From Berlitz to AI: How Sales English Training Evolved (2015-2026)
From Berlitz to AI: How Sales English Training Evolved (2015-2026)
Sales English training — the category of language and communication development specifically focused on helping non-native speakers sell effectively in English — has gone through four distinct generations in roughly two decades. Each generation solved a specific problem with the previous one. Each introduced its own limitations. The current generation, built on AI-driven simulation, is the first to address the fundamental variable that all prior generations missed: practicing under real-time pressure with immediate feedback.
This article traces the full evolution: what each era offered, who it served, what it failed to deliver, and where the field is in 2026. Platforms like DojoSales represent a specific position within the current generation — built specifically for the intersection of language fluency and sales performance, which earlier generations addressed separately if at all.
The Berlitz Era (2000-2015): Classroom and Teacher
Berlitz was founded in 1878 and spent its first 120 years building the dominant global model of language instruction: immersive, teacher-led classroom sessions. By the early 2000s, Berlitz had developed a corporate training division serving Fortune 500 companies with language training for international business travel, cross-cultural communication, and professional communication in English.
The model worked well for what it was designed to do: give non-native professionals enough spoken English to function in international business settings. A German engineer attending a meeting in London. A Japanese executive presenting to a US board. A French lawyer negotiating a contract in New York. The Berlitz model gave these professionals enough foundational English and enough cultural orientation to navigate these situations without major embarrassment.
What it did not do:
It did not address sales specifically. The corporate Berlitz curriculum was business English — general professional communication. Sales conversations have a structure, vocabulary, and dynamic that differs from general business conversation in important ways. Objection handling, pipeline language, cold opening techniques, closing sequences — none of these appeared in standard business English curriculum.
It did not simulate pressure. Classroom instruction, by design, is low-stakes. The teacher is supportive, the pace is controllable, and the student knows they cannot lose a deal in the room. The pressure that produces call anxiety — and that training needs to simulate to produce transfer — was absent. Skills learned in the classroom did not transfer reliably to high-stakes call environments.
It was expensive and slow. A typical corporate Berlitz engagement ran 40 to 80 hours of instruction over several months, at rates that made it inaccessible to individual contributors. It was a management perk, not a rep skill-building tool.
By 2015, the rise of affordable internet connectivity and high-quality video calling had made the cost disadvantage of in-person instruction untenable. The online era began.
The Online Shift (2015-2020): Preply, iTalki, Talaera
The 2015-2020 period produced an explosion of online language tutoring platforms. Preply, iTalki, Cambly, and specialist platforms like Talaera (focused specifically on business English) democratized access to native-speaker tutoring at a fraction of Berlitz's cost.
The model was simple: marketplace connecting non-native learners with native English speakers or qualified language tutors via video call, charged by the session. A learner could book a 50-minute session with a native English speaker in the United States for $15-$40, compared to hundreds of dollars for an equivalent in-person Berlitz session.
This was a significant improvement on the access problem. For the first time, a mid-level sales rep at a European startup could afford regular spoken English practice with a native speaker. The volume of practice became achievable.
The limitations were structural:
Tutors were not salespeople. The native speakers on these platforms were generally English teachers, freelancers, or professionals with general language coaching experience. Asking them to simulate a skeptical VP of Engineering in a late-stage SaaS evaluation was outside their expertise. The "business English conversation practice" these sessions produced was valuable for general fluency but did not prepare reps for the specific dynamics of sales calls.
Feedback was general, not behavioral. A language tutor giving feedback on a simulated sales pitch would typically focus on grammar, vocabulary, and pronunciation — the language tutor's area of expertise. The sales-specific feedback — "your response to the pricing objection was too hedged, you need to be more direct about the ROI claim" — required sales expertise the tutor didn't have.
The pressure problem remained. Talking to a supportive tutor on a platform where both parties know it's practice still does not replicate call anxiety. The affective filter that blocks language access under real call pressure is not triggered by low-stakes conversations with a helpful native speaker.
