Eleven companies were studied. Some reached $10M in under two years; some spent a decade and stalled; two are shrinking. This page is what the difference was, written as instructions. Every step names the company that proves it and what happens if you skip it. Read it in order — the order is the strategy.
Sell it before you build it. Close the first hundred customers yourself and write down every objection. Publish your price, and collect the money before you do the work — that is what makes advertising pay for itself. Buy the search terms of people who were about to build it themselves, not your competitors' names. Win one language market completely before you touch a second. Remove the buyer's risk instead of cutting your price. Hire salespeople only when you run out of capacity, not when you run out of ideas. Then, and only then, spend money to go faster — and build the measurement before you build the spend.
Each phase has an exit condition. Do not start the next one until you have met it — every failure in this research came from moving too early.
| Phase | Steps | What you are doing | Move on when… |
|---|---|---|---|
| A · Prove it | 01–03 | Finding out whether anyone pays, with no product and no capital | ~100 paying customers closed by a founder, and every objection written down somewhere searchable |
| B · Fund itself | 04–06 | Making acquisition pay for itself, so you never need a round to buy growth | Advertising is profitable on its own cash — the money from one customer covers the cost of finding the next |
| C · Concentrate | 07–08 | Owning one market completely instead of being small everywhere | Over half your demand comes from one language market, and retention is holding |
| D · Buy speed | 09–12 | Spending money and hiring people to go faster at something already working | You are not finished — you are now running a machine. See the guardrail. |
Before anything else, be honest about what kind of company you are. If you already run paid advertising at scale, your engine is paid. If you have a founder with an audience, your engine is that audience. If you came out of a services business, your engine is your client base. Do not choose the motion that looks best on paper — choose the one you can already execute on Monday.
Take money for the thing before the thing exists. Resell a competitor's product, or sell your own from a one-page description. You are not testing an idea — you are finding out whether a real person will hand over real money, and hearing exactly why they will not.
No sales team, no playbook. The founder does the calls. Then — the part people skip — log every conversation: what they asked, what they pushed back on, the words they used. That record becomes your ad copy, your pricing tiers and your onboarding flow later.
Put real numbers on a public page. Tiers, what is included, what it costs. Not "contact us".
Sell a prepaid pack, a credit bundle or an annual plan up front — self-serve, no sales call. Cash arrives before you deliver. That cash pays for the advertising that finds the next customer, and the loop closes on itself.
When you advertise, do not bid on your competitors' names. Bid on the tools your customer would use to solve the problem the hard way — the developer platforms, the raw infrastructure, the incumbent system they are stuck with.
twilio, vapi, retell and placetel. Trengo
buys whatsapp api and whatsapp business api, all growing 35–53%. Meanwhile Ada
spends 23% of its entire paid budget on the single word cognigy, and Sierra, Fin
and Ada all bid on each other.Pick one country or one language and get more than half your demand from it. Localise properly — website, ads, support, content — not just the interface strings.
When someone hesitates, the instinct is to discount. Do the other thing: make the risk disappear. Charge only when it works. Refund if it does not. Guarantee the outcome in writing.
Not when you run out of ideas. Hire sellers once the product already converts, the pipeline is automated and you have referenceable customers — so they scale something that works rather than discovering what works.
Start with whatever optimises cash — prepaid, credits, usage. Move to committed subscriptions once you can prove customers stay. Not before.
Not a "country manager" executing a plan from head office. One person who owns the market: the website, the advertising budget, local hiring, local partnerships. Treat them as the founder of that market.
Once self-serve, sales and advertising all work, add channels — but prefer the ones your own product generates. The best lead list is the one your product manufactures for free.
fonio did everything above and reached $10M ARR in under two years. Then in June 2026 it raised $17M led by 20VC at a $140M valuation and began hiring a Head of Growth to own a "7-figure monthly marketing motion." At roughly $833k of monthly revenue, that is spending at or above what the business earns.
Two numbers make that risky rather than simply bold. Their CEO puts net revenue retention "a little under 100%" — the existing customer base is flat-to-shrinking, so every euro of growth is bought and none of it compounds. And their own Chief Commercial Officer, while hiring a revenue-operations lead, wrote publicly: "the analytics are not keeping up. I'm currently relying heavily on gut feeling and sense checking multiple dashboards that show different values… it will break soon."
