What 5,443 invoice app reviews actually say

The store rating and the reviews disagree

Every app in the sample carries a summary rating of 4.8 or 4.9 stars. Score the written reviews we collected for those same apps and the picture changes: the per-app means land between 3.15 and 4.58. Every one is lower than the store rating. The gap averages 0.9 stars, runs from 0.30 at the narrow end to 1.65 at the wide end, and exceeds 1.2 stars for three of the eight.

One thing to be exact about, because it is the obvious objection: these are not two measurements of the same window. The store rating is each app’s current displayed summary; the written mean is computed over our sampled corpus, some of which goes back to 2011. It is still the comparison a person standing in the App Store actually makes — but it is a current headline number against an accumulated written record.

That is not evidence of anyone gaming anything. It is what you would expect from two signals produced by different behaviour, and Apple documents the machinery. Anyone can rate an app directly on its product page — but developers can also solicit ratings through an in-app prompt, which Apple permits up to three times in a 365-day period, and Apple’s own guidance is to ask when someone is “most likely to feel satisfaction”, after they complete a task rather than interrupting them. A written review takes a keyboard and five minutes, and overwhelmingly gets written by someone with a reason.

There is a second mechanism, and it is the one most people miss: a developer can reset the summary rating when a new version ships, but the written reviews stay on the product page. One is a number that can be wiped after a bad release. The other is a record that accumulates.

Almost nobody is lukewarm

The distribution matters more than the average, because it is not shaped like an average. Of the 5,443 reviews:

Two peaks, and a trough where the considered opinions should be. Fewer than one review in seven falls between the extremes. The polarisation is in the data; the explanation is ours, and it is this: that shape is what dependence looks like in a utility product. People who rely on an app to bill from the driveway review it harshly the day it stops letting them.

The complaint has not changed in seven years

Below are five of the strongest recurring negative themes, ranked by how many distinct reviewers raise them. Four are the same complaint wearing different numbers — the subscription went up, or the free tier stops at three, or at ten. The fifth is not about money at all: it is reviewers reporting that they lost invoices they had already sent.

Five of the strongest, rather than the five strongest, is deliberate. The clustering leaves near-duplicates — two separately labelled themes can describe the same price rise — so the table is a fair picture of what recurs, but not an exact deduplicated league table.

What makes any of them worth reporting is not the headcount. It is how far each one spreads across vendors, releases and years.

A complaint that appears across five competitors, thirty-odd releases and seven and a half years is not a vendor having a bad quarter. It is a packaging pattern that recurs across the apps in our sample, and it has been stable for longer than most of them have had their current names. Reviewers name what they are paying in their own words — two hundred dollars a year, twenty a month, two hundred and forty a year — and the figures recur often enough to be a theme in their own right.

The economic point underneath it is narrower than “caps are greedy”, and we cannot see any vendor’s books, so this is reasoning about the shape of the product rather than about anyone’s accounts. Whatever an invoicing service costs to run — storage, sync, email delivery, support — the marginal cost of one additional invoice from an existing customer is tiny next to the subscription price. That makes a monthly invoice cap primarily a packaging decision rather than the kind of hard resource constraint you see in compute- or storage-heavy products, where a quota tracks a real bill. Packaging is a legitimate way to sell software. The narrower point is simply that an invoice cap is a commercial product limit, not a technical ceiling on creating another invoice. What a monthly cap does to lumpy work goes into what that feels like on the Friday of a busy week.

What people praise is one word

The largest theme in the entire corpus is not a complaint. It is easy — 789 reviewers, every app in the sample, and a span of 138 months. Nothing else comes close; the second-strongest positive theme is “easy to do invoices”, at 239 reviewers.

Set that against the requests. The corpus contains 733 reviews making an explicit ask. Rank the content words in them and the leaders are better, customer, support, service, improve, price and subscription. More features are conspicuously absent from the top of that list: apart from payment, no capability noun makes it. The asks that recur are to be charged less and answered faster.

Our read of that — and it is a read, not something the word counts prove — is that speed and service matter more in this category than feature breadth, and that there is very little here to win on by building more. What is left to compete on is the number of seconds between “I finished the job” and “the customer has the bill”.

The best-liked app is nearly invisible in our search sample

The highest written-review score in the sample belongs to Zoho Invoice, at 4.58 — more than half a star clear of the next app, and the only one whose written record lands near its store rating. Zoho publishes a $0 plan for it.

