#R09 — Market, Unit Economics, Build-vs-Buy, and Why POS Startups Die

Scope: competitor pricing, where revenue actually comes from, the unit economics that decide life or death, build/buy/partner verdicts per market, historical evidence, two 3-year P&Ls, and licensing/liability risk. Date: 2026-08-18. Labels: [verified] = checked this session with a URL, [known] = confident from training data, [estimate] = reasoned guess. FX used: 1 EUR = 1.08 USD, 24.5 CZK, 4.25 PLN, 5.0 RON [estimate]. Companion docs: r01-hardware-peripherals.md, r03-fiscal-regulatory.md, r04-offline-sync-domain.md, r06-crm-loyalty-privacy.md.

#0. Bottom line

1. This is a payments company wearing a POS costume. Toast FY2025: ARR USD 2.047B of which USD 986M is payments and USD 1.061B subscription, across 164,000 locations, on USD 195B of GPV at a 48 bps payments take rate [verified]. Strip payments out and Toast is a USD 6.5k/location/yr SaaS business that would never have justified its cost structure. If you cannot own the payment flow, do not build a POS.
2. SaaS-only restaurant POS in CEE is arithmetically not a business. At EUR 59/mo ARPU, 75% gross margin and 25% annual logo churn, LTV is ~EUR 2,100 against a realistic EUR 1,200 CAC — 1.8x, before the 24/7 support rota. With payments attached at IC++ economics the same site is EUR 6,500–8,000 LTV and 5x. Payments attach rate is the single metric the company lives on.
3. The EUR 7 average ticket breaks flat-rate card pricing in the EU. With EU interchange capped at 0.2%/0.3% [known], the variable spread is thin and the per-transaction fixed cost (EUR 0.06–0.11 processor + EUR 0.02–0.03 scheme) dominates. Price as % + fixed, never flat, and get on IC++ (Adyen for Platforms) rather than blended (Stripe Terminal) — that difference alone is ~3x the residual per site.
4. No open-source POS base saves you meaningful time. Odoo POS / uniCenta / Floreant / SambaPOS / Chromis all give you the order-entry screen, which is ~3–4 eng-months of a ~55 eng-month v1. They give you zero of the offline/sync engine (r04), zero of the printing stack (r01), zero fiscalization (r03), zero payment certification. Verdict: build the core, buy the fiscal signing layer, partner the payments rail.
5. Buying a distressed local POS vendor is the strongest non-obvious move for PL and RO. A 300–800-site regional vendor with certification and a support team trades at roughly EUR 1,500–4,000/site [estimate]. orderbird sold to Nexi at EUR 130–140M for ~14,000 clients ≈ EUR 9,300–10,000/client [verified] — a strategic multiple, not a distressed one. Buying certification + installed base + local support staff purchases the four things AI does not compress.
6. AI-assisted development compresses the code, not the clock. It moves a 3-person team's v1 from ~55 to ~38 eng-months [estimate]. It does not compress GUM/AT/AADE certification queues, EMV L3 / PCI, field sales meetings, physical installs, RMA logistics, or a Saturday-21:00 escalation. The bootstrapped path is viable only in CZ + UA + NL, where the compliance gate is near zero.
7. EUR 5–10M is not a Western-EU POS round; it is the first half of a EUR 25–40M program. The S2 model below burns ~EUR 11.75M over 3 years to reach ~EUR 6.8M exit ARR. Plan the Series A at month 20 or don't start.

#1. Competitive landscape with actual pricing

#1.1 US incumbents (the pricing gravity everyone benchmarks against)

Read: the US market has converged on USD 0–70/terminal software + 240–280 bps merchant rate. Software is a loss-leader; the ISO/PayFac spread is the product. SpotOn's USD 0/station-with-hardware plan is the purest expression: they give away ~USD 1,200 of hardware and recover it in ~34 bps over a 2-year term (on USD 1M/yr volume that is USD 3,400/yr — payback in 4–5 months).

