The price of a SaaS is not a number to guess: it is the rule by which the value you give customers turns into revenue. For a European founder there is a second step that many discover late: between the price written on the page and what you collect there is VAT, and its rule depends on who buys (a business or a consumer) and on where they are. This guide puts the two things together. First it compares the five pricing models on ten invented customers, because the choice depends on how much the value grows with what you measure. Then it explains, from the text of the European directive, what changes in the net price and in the return when you sell in other countries. Finally it shows how many trials you need to know whether a new price works.
The five pricing models, and where they break
A pricing model is defined by two choices: what the customer pays for (a subscription, the users, consumption, a tier) and what grows when the customer grows. The table lines them up. No number is a benchmark: they are descriptions of the mechanisms, which hold whatever the sector.
| Model | The customer pays | Revenue grows with | Where it breaks |
|---|---|---|---|
| Flat | A fixed amount for everyone | The number of customers | Customers who are worth a lot pay the same as those worth little |
| Tiered | The amount of the tier they fall into | Customers and moves between tiers | The thresholds between tiers push people to stay just below |
| Per user | An amount for each user | The customer's users | The customer avoids adding users |
| Usage-based | An amount for each unit used | The customer's usage | A month's revenue is unpredictable |
| Freemium | Nothing, until they move to a paid plan | The share of those who convert | Those who do not convert still have a cost of service |
The decisive column is the third: the right model is the one in which what you measure grows together with the value the customer gets. If a small customer and a large one get the same value, flat is correct; if the value grows with users, the per-user price follows it; if it grows with usage, consumption does. Freemium is more an acquisition strategy than a pricing model, because it requires a paid plan downstream, and its cost (support and infrastructure for those who do not pay) has to be put on the budget before deciding to use it.
Ten invented customers: who captures more value
To make the comparison concrete, the program below uses ten invented customers of a B2B SaaS, each with users, reports per month and a monthly value that they attribute to the product. For each model it looks for the price that maximises revenue, assuming that a customer buys only if they pay less than the product is worth to them. The check at the end verifies that the optimal flat price coincides with the value of one of the customers, as it must. I ran it twice with Python 3, with the same output.
# Ten INVENTED customers of a B2B SaaS: (users, reports per month, what the product is worth to them, EUR/month).
CUSTOMERS = [
(2, 20, 60), (3, 35, 90), (4, 30, 120), (5, 80, 150), (8, 60, 260),
(12, 150, 420), (15, 90, 480), (25, 300, 900), (40, 250, 1300), (60, 700, 2600),
]
def flat_revenue(price):
return sum(price for _, _, v in CUSTOMERS if price <= v)
def per_user_revenue(price):
return sum(price * u for u, _, v in CUSTOMERS if price * u <= v)
def usage_revenue(price):
return sum(price * r for _, r, v in CUSTOMERS if price * r <= v)
def tiered_revenue(tiers):
"""tiers: list of (maximum users, price). Each customer gets the tier that covers their users."""
total = 0
for u, _, v in CUSTOMERS:
price = next((p for maximum, p in tiers if u <= maximum), None)
if price is not None and price <= v:
total += price
return total
def best(function, grid):
return max(((function(p), p) for p in grid), key=lambda t: (t[0], -t[1]))
grid_eur = range(1, 3001)
grid_cents = [p / 10 for p in range(1, 301)] # from 0.10 to 30.00 EUR
r_flat, p_flat = best(flat_revenue, grid_eur)
r_usr, p_usr = best(per_user_revenue, grid_eur)
r_use, p_use = best(usage_revenue, grid_cents)
tiers = [(5, 99), (20, 399), (999, 1500)]
r_tiers = tiered_revenue(tiers)
print(f"flat price {p_flat:>6} EUR/month -> revenue {r_flat:>6}")
print(f"per user price {p_usr:>6} EUR/user -> revenue {r_usr:>6}")
print(f"per use price {p_use:>6} EUR/report -> revenue {r_use:>6.0f}")
print(f"3 tiers {tiers} -> revenue {r_tiers:>6}")
print("total value the customers attribute to the product:", sum(v for _, _, v in CUSTOMERS))