Talaera made progress on the business English specificity problem — its curriculum was designed around professional communication scenarios and its tutors were selected for business experience. But even Talaera's model was curriculum-driven rather than pressure-simulation-driven.
The AI Revolution (2020-2024): Hyperbound, Second Nature, Cold Call Gym
The early 2020s produced a new category: AI roleplay platforms for sales training. These platforms replaced the human tutor with an AI buyer persona — a simulated prospect built on large language models that could respond to a rep's pitch, handle objections, and generate a realistic real-time sales conversation.
Hyperbound, Second Nature, and Cold Call Gym were among the early leaders. The value proposition was compelling: reps could run simulated cold calls and discovery calls at any hour, at scale, without scheduling constraints or human facilitator costs. The AI could play a skeptical CFO, a friendly but non-committal IT manager, or a hostile competitor-loyal buyer — across as many repetitions as the rep wanted.
This was the first generation to partially solve the pressure problem. Reps reported that AI simulations — particularly cold call simulations — did produce some level of anxiety response, more than human role-play did. The reason is likely that the AI is unpredictable: unlike a colleague playing a buyer, the AI doesn't give you verbal hints, doesn't break character with sympathy, and generates responses you haven't seen before. The unpredictability is enough to trigger at least a partial anxiety response.
What the first generation of AI sales training platforms missed:
Language-specific feedback. These platforms were built for sales training, not language training. Their feedback engines evaluated whether the rep used a good discovery question or whether their objection handling was structured correctly. They did not evaluate whether the rep's filler word density was too high, whether they were using hedging language that leaked uncertainty, or whether their contractions were deployed correctly. For native English speakers, this omission was irrelevant. For non-native speakers, it was the gap that mattered most.
The language-sales intersection. A non-native rep's call performance problem is not purely linguistic and not purely sales. It is the intersection: they know the sales framework, they know the English, but the anxiety of performing the English in a sales context causes both to degrade simultaneously. First-generation AI sales training platforms built for native speakers did not address this intersection.
Persona realism calibrated to cultural context. The buyer personas on early platforms were primarily calibrated to US market dynamics. The platform design assumed a US or native-English-speaking rep. Non-native reps using these platforms were practicing against buyer personas calibrated for a different starting context.
Despite these limitations, the 2020-2024 AI platforms established the core infrastructure: AI roleplay at scale, async access, feedback loops, and persona variation. The missing piece was the language layer.
The Missing Piece (2024-2026): Language and Sales Combined
By 2024, two adjacent markets — AI sales training and language learning tech — had developed substantial capabilities that did not overlap. Sales training platforms had sophisticated role-play, objection trees, and sales methodology feedback. Language platforms like Duolingo and Babbel had gamified vocabulary and grammar practice. Neither addressed the specific problem of a sales rep who knew both English and sales methodology but couldn't deploy them together under call pressure.
The gap became visible as more European and Asian companies scaled US-market go-to-market motions. These companies were hiring sales reps who had passed English proficiency tests, who understood US sales methodology, but who were losing deals in their first three to six months in market — and the loss wasn't attributable to product-market fit or pricing. It was attributable to call quality: specifically, to the hesitation, hedging, and freeze patterns that accent bias research and affective filter research predicted.
The commercial signal was clear: a platform that combined AI roleplay with language-specific feedback — that could identify when a rep said "euh" instead of pausing, flag the use of "I think maybe" as a credibility signal, and give feedback on whether the opening 60 seconds was confident enough to survive the accent familiarization window — would address a pain point that no existing platform was addressing.
This is the category that platforms like DojoSales were built to serve: non-native English speakers who need sales call practice specifically, with feedback that addresses both the sales performance layer and the language confidence layer simultaneously.
Where We Are Now: The AI Dojo Model
The current state of sales English training in 2026 is defined by the convergence of three capabilities that the prior eras developed separately:
Realistic AI personas — available from the 2020-2024 generation of platforms, now refined with cultural calibration, industry-specific vocabulary, and difficulty progression. A simulated buyer in 2026 is more realistic than anything available in 2020.