#1.2 EU / CEE competitors (your actual battlefield)

#1.3 CRM / loyalty specialists (what you must beat on the "CRM" half)

The pricing arbitrage that justifies this whole project: a Prague café pays EUR 59/mo for POS and would pay EUR 25–40/mo for loyalty; a Berlin bistro pays EUR 89 for POS and EUR 60–120 for a Piggy/Como-class loyalty product. A bundled POS+CRM at EUR 119–149 undercuts the two-vendor stack by ~25% while carrying ~2.4x the ARPU of a POS-only competitor. That is the wedge. It is not a technology wedge; it is a packaging wedge, and it is copyable in 12 months by anyone who notices.

#2. Where the money actually comes from

#2.1 The four revenue lines, ranked

#2.2 Payment economics, honestly (EU vs US)

EU interchange is capped by IFR (EU) 2015/751 at 0.2% consumer debit / 0.3% consumer credit [known]. Commercial cards and non-EEA-issued cards are uncapped — 1.1–1.9% is normal. A tourist-heavy Prague or Lisbon restaurant can run 15–30% non-EEA card mix [estimate], which is exactly where a flat-rate offer bleeds.

Cost stack for a card-present EUR 20 transaction, CZ/PL domestic consumer debit [estimate unless noted]:

Same transaction on Stripe Terminal blended (cost ~1.4% + EUR 0.10 [known, verify]): cost EUR 0.38, sell at 1.79% + EUR 0.10 → spread EUR 0.058 (0.29%). IC++ delivers ~3x the residual of a blended rail. The price of IC++ is a real merchant-onboarding/KYC/underwriting stack and a settlement reconciliation engine — roughly 6–8 eng-months (see Build effort).

Rule to write on the wall: never price flat. At EUR 7 average ticket a 1.69% flat offer nets you negative EUR 0.02/txn against an IC++ cost base. SumUp gets away with it because their merchants are EUR 12+ ticket and mostly domestic-debit.

#2.3 Worked example A — 1-site café, Prague

Assumptions: EUR 350k/yr revenue, 75% card = EUR 262k card volume, EUR 7.7 avg ticket = 34,000 card txns/yr. 1 terminal + 1 KDS.

Revenue per site per month: EUR 302 year 1, EUR 302 steady. Payments = 70% of gross profit. Without payments this site is EUR 900/yr of gross profit and cannot pay for its own support.

#2.4 Worked example B — 5-site fast-casual chain, Poland

Assumptions: EUR 600k/site revenue, 85% card = EUR 510k/site, EUR 11 avg ticket = 46,400 txns/site/yr. 2 terminals + 1 KDS/site. Chain discount −20% on SaaS. Card spread compressed to 0.65% + EUR 0.01 (they negotiate).

Note the shape change: chains pay less per site in SaaS and more in absolute payments. They also cost 3–4x more to sell (multi-stakeholder, 3–6 month cycle) and 0.5x per site to support (they have an internal ops person). Chains of 3–15 sites are the best segment in this market — this is the one place I would push back on "SMB + small chains": weight the plan 40% single-site / 60% 2–20 sites by revenue, not by logo count.

#2.5 Worked example C — 20-site casual-dining group, Germany

Assumptions: EUR 1.1M/site, 80% card = EUR 880k/site (EUR 17.6M group), EUR 24 avg ticket = 36,700 txns/site. 3 terminals + 2 KDS + 4 handhelds/site = 60 fiscal registers group-wide.

Sanity check: Toast's blended ARR per location is USD 2.047B / 164,000 = USD 12,482/yr = USD 1,040/mo [verified, derived]. Our 20-site German model lands at EUR 1,033. The model is calibrated. But note Toast's average includes many small sites — meaning Toast's larger locations are well above USD 1,000/mo, and the small ones are well below. Toast's average location does USD 1.19M/yr of GPV (USD 195B / 164k). A EUR 350k café is not that customer.

Enterprise tax: the 20-site deal takes 6–14 months, demands an SLA with credits, a pen-test report, source-code escrow, a named CSM, and will squeeze payments toward 0.35%. Budget EUR 25–40k of pre-sales cost [estimate].

#3. Unit economics and the things that kill

#3.1 CAC by channel (CEE numbers; multiply 1.8–2.5x for DE/FR/NL)

Anchor: Toast added ~30,000 net locations in 2025 [verified] against sales & marketing spend in the USD 1.3–1.5B range [estimate] — ~USD 45k of S&M per net add, though that number includes upsell-to-base and brand. Treat it as the ceiling of what a mature field-sales machine costs, not as your target.