That plan is not unlimited, and it is worth being precise because the claim gets repeated carelessly: Zoho’s own pricing page sets it at up to 500 invoices a year, two users and three projects, with “Powered by Zoho Invoice” branding on the document. But 500 a year is roughly forty a month — an order of magnitude past the three- and ten-a-month caps that generate the complaints above. For many solo service businesses that behaves like an uncapped plan rather than a limit they will run into, though anyone billing weekly for several clients should still do the arithmetic.

Now the part that should worry anyone who believes better products win. On 17 August 2026 we queried Apple’s public search API on the US storefront for eight plain invoicing terms — “invoice”, “invoice maker”, “invoice app”, “invoicing”, “invoice generator”, “invoice and estimate”, “billing and invoicing” and “small business invoice” — and recorded the top five results for each. Forty slots.

Zoho Invoice took none of them. It reached the top ten for only three of the eight terms, at positions seven, seven and eight. The best-liked app in our corpus, and the one whose written record most nearly matches its store rating, is close to invisible in a plain search for the thing it does. The most visible app in the sample, Invoice Simple, held seven of the forty; Invoice2go held five; Square Invoices and Invoice Fly held three each. No app came close to owning the results — visibility is fragmented rather than monopolised — but the visibility ranking looked very different from the written-review ranking.

Scope, precisely: that is one storefront, on one day, through the search API rather than the App Store app’s own ranking, and eight terms we fixed before looking at any results. It is a sampled proxy for discoverability, not universal distribution data. Within those limits it is the most commercially important thing in this dataset: being free and best-liked did not make Zoho Invoice prominent in these searches.

How this was done

The eight apps are the dedicated invoicing apps from the competitor set we assembled for our own product research, found through Apple’s public search API rather than hand-picked. A ninth app in that set, Wave, is a general small-business accounting suite rather than an invoicing app, so it sits outside this comparison. This is a working competitor set, not a census: there are invoicing apps on the App Store that are not in it, and no claim here should be read as covering the whole category. Reviews were collected from the US and GB storefronts.

5,443 reviews went in. An extraction pass produced 918 candidate themes. 112 cleared the strength threshold — roughly one in eight. Those 112 were then classified by how they behave over time, which is where most of them stopped being useful:

• 68 enduring. Recurring across enough reviewers, versions and months to be a property of the product rather than of a release. These are the only ones we acted on.
• 24 version-specific. Real recurrence, but concentrated in two or three builds — enough to notice, not enough to call structural.
• 12 resolved. Genuine complaints that stop appearing after a particular version. Someone fixed them.
• 8 unknown. Too few distinct voices, or dates too sparse, to establish whether they persist at all.

That is the whole 112. The classification is deliberately harsh: a theme has to survive across builds and years before it earns any weight.

Two mechanics did most of that filtering, and both are worth stealing:

• Weight breadth over volume. Reviewer count saturates fast and rewards whoever had the worst week. Duration, version spread and competitor spread carry most of the weight instead. Eighteen detailed reviews spread over three years beat forty near-identical ones from a single bad release.
• Separate transient from enduring using dates and versions, never sentiment. “They shipped a broken build” and “their pricing model fails people” are indistinguishable inside one release and completely distinct across thirty. Anger reads the same in both; the version spread does not.

What this cannot tell you

Two limits, both structural, and neither fixable by collecting more of the same data.

Reviews miss the people who leave silently. A review corpus can only contain people who wrote something. It cannot observe the person who installed an app, hit something they disliked in the first week, deleted it and never left a word — and nothing in this data indicates how many of those there are. So the complaints here are the complaints of people who cared enough to type, which is not the same as the complaints of everyone who tried these apps.

The sample is of unknown proportion. Apple’s public review feed returns roughly 500 reviews per app per territory and does not disclose the total. Every count on this page is therefore a count within a sample whose denominator we do not have. Ratios between themes in the same corpus are reasonably trustworthy. Absolute prevalence — “X per cent of all customers think this” — is not recoverable from this data, which is why no figure on this page is phrased that way.

What we changed because of it

Two of these findings are actionable by anyone building here, and we acted on both. The metering theme is why Invoice Tiger has no monthly counter. The “lost previous invoices” theme — 35 reviewers, five apps, seven and a half years — is the one that shaped the design most, because it is the complaint with a real consequence: a bill you cannot reproduce when a customer queries it eight months later.

The third finding, the one about being liked and being findable coming apart, is not actionable by building anything. It is the reason this analysis exists as a public page rather than an internal memo. If you are choosing an invoicing app, the two questions worth asking are what happens to your records if you stop paying, and whether the free tier resets on a calendar that has anything to do with how your work actually arrives. Whether you need an accounting platform at all is the question underneath both of them.