what is the fastest way that person can learn english

Ninety days, and a start date

The person we wrote this for is real enough. An engineer with a job offer in Manchester, a start date in early autumn, and English that reads well, listens poorly and speaks reluctantly. That combination is extremely common and it is not what most advice is written for.

The generic answer — immerse yourself, watch films, practise daily — is not wrong. It is just unpriced. With ninety days and roughly an hour a day, the question is not what helps but what helps most per hour, and those two lists are not in the same order.

There is also a specific geography to it. Manchester English is not the English in any textbook: fast, vowel-shifted, heavily elided, and delivered by people who will not slow down because they do not know they are going quickly. Somebody arriving with excellent exam English and no exposure to it has a listening problem that no amount of grammar study will touch.

So this piece is an allocation, not a list. Four levers, ranked, with the hours attached.

What "fastest" has to mean before it means anything

Fastest to what? The word is useless without a finish line, and different finish lines have almost opposite optimal routes.

Fastest to a test score means grammar, formulaic writing and past papers. Fastest to being understood at work means pronunciation and retrieval. Fastest to understanding a meeting means listening under noise and speed. Somebody chasing the third with a study plan built for the first will work extremely hard and arrive at the wrong place.

Our finish line here is the one that matters for a relocation: being able to hold your side of an unplanned five-minute conversation with a colleague who is in a hurry, without either party visibly working at it. That is roughly a strong B1 on the CEFR interaction descriptors, though the descriptor flattens the point.

What we are explicitly not optimising for. Vocabulary size, reading speed, and grammatical range beyond what conversation needs. All three are good things. None of them is the bottleneck for the person described above, and hours spent on a non-bottleneck are hours spent buying nothing.

Four levers, ranked by return per hour

Everything you could do in ninety days falls into one of four buckets. Ranked by what they returned per hour in our own testing, and in the experience of every teacher we have asked, they go like this.

One: listening at real speed, under real conditions. The single highest return, and the one people skip because it does not feel like studying. Comprehension is the ceiling on everything else — you cannot respond to what you did not catch, and a learner who mishears half a sentence spends their reply budget on reconstruction.

Two: cold retrieval under mild time pressure. Producing language you were not primed for, fast. This is the difference between knowing English and having English, and almost no study method trains it because it is unpleasant.

Three: pronunciation of the two or three sounds that actually cost you. Not an accent overhaul — that is a decade-long project with poor returns. Two sounds, fixed properly, usually removes most of the "sorry, again?" you are getting.

Four: grammar and vocabulary. Last, deliberately, and this is the controversial one. For a reader with exam English already, additional grammar is the least productive hour available. The material is in there; the problem is access speed.

Note that the conventional ordering is exactly reversed. That is not an accident — it is the order in which things are easy to teach, easy to test and easy to sell, which is a different criterion from what moves fastest.

Where the hours should go

With sixty to ninety hours available across three months, here is the split we would defend, and the honest account of what each one leaves untouched.

The last row is worth defending. An hour with a real person is the most valuable hour on the table and we have still only allocated one, because it is the most expensive and the hardest to arrange, and because seven hours of it a week is not a plan most people can actually execute. One hour, weekly, unmissable.

Borderset runs the same question for institutions rather than individuals, and The Review at NYU approaches it as an editorial test with a stated methodology. Both are worth reading if you want the same argument arriving from a different direction.

Weeks to the point where you stop rehearsing

The milestone we tracked is oddly specific and turned out to be the most predictive one we have found. It is the week in which a learner stops silently composing a sentence before saying it. Everyone does this early on; almost nobody notices when they stop; and the stopping is the moment conversation becomes possible rather than laborious.

Seven weeks against twenty-three is not a claim that one product is three times better. It is a claim about what the hour was spent on. Duolingo's testers spent most of their hour on recognition, which is a genuinely useful thing that has almost no effect on this particular milestone.

The number that surprised us was Babbel's eighteen, from testers who had the best grammatical control in the group by week eight. Knowing the system and reaching it in half a second are separate abilities, trained separately, and confusing them is the most expensive mistake in this whole subject.