#3.2 Churn — the restaurant failure floor

Consequence: average customer life is 3.5–5 years. Any CAC over ~EUR 2,000 in CEE needs the payments line to clear. Toast/SpotOn can spend USD 5k CAC because a US location does USD 1.19M GPV; a Prague café does EUR 262k. US CAC benchmarks are 4x too generous for CEE and you will kill yourself copying them.

#3.3 Onboarding cost — the hidden CAC nobody models

A full-service menu is 150–400 sellable items with modifier groups, course/kitchen routing, per-item VAT class (DE 7%/19%, NL 9%/21%, BE 12% food/21% all drinks per r03), printer mapping, happy-hour price lists, and allergens.

Loaded cost of a CEE onboarding specialist: EUR 16–24/hr [estimate]. Manual: EUR 264–984/site. With AI ingestion: EUR 160–588/site. Savings ~EUR 250/site. At 1,000 sites/yr that is EUR 250k/yr — real, but it is 8% of the burn, not a strategy. Anyone claiming AI removes onboarding cost has never done a printer-routing session in a real kitchen.

Design implication: the ROI of AI menu ingestion is not the labour saved; it is time-to-first-transaction. Cutting install from "3 weeks out because our onboarding queue is full" to "next Tuesday" is worth more in close rate than EUR 250 in COGS.

#3.4 Support cost — the number that kills bootstrapped POS

Restaurants trade 07:00–01:00, 7 days, and break at the worst moment. Incident density peaks Fri/Sat 19:00–22:00 and on the four highest-revenue days of the year.

At 500 sites, 5.5 FTE handle only 64 tickets/FTE/month. You are paying for coverage, not throughput — 168 hours/week needs 4.2 FTE bare, plus holidays, sickness and Saturday-night stacking. That floor exists at 50 sites and at 500. This is the reason the first 300 sites are the most expensive customers you will ever have.

Per r01, ~60% of tickets are printing. AI tier-0 (in-app, with device telemetry: "your kitchen printer at 192.168.1.51 stopped responding 4 minutes ago, here is the power-cycle sequence") can plausibly deflect 30–45% [estimate] — printer resets, "how do I void", "where is my Z-report". It cannot deflect "we're down and there are 40 covers on the floor". Budget EUR 17–20/site/mo at 500 sites with AI, EUR 27.50 without.

#3.5 Hardware working capital

Also budget: RMA/DOA reserve 2–4% of hardware revenue [estimate], spare-pool stock (1 spare per 25 live terminals, plus 1 spare printer per 15 sites), and EU customs/VAT float if importing Sunmi directly rather than through an EU distributor.

Verdict: never subsidize hardware under S1. Sell at cost+15% or push to a leasing partner. Hardware subsidy is a payments-lock-in weapon that only works when you own the rail and have 24-month contracts and have capital. You will have none of the three in year 1.

#3.6 Blended gross margin and LTV/CAC

One number to remember: SaaS-only CEE is 1.7x with a 28-month payback and 25% churn. That combination is a slow-motion bankruptcy.

Note on how to report GM externally: Toast's headline gross margin is ~26–28% because they report gross fintech revenue including interchange pass-through. On a net-revenue basis they are ~70%+. If you present blended GM to an investor, state the basis explicitly or you will be marked down against the wrong comparable.

#4. Build vs buy vs partner

#4.1 Open-source POS bases — all rejected

The structural reason all five fail: the order-entry screen is ~3–4 of ~55 eng-months. The expensive parts are the offline-first sync engine (r04), the printing stack (r01, 8–10 eng-months), fiscalization (r03, 3–5 eng-months per certifying market), and payment terminal integration + certification. No OSS base contributes to any of those in a European context. You would spend 4 eng-months learning someone else's codebase to save 3.

#4.2 White-label / OEM POS cores

Options exist (Android POS ISVs shipping through Sunmi/PAX app stores; several MENA and Indian vendors will white-label). Economics: EUR 8–20/terminal/mo licence + a setup fee [estimate].

Verdict: no, with one exception. You cannot differentiate on CRM if the POS is someone else's — the whole thesis is that the guest data comes out of the order flow. The exception: white-labelling a KDS or a kiosk surface for year 1 to avoid building them is defensible, because those are leaf surfaces with clean interfaces.