# Independent check of the flat price: the optimal price coincides with one of the customers' values.
assert max(flat_revenue(v) for _, _, v in CUSTOMERS) == r_flat
# Output: flat price 900 EUR/month -> revenue 2700
# per user price 30 EUR/user -> revenue 5220
# per use price 2.8 EUR/report -> revenue 4480
# 3 tiers [(5, 99), (20, 399), (999, 1500)] -> revenue 2496
# total value the customers attribute to the product: 6380
The maximum revenue of the flat model is 2,700 euros a month, 42% of the total value (6,380) that the customers attribute to the product: at 900 euros only those who are worth at least 900 buy, and seven customers out of ten drop out. The per-user price at 30 euros brings 5,220 (82%), usage at 2.80 euros per report 4,480 (70%). The three tiers that I picked at random (99, 399, 1,500 euros) bring 2,496, 39%: worse than flat. This is not a ranking of the models. It says that in my data the value grows with users, and therefore the per-user price captures it, while tiers set without looking at the data can do worse than a single price. With your data the ranking may change, and the program can be redone in ten lines on your customers.
Anchoring and the decoy effect: what the studies say
Two ideas from the psychology of price circulate in every pricing guide, and it is worth knowing where they come from. Anchoring comes from Tversky and Kahneman: in their 1974 article on judgment under uncertainty they describe adjustment from an anchor as one of three heuristics that people use to estimate under uncertainty, heuristics that are highly economical and usually effective, but that lead to systematic and predictable errors. The decoy effect, the idea of adding to the price list an option that is worse than another on every aspect in order to shift the choice, goes back to a 1982 study by Huber, Payne and Puto, published in the Journal of Consumer Research with the title "Adding Asymmetrically Dominated Alternatives: Violations of Regularity and the Similarity Hypothesis". Of this second article I only checked the title and metadata on Crossref, not the abstract: I do not report its results.
Both are studies on people's decisions in experimental situations, not measurements on SaaS price lists. That is why, in this article, you will not find a percentage of customers who "move" towards a middle tier: neither source supplies one for your product. If you want to use a decoy in the price list, the right question is whether it works on your customers, and the answer comes from a test, as in the last section.
From price to takings: what VAT does
VAT is not a cost to you, but it changes the price customers see and, if you get the rule wrong, the amount you are left with. The basis is Directive 2006/112/EC, which I read in the consolidated version of 14 April 2025, the latest on EUR-Lex: the articles cited below have the same text in the consolidated versions of 1 January and 20 March 2025 too. Directive (EU) 2025/516 (the so-called ViDA) amends other articles with effect from future dates: from 1 January 2027 Article 59c (paragraph 1, point (b), and paragraph 3, without touching the EUR 10,000 threshold) and from 1 July 2028 among others Articles 369b and 369g. Articles 44, 58, 196, 369d, 369f, 369h and 369i are not touched, but the dates should be rechecked with the accountant. The rule changes depending on who buys.
If the customer is a taxable person, meaning a business or a professional acting as such, Article 44 places the service in the place where the customer has established their business. Article 196 provides that VAT is payable by the customer when the service is supplied by a taxable person not established in the territory of that State: this is the reverse-charge mechanism. For the supplier it means that the invoice to a customer in another country does not carry the VAT of that country. To apply it, however, you must know that the customer really is a taxable person: how to document it (for example with the VAT identification number) is a point to agree with the accountant, because in this session I did not read the implementing rule that governs the proof.
If the customer is a consumer, for services supplied electronically Article 58 places the service in the place where the customer is established or has their permanent address or usually resides: the VAT is that of the customer's country. There is an exception for small volumes, in Article 59c: if you are established in a single Member State and the net value of your B2C sales to other Member States does not exceed 10,000 euros in the current calendar year, and did not exceed it in the previous year, Article 58 does not apply and the VAT of your country remains; if the threshold is exceeded, Article 58 applies from that point. For B2C sales beyond the threshold Article 369b admits the special one-stop-shop scheme (OSS), in which you are identified for VAT purposes for those transactions only in the Member State of identification (Article 369d).