Seven figures a month, sub-100% retention, and no attribution you can trust. That is the single largest risk across all eleven companies, and it sits inside the one we would otherwise copy wholesale.
Never increase spend faster than you can measure it. Before Phase D you need to know — reliably, from one source — what a customer costs to acquire, how many days the payback takes, what they are worth over twelve months, and whether the base grows or leaks without new sales.
Judge this business on return on marketing spend and payback period, not on ARR. Growth bought at negative retention looks identical to real growth on a chart — right up until the spend stops.
Each is a real company that did most things right and one thing wrong.
Stonly — Paris and Kraków engineering, an English-only website with no French marketing pages at all, two people in New York, and its largest market is the United States. It under-serves both. No round since January 2022, and it has since removed its published pricing.
The lesson: pick the market you can actually reach, then commit the whole company to it.
BlueTweak — around 480 competitor-comparison blog posts pulling "free alternative" searchers into a demo-gated, four-week-implementation, per-agent enterprise product. Average visit: 7 seconds. Zero branded search. Filed revenue down 30% from its 2023 peak; 2025 was loss-making.
The lesson: traffic is not demand. Match the funnel to the buyer, or a good product will not save you.
Trengo — eleven years old, $38M raised, $17.9M ARR, and as of 19 August 2026 its AI agent for e-commerce was still in hand-picked closed beta while a 22-month-old competitor was at $10M ARR with its own voice models. Traffic up 3.6% last month, 1.5% paid, no round since 2021.
The lesson: word of mouth is durable and it does not compound. There is a ceiling around $18M without an acquisition engine.
Ada — founded 2016, eight years before the AI-native cohort. Spent 2022–23 shrinking 40% across three layoffs while re-platforming from scripted flows to LLMs. The product transition worked; the market one did not. "350+ customers" has been cited unchanged from 2021 to 2026, and it has not raised since.
The lesson: being first in a category is worth nothing if you are rebuilding during the two years the category gets decided.
Deliberately short. Anything not on this list is a distraction at that stage.
| Phase | The one number that matters | Guardrails | Ignore for now |
|---|---|---|---|
| A · Prove it | Paying customers closed by a founder | Objections logged and searchable; how long a call takes to close | ARR, traffic, brand, headcount |
| B · Fund itself | Return on marketing spend | Payback in days; cash collected before delivery; share of revenue that is self-serve | Enterprise logos, analyst coverage |
| C · Concentrate | Share of demand from your home market | Net revenue retention; monthly churn; 12-month customer value | Number of markets, number of integrations |
| D · Buy speed | Revenue per employee | NRR above 100% before spend rises; payback still inside target; one source of truth for attribution | Headline ARR growth on its own — it hides everything above |
Revenue per employee across this set: Trengo ~$164k · fonio ~$123k · BlueTweak ~$75k and falling. Net revenue retention: Parloa 150% · Fin 146% · Ada 146% · fonio just under 100%.
If your NRR is below 100%, you do not have a growth engine — you have a customer-replacement treadmill, and paid spend makes it spin faster rather than climb.
Research notes behind the playbook — the original brief's questions, the wider vendor landscape, and the pipeline of companies not yet covered. Kept for reference, not for presentation.
Only fonio did, and it is not close. 63.1% of fonio's traffic is European (Germany 42.8%, Austria 10.4%, France 5.2%, Poland 4.7%) with every one of those markets growing 17–52%. Sierra 6.0% European, My AskAI 8.9%, Fin 3.7%, Ada 2.9%.
Two nuances worth raising. Fin is Irish-founded and 40% of its traffic is American — being a European company is not the same as selling to Europe. And fonio's third-largest market is Brazil at 10.0%, its fastest-growing at +66%, ahead of Austria and double France. If the GTM-Lead-per-market model produced that, it generalises past Europe, and Brazil is the case to interrogate.
Paid share of all traffic: Fin 31.9% · fonio 24.2% · Ada 12.9% · Sierra 7.6% · My AskAI <1%. Paid is not a segment story — Fin sells SMB through enterprise and is the most paid-dependent; Sierra sells Fortune 50 and barely uses it. What separates them is who pays the CAC: Fin and fonio have self-serve floors a card can buy; Sierra and Ada do not.