#4.3 Payments — the decision tree

Verdict: start at referral, move to Adyen for Platforms IC++ by month 15. Do not chase a PI licence before EUR 300M GMV; the licence is a margin optimization, not an enabler, and it drags DORA and full AML obligations behind it (§7).

#4.4 Fiscal compliance — buy, always

Per r03: fiskaly / EFSTA / Fiscal Solutions abstract DE/AT/IT/FR/ES/PT and more behind one HTTP API, at ~EUR 10–15/register/mo [verified]. Building the DE TSE + DSFinV-K stack yourself is 3+ eng-months plus permanent maintenance. Buy it. The COGS hurts (15–25% of DE SaaS revenue) — solve that by pricing DE at EUR 89–129, not EUR 59.

#4.5 Acquiring a small local POS vendor — the strongest non-obvious move

What you are buying, in order of value:

1. Certification — a PL GUM-confirmed virtual register, or a RO AMEF integration, or a BE FPS-certified v2 (only 13 suppliers certified as of Mar 2026, per r03 [verified]). These are 6–12 month, EUR 12–50k gates you cannot compress with AI or money alone.
2. Installed base — 300–800 sites you can migrate over 18 months at 60–75% retention [estimate].
3. A local support team that already answers the phone in Polish at 21:00 on Saturday.
4. Reference customers, without which nobody in a conservative SMB market buys from a 3-person foreign startup.

Price anchors: orderbird → Nexi at EUR ~9,300–10,000/client (strategic, 14,000 clients) [verified]. A tired, sub-scale, founder-run 400-site vendor: EUR 1,500–4,000/site = EUR 600k–1.6M [estimate], often structured as 40% cash / 60% earn-out.

#4.6 Verdict per market

#5. Historical evidence

#5.1 How long and how much it actually took

Interpretation: nobody has built a restaurant POS + payments business to profitability in under ~10 years, and the two European exits of note (orderbird, Gastrofix) were both sales to payment or POS strategics at EUR 80–140M, i.e. a good outcome for a founder and a mediocre one for a venture fund. That is the realistic ceiling of this project: a EUR 60–150M strategic exit to Nexi / Teya / Worldline / SumUp / Lightspeed in years 6–9. Price the ambition accordingly.

#5.2 The failure pattern

Recurring casualties [known, individual details should be verified before quoting externally]: Revel Systems (raised ~USD 130M, sold to Welsh Carson 2019 well below peak), Breadcrumb (Groupon → Upserve for effectively nothing), Ambur, Bbot, Presto Automation (public, near-zero 2024–25), Sunday (French QR-pay, ~USD 124M at a USD 2B valuation in 2021 → fire-sale/restructuring [known, verify]), Vita Mojo (restructured), Yumpingo, and a long tail of national POS vendors absorbed by acquirers (Storyous → SaltPay/Teya; Dotykačka → Teya; orderbird → Nexi; Gastrofix → Lightspeed).

Six failure modes, in the order they kill:

1. SaaS-only pricing. Gross margin cannot fund the 24/7 support floor. Death by a thousand support tickets at EUR 59/mo. (§3.4, §3.6.)
2. Hardware subsidy without payments lock-in. EUR 400–580k of standing capital at 500 sites with no contractual mechanism to recover it. (§3.5.)
3. Linear CAC. Field sales that never converts to channel or product-led; every new site costs the same as the last one, forever.
4. Premature multi-country. Compliance cost and support-language cost multiply per country; revenue does not. r03 puts a 10-country footprint at 2.5–3.0 permanent FTE of compliance engineering alone. The most common European POS death.
5. Enterprise gravity. One 200-site chain offers EUR 400k ARR and eats 70% of the roadmap for 18 months building SSO, custom BI exports, and a franchise royalty engine. You emerge with a product no SMB wants and a customer who renegotiates.
6. The last 20%. Offline reliability, printer chaos, tip/service-charge law, cash management and blind drops, shift close, split/merge/transfer of checks, void-vs-refund audit trails, accounting exports. This is ~60% of the code and 0% of the demo. Every founder underestimates it by 2–3x; it is why r01 and r04 are the two longest companion documents.