Under the OSS scheme you submit a return for each calendar quarter, by the end of the following month (Article 369f), with the detail for each Member State of consumption and by rate (Article 369g), in euro (Article 369h), and you pay at the expiry of the filing deadline (Article 369i). These rules concern those established in the Union: for those outside there is a different scheme, which I do not cover here.
The standard rates of the three countries used in the example were read at the source: Italy applies 22% (article 16 of Presidential Decree (DPR) 633/1972, in force until 31-12-2026, then article 34 of the VAT consolidated act, Legislative Decree 10/2026, which carries the same 22%), Germany 19% (§ 12, paragraph 1, UStG), Spain 21% (article 90 of Ley 37/1992). Reduced rates exist for listed goods and services; whether your service falls into one of them is established by the rule of the customer's country and has to be asked of the accountant, not assumed. The program computes what follows the price and the quarterly return.
from datetime import date
from decimal import Decimal as D, ROUND_HALF_UP
from fractions import Fraction as F
from itertools import accumulate
# Standard rates read at the source on 2026-11-13: Italy 22% (art. 16 DPR 633/1972),
# Germany 19% (§ 12 UStG), Spain 21% (art. 90 Ley 37/1992).
RATES = {"Italy": D("0.22"), "Germany": D("0.19"), "Spain": D("0.21")}
CENT = D("0.01")
THRESHOLD = D("10000") # art. 59c(1)(c) of Directive 2006/112/EC
def cent(x):
return x.quantize(CENT, rounding=ROUND_HALF_UP)
# 1) The same list price VAT INCLUDED, in three countries (sale to a consumer, B2C)
print("price including VAT 29.00 EUR:")
for country, a in RATES.items():
net = cent(D("29.00") / (1 + a))
print(f" {country:<9} net {net} VAT {D('29.00') - net}")
# 2) The same NET revenue of 24.00 EUR, and what the customer sees
print("net 24.00 EUR, price the customer sees:")
for country, a in RATES.items():
print(f" {country:<9} {cent(D('24.00') * (1 + a))}")
# 3) A supplier established in Italy, INVENTED B2C sales to consumers in other Member States
# (date, customer's country, net amount in EUR). Previous calendar year: threshold not exceeded.
sales = [
(date(2026, 2, 10), "Germany", D("2400.00")), (date(2026, 3, 5), "Spain", D("1900.00")),
(date(2026, 4, 20), "Germany", D("3100.00")), (date(2026, 5, 18), "Spain", D("2600.00")),
(date(2026, 7, 9), "Germany", D("4000.00")), (date(2026, 8, 25), "Spain", D("1500.00")),
]
cumulative = D("0")
exceeded_on = None
cumulative_at_threshold = None
due_oss = {}
due_italy = D("0")
for day, country, net in sales:
cumulative += net
if exceeded_on is None and cumulative > THRESHOLD:
exceeded_on, cumulative_at_threshold = day, cumulative
if exceeded_on is None: # below the threshold: Italian VAT
due_italy += cent(net * RATES["Italy"])
else: # threshold exceeded: VAT of the customer's country
due_oss[country] = due_oss.get(country, D("0")) + cent(net * RATES[country])
print("threshold exceeded on:", exceeded_on, "- cumulative at that date:", cumulative_at_threshold)
print("Italian VAT on the sales before the threshold:", due_italy)
for country, vat in due_oss.items():
print(f"VAT to declare to {country} (OSS scheme):", vat)