Every company in the set bids on competitors' brand names. Ada spends 23% of its paid
search on the word cognigy and also buys sierra ai and intercom fin.
Sierra buys eleven labs (22%) and decagon. Fin buys decagon,
gorgias, parloa.
fonio does something smarter. It buys twilio,
vapi, retell, placetel — developer infrastructure and incumbent
telephony. Someone searching "twilio" is not comparing AI receptionists; they are about to
build one themselves. fonio intercepts them where the competing offer is six engineer-months
rather than a rival product. Cheaper traffic, weaker competition, higher intent.
Sierra proves it can be a genuine acquisition channel, and we can measure it. 31.4% of
Sierra's traffic is organic search, and the two largest non-branded terms inside it are
bret taylor (5.41%, +69%) and clay bavor (3.70%, +236%). People search the
founder and land on the company.
But note what it requires: a former Salesforce co-CEO who chairs OpenAI's board. That is
not a channel you decide to build. The replicable version is the artefact, not the person —
tau bench, the benchmark Sierra published and open-sourced, pulls 3.60%, nearly as much as
the CEO's name.
The other model is My AskAI's: two founders with 2,168 and 1,431 followers producing 66.9% direct traffic off personal accounts while the brand account sits dormant. Small audiences, high trust, and it is the entire top of funnel. fonio's version sits between the two — Keinrath on LinkedIn daily during the founder-led phase, then handed off.
| Company | Primary channel | The expertise it rests on | Transferable? |
|---|---|---|---|
| fonio | Paid social + paid search, self-funding | Performance marketing (Benedikt Brauner) + telecom infrastructure depth | Yes — closest to us |
| Sierra | Founder reputation → organic; enterprise field sales | Two decades of Fortune 500 executive relationships | No — not buyable |
| Fin | Installed base, then paid at scale + published price war | 15 years of Intercom distribution; in-house model team | The pricing play, yes |
| Ada | Marketplace/partner integrations; vertical proof | Frontline support operations knowledge; airline systems | Vertical depth, yes |
| My AskAI | Founder audience → direct; competitor SEO; marketplaces | Indie-hacker credibility; a 2-person cost base | The SEO farm, yes |
Sierra Summit (Nov 2025) is the strongest example: a first customer conference with 8 product launches in one day, keynoted by customers' own CEOs rather than by developers — consistent with a top-down motion. Fin runs "Pioneer" and seasonal "Built For You" releases as predictable press moments. fonio and Ada show no significant event motion. My AskAI's equivalent is Product Hunt — nine launches at roughly monthly cadence.
Three distinct shapes. Sierra rents distribution — SoftBank as exclusive Japan reseller, an investor doubling as a channel — and buys geographies via four 3-to-10-person acqui-hires in seven months. Ada and My AskAI live inside the incumbent — ~25 helpdesk integrations and marketplace listings, an explicit non-rip-and-replace position that also makes the partner a future competitor. fonio built a partner programme (David Leibovitz) plus ~500 partners closed personally by the founder.
The one alternative channel nobody else has: fonio's SDRs call people who ran a test call on the website. The product manufactures its own highest-intent list. No data purchase, no cold list, and the prospect has already heard the product work.
The data supports that, and dates it precisely. Every pricing-model innovation in the category happened in a 26-month window: Fin ships outcome pricing at $0.99 (2023) · Ada switches to per-resolution (Oct 2023) · Sierra formalises outcome pricing (Dec 2024) · fonio moves credits → subscription (Feb 2026) · Ada reverses back to volume (Jul 2025). And the consolidation has already started — Salesforce agreed to buy Fin for $3.6B in June 2026. If we are entering this category, we are entering it after the pricing experiments and during the consolidation.
From the Everhelp competitive deck. This is the most useful thing in it: it puts eight vendors on one axis of revenue and headcount, and four of them sit in or near the $2–30M band the brief asked for. It also contains the single most strategically relevant find in the whole research programme so far — see the callout below.