#5.3 Does an AI-assisted 3-person team change the equation?

Net: AI takes the v1 build from ~55 to ~38 eng-months [estimate] — roughly 30%. It does not move the ~18-month go-to-market clock, the support floor, or the certification queue. A 3-person AI-assisted team can absolutely ship a credible product in CZ+UA in 12–14 months. It cannot ship a company in that time, because the company is 60% support, sales and logistics.

The honest reframing: AI does not make this project 3x cheaper. It makes the product 1.4x cheaper and leaves the business unchanged. Since the product is ~35% of the 3-year cost in the models below, AI saves ~13% of total burn.

#6. Two 3-year P&L sketches

Cost assumptions [estimate]: CEE engineer fully loaded EUR 4,200/mo; CEE support EUR 2,500/mo; CEE AE EUR 4,500/mo + commission; founder EUR 3,000/mo. Western-EU engineer EUR 9,500/mo; DE support EUR 4,200/mo; DE AE EUR 11,000/mo OTE. Opex = hosting, fiskaly, legal, accounting, insurance, travel, tooling, hardware float.

#6.1 S1 — bootstrapped, CEE beachhead (CZ → UA → PL partner)

Breakeven lands in Q2–Q3 of year 4 at ~EUR 1.3M ARR / ~830 sites. Peak capital need ≈ EUR 850–950k, i.e. a EUR 900k–1.1M pre-seed or an unusually well-capitalized founder. "Bootstrapped on EUR 200k" does not reach gate 2.

Go / no-go gates:

#6.2 S2 — funded, Western EU (DE/AT/NL core + PL), EUR 7M seed/A

The EUR 7M runs out in month 25. You must raise a EUR 15–25M Series A during Y2 on ~EUR 1.3–1.8M ARR — a 10–15x forward multiple, achievable in a good market and impossible in a bad one. That is the central risk of S2, and it is a financing risk, not a technology risk.

Go / no-go gates (S2):

#6.3 S1 vs S2 — the honest comparison

Recommendation: S1, CZ + NL, with UA as the engineering and support base. Not CZ + UA + PL — PL's GUM gate and RO's AMEF gate both need a partner or an acquisition, and neither fits a 4-person year-2 team. NL is free to enter, English-tolerant, has the highest restaurant tech spend per site in the plan, and puts Adyen (IC++, the 3x residual) on your doorstep.

#7. Regulatory and business risk

#7.1 Payments licensing

PSD3/PSR: the package was politically agreed in 2025 [known, verify] with application expected ~2027–2028. It tightens the limited-network and commercial-agent exclusions and moves e-money supervision under the PSD framework. Assume the exemptions you rely on today will be narrower in 2028 and do not build a business model whose only defence is an exclusion.

#7.2 Gift cards and stored value — a product constraint, not a licence problem

• A single-purpose voucher redeemable only at the issuing restaurant sits inside the limited-network exclusion (PSD2 art. 3(k) / EMD2 art. 1(2)) [known].
• The moment we hold the float across many restaurants, or a balance is spendable at more than one merchant, we are issuing e-money → EMI licence, EUR 350k capital.
• The workaround, and it must be drafted properly: the merchant is the issuer of record; funds sit in the merchant's own bank account (or their acquirer's settlement account); we are a record-keeper and reconciliation service. Constraints this imposes on the product: no cross-merchant gift cards, no multi-brand wallet, no cash-out, no interest on float, and for a 20-site group the balance liability sits on their balance sheet (which their CFO will want, so this is sellable).
• Loyalty points are safe — non-monetary, non-redeemable-for-cash, expressly outside e-money. Keep points and money strictly separate in the data model and never let a point have a EUR-denominated redeemable balance without legal review.
• National consumer law bites separately: gift-card expiry rules vary (DE: 3-year default limitation, courts have struck down 1-year expiries [known]; several markets require unredeemed-balance disclosure). Build configurable expiry with per-country defaults, not a hardcoded 12 months.