# Independent check: partial sums and fractions, without the shared state of the loop above.
sums = list(accumulate(v[2] for v in sales))
expected = sum(
F(v[2]) * F(RATES["Italy"] if total <= THRESHOLD else RATES[v[1]])
for v, total in zip(sales, sums)
)
assert abs(expected - F(due_italy + sum(due_oss.values()))) < F(1, 10)
print("check with fractions, total VAT:", float(expected))
# Output: price including VAT 29.00 EUR:
# Italy net 23.77 VAT 5.23
# Germany net 24.37 VAT 4.63
# Spain net 23.97 VAT 5.03
# net 24.00 EUR, price the customer sees:
# Italy 29.28
# Germany 28.56
# Spain 29.04
# threshold exceeded on: 2026-07-09 - cumulative at that date: 14000.00
# Italian VAT on the sales before the threshold: 2200.00
# VAT to declare to Germany (OSS scheme): 760.00
# VAT to declare to Spain (OSS scheme): 315.00
# check with fractions, total VAT: 3275.0
The same 29-euro list price including VAT leaves 23.77 euros in Italy, 24.37 in Germany and 23.97 in Spain: the uniform price for customers is not uniform takings for you. If instead you want to collect 24 euros net, the customer will see 29.28, 28.56 and 29.04. The second part simulates a supplier established in Italy with six invented B2C sales to German and Spanish customers: the 10,000-euro threshold (cumulative 14,000) is exceeded on 9 July, the sales before that date carry 2,200 euros of Italian VAT, those from 9 July 760 euros to declare to Germany and 315 to Spain, through the OSS. The text of the directive says that Article 58 applies "as of that time": how to treat the single sale that makes the threshold be exceeded is a point to clarify with the accountant, not to deduce from my example.
Testing a price without wasting traffic
Changing the price and watching whether sales rise is not a test: the difference between two periods may be noise. A well-run test compares two groups of visitors who see different prices in the same period, and the question to ask first is how many people are needed. The program computes the sample with the normal approximation for two proportions, two-tailed, with a 5% significance level and 80% power, and checks the result with a fixed-seed simulation.
import math
import random
from statistics import NormalDist
N = NormalDist()
ALPHA, POWER = 0.05, 0.80
def n_per_group(p1, p2):
"""Trials needed for each group, normal approximation for two proportions (two-tailed test)."""
za, zb = N.inv_cdf(1 - ALPHA / 2), N.inv_cdf(POWER)
pm = (p1 + p2) / 2
num = (za * math.sqrt(2 * pm * (1 - pm)) + zb * math.sqrt(p1 * (1 - p1) + p2 * (1 - p2))) ** 2
return math.ceil(num / (p1 - p2) ** 2)
def simulated_power(p1, p2, n, repetitions, seed):
"""How many times out of `repetitions` the two-tailed z test rejects the null hypothesis."""
rng = random.Random(seed)
z_crit = N.inv_cdf(1 - ALPHA / 2)
rejections = 0
for _ in range(repetitions):
x1, x2 = rng.binomialvariate(n, p1), rng.binomialvariate(n, p2)
pooled = (x1 + x2) / (2 * n)
se = math.sqrt(2 * pooled * (1 - pooled) / n)
if se > 0 and abs(x1 / n - x2 / n) / se > z_crit:
rejections += 1
return rejections / repetitions
for p1, p2 in ((0.05, 0.04), (0.05, 0.03), (0.05, 0.045)):
n = n_per_group(p1, p2)
print(f"conversion {p1:.1%} against {p2:.1%}: {n} trials per group, {2 * n} in total")
n = n_per_group(0.05, 0.04)
print("simulated power with that n (2000 repetitions, seed 7):", simulated_power(0.05, 0.04, n, 2000, 7))
print("same test with a tenth of the sample:", simulated_power(0.05, 0.04, n // 10, 2000, 7))
# Output: conversion 5.0% against 4.0%: 6745 trials per group, 13490 in total
# conversion 5.0% against 3.0%: 1506 trials per group, 3012 in total
# conversion 5.0% against 4.5%: 28408 trials per group, 56816 in total
# simulated power with that n (2000 repetitions, seed 7): 0.7955
# same test with a tenth of the sample: 0.142
The result dampens enthusiasm: to distinguish a 5% conversion from a 4% one you need about 6,745 visitors per group, 13,490 in total; for a difference from 5% to 3% 1,506 per group are enough, and for 5% against 4.5% you need 28,408. The simulation confirms the calculation: with that sample the test rejects the null hypothesis in about 80% of the repetitions (0.7955, with a simulation error of about one percentage point), and with a tenth of the sample only in 14%. If you have a few hundred visitors a month, an A/B test on price cannot tell you anything about small variations: you can only compare with customer interviews or change the price and evaluate over a longer window, accepting the uncertainty.
