Speak
Each of these is a real decision Speak is working through, written so you can make the call yourself — incomplete information, someone who pushes back, no answer key.
Speak is a subscription-based, mobile-first language-learning platform that helps individual learners and workplace teams build practical speaking ability through structured lessons, phrase practice, and conversations with an AI tutor. Its focus is voice-first practice with immediate feedback on pronunciation, grammar, and word choice. Speak for Business (S4B) is Speak's enterprise arm, which helps employees get comfortable speaking the languages their business success depends on. Speak for Business has grown to hundreds of logos globally, with its deepest footprint in Korea and Japan. Speak reached a $1B valuation with its Series C round as of December 2024. Speak has a team of more than 90 people based throughout San Francisco, Seoul, Tokyo, Taipei, and Ljubljana.
Your first one is free. No card.
37 live right now.
The rest
Target the next lesson after feedback stops feeling personal
Learners receive specific corrections, yet their next speaking lesson still feels generic. You have to weigh a useful prediction against content quality and an overconfident learning claim.
Choose live feedback before the wider rollout locks it in
Reviewed learner turns show both needless corrections and missed errors in live conversations. You have to weigh wider access against feedback learners can trust.
Trace speech failures before Android triage misses the release window
Learners are repeating answers or leaving exercises when speech-to-text records a different sentence. You must decide what AI can rank from traces and what engineers must confirm before a fix ships.
Preserve hard exchanges so learners can retry them later
Learners cannot return to the exact spoken exchange where they froze after a session ends. You have to weigh a quick retry path against preserving the flow of live practice.
Protect Spanish pauses while the live tutor rollout expands
Spanish learners are being interrupted during live answers and some leave before the lesson ends. You must weigh faster tutor feedback against giving learners enough space to speak.
Prevent double charges while the paywall test stays live
Some annual subscribers are charged twice after payment approval, and their entitlement records do not always match. You have to protect payment correctness without throwing away the experiment’s conversion signal.
Launch a peer cohort of 150 learners without replacing AI feedback
Early English learners are finishing lessons but losing the reason to return and speak again. You have to weigh peer momentum against a community space that stays safe and useful.
Steer the cancellation thread before a promise commits the team
A public thread is mixing disputed renewal charges with doubts about the tutor, and new replies are spreading the concern. You have to weigh fast reassurance against promises CX and Product can actually stand behind.
Rebuild first-conversation confidence after the tutor mishears learners
New English learners are leaving after the tutor responds to a phrase they did not intend. You have to weigh fast reassurance against collecting feedback precise enough to guide follow-up.
Shape a reflection flow without claiming a proficiency score
Learners value individual corrections but still cannot tell whether their speaking is improving. You have to weigh a motivating progress story against evidence the app can honestly support.
Decide how learners correct the tutor once Japanese voice assets lock
Beta recordings show learners quitting after recognition mismatches that machine summaries often miss. You have to weigh a recoverable conversation against latency, localization lock-in, and a fixed release date.
Clarify the mic state before learners leave their first response
Forty-one percent of first-time learners leave the recording step without submitting. You have to weigh a more explicit live recording state against a focused lesson flow.
Settle which Taiwan test to run when Meta Ads and TikTok disagree
Meta Ads and TikTok disagree on which Taiwan messages bring learners who start speaking practice. You must decide what AI can recommend and what the strategist must still judge.
Win 1,200 paid subscribers from travel-bound learners in eight weeks
Travel-bound learners are comparing practical speaking practice with cheaper AI conversation apps. You have to weigh immediate departure urgency against a subscriber path that lasts beyond the trip.
Save promotion renewals while the price promise is still live
Promotion cohorts are cancelling before their second charge, putting the seasonal LTV:CAC target at risk. You must weigh a clearer value journey against a stronger price intervention when only one can use the remaining budget.
Fix the TikTok path to paid practice before the weekly budget shift
TikTok first-week subscription conversion is 4.1%, versus 5.5% required to hold the channel’s economics. You have to weigh a faster message change against preserving an honest first-session expectation.