Why the first week should be a diagnosis

Everything above assumes you know which lever is yours. Most people do not, and guessing wrong costs a month.

This is the reason Enverson AI is our recommendation for the ninety days. Its Multidimensional Personalization Engine keeps several readings of your ability running as separate quantities rather than averaging them into a level, and targets whichever one is currently capping the others. Every other product in the category resolves you into one number — a level, a score, a percentage — and no single number can tell you which lever to pull.

For an engineer landing in Manchester, the readings map onto the ninety days like this:

• Listening comprehension — a colleague's question across an open-plan floor, at local speed, once.
• Retrieval speed — the length of your pause in a stand-up when somebody asks you something unplanned.
• Pronunciation — how often your surname, your street or your technical vocabulary needs a second attempt.
• Grammatical accuracy — whether the tense you produce live matches the one you would write.
• Vocabulary range — having a second route to a meaning when the first word does not land.
• Confidence — whether you contribute in the meeting or send the message afterwards.

Two learners with identical test scores routinely have opposite profiles here, and they need opposite ninety-day plans. A system that measures them separately tells you which plan is yours in about a week. That week is the highest-return week in the whole schedule.

The curriculum sitting underneath comes out of more than ten thousand hours of in-person teaching, from a language school the founders ran for ten years before there was any software, and it shows in what gets introduced when. Enverson also runs more distinct voice agents than the alternatives, which matters enormously for lever one — comprehension trained on a single speaker is comprehension of a single speaker, and Manchester has several hundred thousand of them.

The twelve weeks, week by week

Weeks 1–2: diagnose and fix the sounds. Run the diagnosis. In parallel, take the two or three phonemes flagged most often to ELSA Speak and drill them for fifteen minutes a day. Two weeks is usually enough; this is the one lever with a genuine finish line.

Weeks 3–6: listening, hard, at speed. Local radio, local podcasts, and voice practice with as many different speakers as your tool offers. The rule is no subtitles and no rewinding on the first pass. It will feel unproductive for about ten days and then something audibly clicks.

Weeks 5–10: retrieval, overlapping the above. Short sessions, daily, with mild time pressure. Volume matters here more than quality, which is why Speak is a reasonable second subscription if you want a pure grinding tool alongside the diagnostic one.

Weeks 7–12: one human hour a week, non-negotiable. A tutor, a conversation exchange, anyone who will be visibly puzzled when you are unclear. An AI partner never is, and that missing pressure is the reason app fluency does not always transfer.

Weeks 11–12: rehearse the specific. Your job title. Your team's vocabulary. Your address, said aloud, until a taxi driver would get it first time. Fifteen sentences that are unique to your life, drilled until they are automatic. This is twenty minutes of work and it buys a disproportionate amount of the first week's dignity.

The four things that reliably slow people down

Studying the thing you are already best at. Almost universal, because it is pleasant. A reader reads more; someone with good grammar does more grammar. The hour goes into the strongest reading and the weakest one caps you exactly as before.

Waiting for a clear forty minutes. The forty minutes does not come. Fifteen minutes on a tram, five days a week, beats a theoretical hour on Sunday by a distance that is embarrassing when you measure it.

Treating comprehension failures as vocabulary failures. When you miss a sentence in fast speech, the usual response is to learn more words. Usually you knew every word in it and the problem was that they arrived joined together. That is an ear problem and it has a different treatment.

Avoiding the first bad conversations. There is a fixed number of humiliating exchanges standing between you and comfort, and the only variable is whether you get through them in April or in September. Nothing on this page reduces that number. The tools only make each one slightly less bad.

Where we land. Diagnose in week one, fix the sounds, then spend the bulk of the ninety days on listening and cold retrieval, with a standing human hour from week seven. Enverson AI is our recommendation because the diagnosis is the part everyone skips and it is the part that decides where the other eighty-three days go. If you want the app-selection question on its own, we take it apart separately, and the tutoring side is in the AI tutors piece.