. The practice topics in this library are authored, not observed.
4. Sample Accenture email prompts. Same — the two genre names are all that is known.
5. Which grading product Accenture uses. Unresolved, and largely immaterial: prepare to the parameters Accenture publishes.
6. Whether "10 seconds" covers reading plus answering, or answering only.
7. Whether a failed recording can be retried.
8. Whether you are told your score.

Interview

9. The 12-day vs 20–25-day timeline conflict.
10. Whether the panelist sees round #3's output.
11. What weight the interview carries within "overall performance".

Structural

12. Every candidate source describes the batch-2026 drive. There is no 2027-cycle account anywhere. The process could change.
13. n = 2 for Assessment #4; n = 2 for the interview. One campus, one cycle, two voices.

Part 4 — The prioritised prep plan

Ordered by marks-per-hour, given the rubric and the evidence.

Assessment #4

Non-negotiable: fluency improves by repetition under time, not by reading. Every item above except #1 needs a microphone and a recording, not a highlighter.

Interview

Part 5 — What happens next

Phases 4 and 5 build the two libraries against exactly this picture.