Frequently asked questions
Do I have to charge VAT to a business customer in another country?
It depends on the customer's status and on your location. For a service supplied to a taxable person established in another Member State, the directive provides that VAT is payable by the customer (Articles 44 and 196), so your invoice does not carry the VAT of that country. How to prove that the customer is a taxable person has to be clarified with the accountant.
What changes if I sell to consumers in other countries?
For electronic services the VAT of the customer's country generally applies (Article 58), with a threshold of 10,000 euros of annual sales to other Member States below which the VAT of your country remains (Article 59c). To avoid registering in every State you can use the one-stop shop OSS, with a quarterly return.
Is a different price for each country worth it?
I do not have a source that quantifies what a local price list is worth, so I give no percentage. The example only shows that an identical price including VAT leaves different net takings between countries; if this matters for your margin, calculate the effect on your price.
How many customers do I need for an A/B test on price?
It depends on the difference you want to distinguish: in the calculation of this guide, 6,745 visitors per group to see a 4% conversion against 5%, 1,506 to see 3%. With smaller samples the test does not have the power to answer.
What you can do now
You can choose the price metric by looking at how value grows for your customers, and redo the comparison of the five models on your own data. You can separate price from takings: the same list price including VAT is worth different net amounts, and the directive's rules on place, threshold and one-stop shop decide how you sell in your customers' countries. You know how many trials you need before believing a test on price. For what comes next, read how to negotiate the investment in the guide to the term sheet, how to prepare the pitch in the one on Pitch Coach Audio and how to hire in the guide to hiring. To estimate the cash you need with recoverable VAT, there is the runway calculator.
Sources
- Directive 2006/112/EC, consolidated text of 14 April 2025 (EUR-Lex) - read on 2026-11-13 Articles 44, 45, 58, 59c, 196, 369a-369i (Articles 59 quater and 369 bis-369 decies in the Italian-language version); text identical to that of the consolidated versions of 1 January and 20 March 2025. Latest version on EUR-Lex at the date of consultation. Directive (EU) 2025/516, Articles 2, 3 and 6, consulted on 2026-11-13: amendments from 1 January 2027 (59c) and from 1 July 2028 (369b, 369g).
- Decree of the President of the Republic (DPR) of 26 October 1972, no. 633, art. 16 (Normattiva) - 22% rate in the text in force until 31-12-2026; from 1 January 2027 the article is repealed by Legislative Decree 10/2026; read on 2026-11-13. VAT consolidated act (testo unico IVA), Legislative Decree 10/2026, art. 34 (22% rate), read on 2026-11-13.
- UStG § 12 (Gesetze im Internet) - standard rate of 19%, read on 2026-11-13. Ley 37/1992, art. 90 (BOE) - general rate of 21%, consolidated text with last update of 2026-10-07, read on 2026-11-13.
- Tversky and Kahneman, "Judgment under Uncertainty: Heuristics and Biases", Science 185 (1974) - abstract read on 2026-11-13: adjustment from an anchor as a heuristic. Huber, Payne and Puto, "Adding Asymmetrically Dominated Alternatives", Journal of Consumer Research 9 (1982) - title and metadata only (Crossref), abstract not read.
- Calculations: Python 3 programs reported above, run twice with identical output on 2026-11-13; the data are invented.

Commenti
Caricamento commenti...
Sign in to comment