Language-aware feedback engines — the new addition. The ability to analyze a transcript for filler word patterns, hedging language, directness signals, contraction usage, and specific non-native language patterns (lexical calques, register mismatches) and generate surgical feedback pointing to specific timestamps and specific phrases. This is the feedback layer that makes the practice useful for non-native reps rather than just for native reps working on sales technique.
Accessible async infrastructure — the democratization contribution of the 2015-2020 era, now applied to simulation rather than tutoring. A rep in Lyon can run a simulated call with a VP of Sales in San Francisco at any hour without scheduling, without social stakes, and with a per-session cost well under what Berlitz charged per hour.
The convergence produces the dojo model: high-volume, pressure-simulated repetitions with specific feedback calibrated to the non-native rep's specific gap profile. Not language class. Not sales workshop. Deliberate practice at the intersection of both.
What's Next (2027+): Real-Time In-Call Coaching and Accent-Aware AI
The trajectory from here is visible even if the delivery dates are not certain.
Real-time in-call coaching is the next frontier. Rather than post-call feedback on a transcript, AI systems will monitor calls in real time — detecting hesitation patterns, flagging opportunities being missed, and surfacing suggested next steps through a discreet interface. Systems like Chorus and Gong already do post-call analysis; the real-time layer is the logical next step and has been in development across several vendors since 2023.
For non-native reps, real-time coaching has a specific application: the AI detects the onset of a freeze (rising filler density, response latency spike, hedging language pattern) and surfaces a pre-approved bridging phrase that the rep can deploy to recover. This is not replacing the rep's judgment — it is providing the safety net that practice builds over time, available from day one.
Accent-aware AI personas represent a different direction. Current AI roleplay systems generate responses that are accent-neutral — the AI buyer cannot hear the rep, only read the transcript. As voice-to-voice AI simulation matures (as Vapi and similar real-time voice AI platforms develop), the simulation will include full audio — and the AI persona will respond to delivery characteristics including pace, hesitation, and the specific confidence signals that the accent bias research identifies. This creates a feedback loop that no written transcript analysis can replicate.
Personalized difficulty calibration — AI systems that monitor a rep's specific gap profile across sessions and automatically adjust persona difficulty, objection complexity, and scenario novelty to maintain the deliberate practice condition of "slightly beyond current ability" — will be the norm rather than the exception by 2027.
The evolution from Berlitz to AI has been a progression toward specificity, accessibility, and pressure realism. The next generation completes that progression with real-time responsiveness and full audio simulation.
FAQ
Q: Is there still a place for human tutors and coaches in this landscape?
Yes — for specific use cases. Human coaches excel at qualitative strategic feedback, motivation, and the kind of nuanced assessment that catches patterns an AI might miss. The AI dojo model replaces the volume layer (repetition sessions that a human coach cannot cost-effectively provide) while preserving the value of human coaching at the strategic layer. The two are complementary, not competitive.
Q: How does DojoSales fit into this evolution?
DojoSales occupies the 2024-2026 position: language-aware AI sales simulation specifically calibrated for non-native English speakers. It combines the AI roleplay infrastructure of the 2020-2024 generation with a feedback engine designed specifically for non-native language patterns — filler words, hedging language, lexical calques, register mismatches — rather than only sales methodology feedback. It is accessible at $39/month versus hundreds of dollars per session for the Berlitz-era equivalent.
Q: Will AI make human language coaching irrelevant?
For high-volume repetition practice, AI is already more practical than human coaching — lower cost, no scheduling constraints, and no social stakes that discourage risk-taking in practice. For the qualitative, motivational, and strategic layers of skill development, human coaching continues to add value that AI cannot replicate yet. The honest answer is that for the specific use case — non-native reps building English call confidence — AI simulation is better than a language tutor for the practice layer, while a sales coach is better than either for strategic guidance.
For a Berlitz-specific comparison, see DojoSales vs Berlitz: What's Actually Different. For the AI sales training landscape, see Best AI Sales Training for Non-Native English Speakers. For the complete sales English foundation, see The Complete Guide to Sales English.