The deck's slide title is "companies with comparable revenue operate with significantly smaller teams." The sharper framing — and the one that matches how we already measure things — is revenue per employee. On that axis the AI-native pure-plays beat the incumbents by 3–5×, and the two smallest companies in the set are among the most efficient.
| Vendor | Headcount | Rev. 2024 est. $m | Rev / employee | Traffic LTM | Monthly CAGR | Key geos | Note |
|---|---|---|---|---|---|---|---|
| Zendesk | 1,566 | 100+ | ~$64k* | 31.7M | +99% | Tier-1 41%, LatAm 10%, East Asia 8% | Axis truncated — AI line only |
| Kore AI | 1,179 | 55 | ~$47k* | 0.9M | 0% | Tier-1 22%, South Asia 17% | Flat growth |
| Sendbird AI | 289 | 40 | $138k | 2.6M | −5% | Tier-1 18%, South Asia 15% | 52% of traffic is paid search |
| Freshworks | 953 | 30 | ~$31k* | 1.7M | +2% | Tier-1 26%, South Asia 17% | AI line only — badly understated |
| Fin AI wrong entity | 121 | 20 | $165k | 0.3M | +29% | Tier-1 47%, Europe 6% | See the flag in the Fin dossier |
| Hoory AI | 26 | 2 | $77k | 0.2M | +10% | Armenia 31%, LatAm 13%, South Asia 13% | 68% of social is YouTube |
| Quidget AI | 8 | 1 | $125k | 0.0M | +11% | Tier-1 39%, South Asia 17%, Ukraine 8% | Built by SupportYourApp — see below |
| PerfectBot | 10 | 0 | ~$0 | 0.0M | −3% | Tier-1 61%, Europe 26% | Stopped operations |
| fonio (ours, for scale) | 81 | 10 | $123k | ~2.0M | 30%+ MoM rev | Europe 63% | Only European-demand business in the set |
Quidget AI is built by SupportYourApp — verified on SupportYourApp's own site: "Quidget, a customer support automation platform, was developed by SupportYourApp." SupportYourApp is a Ukrainian-founded customer-support outsourcing company: 16+ years, 250+ clients, on the 2026 Inc. 5000 list, human-led AI-powered support outsourcing across call centre, live chat, helpdesk and technical support.
That is Everhelp's own business model — and they have already built and shipped the AI product. Quidget: no-code AI agent, answers up to 80% of Tier-1 questions, 45+ languages, chat / email / voice / WhatsApp / Slack / Telegram / Viber, integrates with Zendesk, Freshdesk and Calendly, free trial, "go live in 2 minutes". Roughly $1M revenue on 8 people — the highest revenue-per-head of any pure-play in the table bar the mislabelled Fin row.
The exec-summary page asks whether AI support is offence or defence for Everhelp. Our closest structural peer has already answered it: offence, as a separate productised brand, spun out of the services business. That is now the single highest-value research target in the programme — not another US unicorn.
52% of its traffic is paid search — more than double any other vendor in either dataset, and it is shrinking at −5% monthly. $40M revenue on 289 people is a good ratio, but a half-paid mix on a declining trend is what buying growth looks like when the organic base isn't compounding. The counter-example to fonio, whose paid sits at 24% alongside a real organic and direct base.
Quidget's social is 70% Reddit and Hoory's is 68% YouTube — the two smallest companies in the table are the only ones not running a LinkedIn-first playbook. Cheap channels, chosen because the expensive one is unaffordable. Worth noting that My AskAI's social is also 100% Reddit. There is a pattern here for sub-$5M companies and it is not LinkedIn.
Screened against five criteria: $2–30M ARR · genuinely European demand (not just an EU address) · AI for customer conversations · SMB/mid-market with published pricing · alive and independent. Twenty-nine companies checked, nine survive.
| Was | Reality |
|---|---|
| aaron.ai | Acquired by Doctolib, 22 May 2024. Absorbed into Doctolib's booking stack — 3,500+ German providers at exit. Not investable, but read it as the precedent transaction. |
| Cognigy | Acquired by NiCE for ~$955M, announced 28 Jul 2025. ARR was ~$37M at exit on $165M raised — already above band and now gone. |
| Parloa | $50M+ ARR, $3B valuation (Series D, Jan 2026), 430 people, NRR 150%. 1.7× above the ceiling and now materially US-weighted. |
Which is the point of screening before researching. The DACH voice-AI market is barbelled: everything with real institutional funding has blown through $30M or been bought, and the segment closest to fonio's actual product is a long tail of sub-$1M vendors. fonio at ~$10M in under 12 months has no true DACH peer at its size in its exact category — which is itself the finding.