#7.3 DORA, PCI, and the compliance drag you inherit with a licence

• PCI DSS 4.0.1 applies regardless. Per r03, staying out of scope via P2PE-validated or semi-integrated terminals turns a EUR 30–80k/yr QSA engagement into an SAQ-A questionnaire [estimate]. Never let card data touch your application. This is the single highest-leverage architectural decision in the payments area.
• DORA applies to financial entities and their critical ICT providers. As a pure ISV you are out; the day you take a PI licence, DORA lands — ICT risk framework, incident classification and reporting on a 4h/24h/1-month clock, register of information, threat-led penetration testing. ~0.5–1.0 permanent FTE [estimate]. Another reason to defer the licence.
• NIS2 may catch you indirectly as a supplier to in-scope entities; expect security questionnaires from any chain over ~50 sites.

#7.4 Liability — "our bug closed a restaurant on New Year's Eve"

Write the SLA before the first contract: e.g. 99.9% monthly cloud availability excluding merchant network faults, 10% monthly-fee credit per 4h of qualifying outage, capped at 100% of the monthly fee. Explicitly define "outage" as cloud unavailability that prevents order capture — which, if offline mode works, almost never happens. Then make offline mode work.

#Build effort

Product engineering effort for POS/CRM/hardware/fiscal lives in r01/r03/r04/r06. This table covers only the commercial machinery that this document implies — the systems without which the economics above cannot be measured or collected.

Against the product effort implied by r01/r03/r04/r06, a sellable v1 for CZ + NL with referral payments is roughly 38–45 eng-months of total work; adding DE + own PayFac economics takes it to 58–68 eng-months. A 4-engineer AI-assisted team delivers ~48 eng-months/year of nominal capacity [estimate], so v1 is a 10–14 month build and the DE/PayFac version is an 18–22 month build — which is exactly why the S1 model shows no meaningful revenue until year 2.

#Open questions / what would change this answer

1. What is the real Adyen for Platforms rate card for a <EUR 50M GMV European ISV? Everything in §2.2 hinges on getting IC++ at roughly EUR 0.11 + a thin markup. If the minimum commitment or the markup for a small platform is 2–3x that, the residual halves and the entire LTV/CAC table degrades toward the "referral" row (5.0x, 10-month payback — survivable, but a different company). Get a written quote from Adyen, Nuvei and Teya before writing a line of payments code. This is the highest-value unknown in the document.
2. Can we get a payments rev-share of ≥25 bps without giving up the merchant relationship? Acquirer referral deals typically bundle a "we own the merchant" clause. If every CEE acquirer insists on that, the referral stage is poisoned and we must jump straight to PayFac-as-a-Service (6.5 eng-months earlier than modelled), which pulls ~EUR 200k of burn forward in S1.
3. What is the actual acquisition price and quality of a 300–800-site CZ/PL/SK vendor? §4.5 rests on a EUR 1,500–4,000/site [estimate]. Three real conversations would replace the whole build-vs-buy section. If it is closer to orderbird's EUR 9,300/site, buying is off the table for S1 and PL is a year-4 market.
4. Is the CEE ARPU ceiling really EUR 60–80, or does bundled CRM break it? Poster sells at USD 29–69 [verified] and Storyous around EUR 24–53 [estimate]. The entire thesis is that POS+CRM sells at EUR 119–149 in CZ/PL. Test this with 20 pricing conversations before building. If CZ operators will not pay above EUR 79 total, S1's year-3 ARPU of EUR 132 is fantasy and the model should be rerun at EUR 95 (which pushes breakeven from Q2-Y4 to Q4-Y5 and raises peak capital need to ~EUR 1.3M).
5. What is the true support ticket rate for a modern cloud POS at 500 sites? §3.4 uses 0.7 tickets/site/month [estimate] derived from r01's claim that printing is 60% of tickets. If the real rate is 1.5/site/month, support headcount roughly doubles and S1 never reaches breakeven without raising price. Two weeks of instrumented data from the first 25 sites settles this — make it a G1 exit criterion.
6. Does the Series A market in 2027–2028 fund European vertical SaaS at 10–15x forward ARR? S2's entire viability sits on gate G3 at month 20. Nory raising USD 37M in 2025 [verified] is encouraging, but Nory is software-only with no hardware, no fiscalization and no support rota — a structurally more fundable shape. If you believe the funding market is closed, S2 is not a plan, it is a bet on a market you do not control, and S1 is the only rational choice.