Build a value review for 6 customers so renewals rest on evidence
Administrators can see learner activity but cannot agree on what proves workplace value. You must weigh a useful value story against claims the programme cannot support.
Defend a renewal when completion no longer proves confidence
A 4,000-seat business program may not renew because executives dispute what learner progress proves. You have to weigh a fast commercial save against claims the evidence cannot support.
Activate finance learners before their participation review
Invited finance employees opened the launch message but have not started speaking practice. You must weigh a focused activation push against the administrator’s single outreach window and the promise the app can actually keep.
Decide what Cantonese help to build instead of localizing first
Reviews request Cantonese support but do not say which learner moment needs it. You must weigh a quick localized response against evidence of what prospective learners will understand and choose.
Commit a placement score while subgroup evidence diverges
Comparable expert-rated speakers show a 0.18 score gap by learner background. You must weigh a fixed expansion date against a score interpretation that learners may carry into placement.
Decide if scoring or wording is at fault before the QA window
312 of 487 recent exits happen within ten seconds of feedback. You must weigh a fast copy change against evidence that the scoring signal itself is wrong.
Clarify CRM variants while lesson promises split across markets
Localized CRM messages are making different promises about practical speaking practice. You have to weigh faster variation against a review boundary that keeps every launched asset credible.
Win trust from career-driven learners without implying a human coach
Career-driven English learners are treating AI speaking apps as interchangeable and asking what proves progress. You have to weigh a sharper career promise against what the lesson experience can honestly support.
Choose teacher or AI voice for Korean ads before media buys begin
The campaign must raise qualified trial starts by 18% without pushing first-lesson complaints above 2%. You have to weigh a stronger conversational promise against what locked teacher footage can honestly deliver.
Match the ad's promise to the first lesson before the swap
Spanish landing-page trial starts are at 8% after learners see the first lesson. You have to weigh a more immediate speaking promise against an honest picture of structured practice.
Shape work practice before learners settle for generic conversations
Career-motivated learners leave general lessons when they cannot rehearse the conversation ahead of them. You must weigh immediate relevance against a learning path that builds real spoken ability.
Choose a progress signal before reports rewrite learner history
Learners and workplace customers are receiving conflicting explanations of English speaking progress. You have to weigh a more responsive learner-state model against evidence that stays credible after reporting begins.
Tune Spanish corrections while learners leave mid-practice
Learners are leaving Spanish speaking exercises after feedback they believe is wrong. You have to weigh tighter feedback quality against catching the errors learners need to practice.
Choose the next test for Korea leads once the labels are mapped
Campaign evidence points in different directions, and Sales questions whether the leads are ready for enterprise conversations. You must decide what AI can recommend, what strategists must judge, and how the team will trust the evidence.
Prove the demand for evidence while Sales reviews 15 leads
Korea paid programs are drawing broad inquiries while L&D administrators need evidence for finance reviews. You must decide what evidence-led motion to test without overstating what the program can prove.
Choose a pipeline path while attribution splits the same accounts
The remaining budget must produce 18 sales-accepted opportunities and ₩900 million in qualified pipeline this quarter. You must weigh a faster event-led route against a paid channel playbook that could scale beyond Korea.
Rework Naver leads while Sales rejects reporting-guide inquiries
Paid leads are filling forms but too few become sales-accepted within 14 days. You have to weigh lower-cost volume against a path Sales can qualify quickly.
Package customer-call practice without making every deal bespoke
Enterprise buyers want speaking practice for real customer conversations, not a broad language catalog. You have to weigh repeatable revenue against the specificity that gets L&D leaders to commit.
Close the 8,000-seat deal without penalties tied to learner progress
An 8,000-employee agreement could slip as procurement seeks remedies tied to learner outcomes. You have to weigh a quarter-end close against commitments the program can genuinely support.
Win the reporting review before procurement closes
A 2,000-seat opportunity is stalled because the buyer needs evidence beyond active usage. You must weigh a compelling outcome case against promises the current reporting cannot support.