No-code AI telephone assistant — inbound answering, appointment booking, routing, transcripts. Same product as fonio, same verticals (medical practices, hotels, gastronomy, public sector, insurance), German-only demand, and it publishes its prices.
"AI communication hub for hospitality" — WhatsApp / email / SMS / webchat inbox plus an AI agent, sold to hotels. 1,000+ properties across 30 countries (St. Regis Mauritius, InterContinental Vienna, Le Grand Bellevue Gstaad).
Omnichannel team inbox (WhatsApp, email, voice, social, chat) with AI agents for what they call "the boring 84%" of conversations. The only company screened that combines in-band revenue, genuinely Benelux-first demand, SMB team pricing in EUR, and independence.
WhatsApp marketing, automation and AI-driven service, heavily Shopify-integrated. Customers are Austrian and German retail: Billa, Bipa, Bauhaus, Air Up, Biogena, Waterdrop — real DACH demand, no ambiguity.
AI agent for support across web chat, WhatsApp, Messenger and email, with tool-calling into back-office systems. Founded ~2017 by Alexander Wijninga.
| Company | Base | ARR | Why it's interesting | Why it's not tier 1 |
|---|---|---|---|---|
| Ebbot | Stockholm, SE | SEK 34.7M ≈ $3.6M FY2025, filed | The only audited revenue trajectory in the set — SEK 7.7M (2021) → 12.4 → 18.3 → 27.5 → 34.7M. Sweden-only demand (Åhléns, Europcar Sweden). | Sales-led, no published pricing, loss-making every year (−SEK 12.1M in FY2025), growth decelerating 51% → 47% → 26%. |
| indigo.ai | Milan, IT | est. $3–6M | Italy-only demand — the strongest single-language signal after Ebbot — and Italian statutory accounts are filed, so exact revenue and headcount are one registry lookup away. | Financials not yet pulled; event/ecosystem-led GTM (Netcomm Forum) rather than paid. |
| Dixa | Copenhagen, DK | $23.4M est. 2024 | Top of band with EUR mid-market pricing published (€89/€139/€179 per agent; Mim AI at a flat €0.35/conversation — a deliberate anti-Zendesk price). Clean acquisition history to study. | ~$155M raised incl. a $105M Series C at a $400M mark — a per-seat platform under return pressure, not an SMB performance business. Verify the UK/US traffic share first. |
| Born Digital | Prague, CZ | est. $5–9M | The leading independent CEE voice-AI vendor. 70 staff, 100+ customers across 10 countries, Czech-language content, EU R&D co-funding. | Financials entirely unverified, founders not even named on the about page, and Gulf exposure dilutes the European-demand score. Enterprise/services motion. |
| Landbot | Barcelona, ES | est. $5–10M | Right commercial shape — no-code, PLG, published self-serve at roughly €80–100/mo, WhatsApp-led. | Demand may be LatAm rather than EU (Mexico, Colombia). No round since the €6.5M Series A in Jan 2021 — five years is a flag. |
| DigitalGenius | London, UK | est. $10–18M | Strong European e-commerce logos (AllSaints, On, air up, Beauty Pie); added voice by 2026. | 13 years old, sales-led at a 10,000+ tickets/month floor, UK not EU, financially opaque, and has taken debt financing. |
| Company | Base | Reason |
|---|---|---|
| Tidio / Lyro | Szczecin, PL | Double fail — and it was on my earlier reserve list. $48.4M est. ARR (Oct 2024), above the ceiling; and SimilarWeb Jul 2026 puts US at 24.3% with pricing in USD. A Polish HQ selling a dollar-priced product to a US/global long tail. |
| Zowie | Warsaw, PL | $12.1M est. — squarely in band, but built from day one for US DTC e-commerce. Demand trap. |
| Synthflow AI | Berlin, DE | The trap in its purest form. $20M Series A (Jun 2025), but enterprise contracts start at $30,000/year in USD, no self-serve tier, no German-language GTM. Berlin address, American business. |
| PolyAI | London, UK | ~$40M ARR, $86M Series D at $750M (Dec 2025), and growth is explicitly US enterprise — US client revenue nearly tripled. |
| Moveo.AI | "Athens" | HQ is actually New York; demand is US and Brazil; only ~$3M raised. Greek engineering, not Greek demand. |
| Rasa | Berlin + US | Wrong product — a developer framework sold to enterprise engineering teams, not a conversation service. Hiring a Head of Sales North America. |
| Certainly | Copenhagen, DK | PE-owned since 2023, pivoted into the Salesforce ecosystem, ~27k visits per quarter with a 24-second average visit. |
| DRUID AI | Bucharest, RO | $31M Series C (Sep 2025) under a new US CEO at a reported ~$300M valuation — a deliberate US enterprise pivot, and the product is broad agentic automation. |
| VIER | Hannover, DE | Genuinely DACH and voice-centric, but a six-company enterprise rollup, 200+ staff, est. €25–45M — probably through the ceiling, and no SMB motion at all. |
| BOTfriends · melibo · DeepOpinion | DE / AT | BOTfriends ~$1.7M and services-weighted (below floor). melibo is chat-only with quote-gated pricing (worth a later look). DeepOpinion does document automation, not conversations. |
| Company | Acquirer | When | What it tells us |
|---|---|---|---|
| aaron.ai | Doctolib | 22 May 2024 | AI phone for clinics, 3,500+ German providers at exit — bought by the vertical incumbent, not a CX vendor |
| Cognigy | NiCE · ~$955M | 28 Jul 2025 | ~$37M ARR at a ~25× multiple, 5.7× to investors on $165M raised |
| Vocalls | CallMiner | Jun 2025 | CEE voice AI exits to a US analytics vendor. Price not public |
| Ultimate.ai | Zendesk | 13 Mar 2024 | Helsinki. Now sold as "Zendesk AI agents" |
| Solvemate | Dixa · in a $43M double deal | Mar 2022 | Berlin. Brand absorbed; it is the ancestor of Dixa's Mim |
| Userlike | Lime Technologies · €19.8M | 30 Apr 2021 | Cologne, ~40 staff, ~2,000 customers. Now "Lime Connect" |
| e-bot7 | LivePerson · reported >$50M | Jul 2021 | Munich. The acquirer's subsequent decline makes it a cautionary tale |
| fluently | fonio.ai | 8 Sep 2025 | fonio is doing its own consolidation — 450 customers bought out of the Linz long tail |
Both completed exits in fonio's exact product category went to vertical incumbents, not CX platforms. aaron.ai → Doctolib (clinics). And fonio itself bought fluently. If that pattern holds, the acquirers to watch for a hospitality player like chatlyn are hotel-tech platforms — Mews, Apaleo, SIHOT — not contact-centre vendors. That changes who the strategic buyer is for anything we build in this space.
German-language search surfaced a long tail of AI phone-answering vendors, almost all sub-$1M and so out of scope as research targets — but they all publish prices, which makes them the cleanest available benchmark for where fonio's floor actually sits.
| Vendor | Base | Published price |
|---|---|---|
| IONOS AI phone | DE | €39/mo (30 calls) → €99/mo unlimited |
| goai | Linz, AT | €39–499/mo + project fees — almost exactly fonio's ladder |
| meiti | DE | €49/mo (Solo) |
| Placetel AI (Telekom-owned) | DE | €69/mo + €9/mo per AI number |
| Telfo | DE | €79/mo (750 min) · €159/mo (1,500 min) |
| HalloPetra | DE | €99/mo (250 min) |
| assistent24 | AT | €149 / €199 / €329/mo net |
| SalesFrank | DE | €0.36/min prepaid (outbound) |
| fonio | AT | €79–499/mo — mid-to-top of this range, not the cheapest |
Two German comparison sites list fonio differently from our brief — one at €99/mo for 1,000 minutes, another as "pricing on request." Neither matches the €79–499 tiers we are using as the anchor for every price comparison in this research. Re-verify fonio's published pricing directly before the deck circulates. Also note that one of those roundups still lists fluently as a live independent option eleven months after fonio acquired it — third-party comparison content in this category goes stale fast.