What the Self-Employed Actually Spend Money On: More for Software Than for an Office

What does the cost structure of a one-person business look like? For large companies, annual reports and balance-sheet analysis answer that. For Germany's roughly four million self-employed, there are mostly estimates and tax guides. We counted instead: twelve months of actual business expenses, quarter by quarter.

The basis is the business expenses of self-employed people and small businesses in Germany from July 2025 to June 2026. For the question of how widespread an expense is, we use every business in that period, including those that joined partway through. For the question of what changes over time, an additional fixed comparison group holds the same businesses across all four quarters: only then is a change a change in behavior rather than in who is being counted. Private withdrawals, transfers between a business's own accounts, and VAT prepayments are excluded; only business expenses are analyzed.

What most of them pay for

Download CSV

Across twelve months, a good quarter of businesses have no software expense at all, but about 78 percent have at least one. That puts software at the top of the list, with no other category close. Subcontractors and freelancers follow at about 64 percent, goods and materials at about 54 percent.

Two figures further down the list say more than the top does. About 43 percent pay for legal or accounting advice: a substantial share of the self-employed buy tax help, but not the majority. And office rent sits in the bottom third at about 26 percent, behind marketing and just ahead of professional development. Only about one business in four pays for office space at all.

Software instead of an office

The comparison that shows the shift most clearly is software against office rent. About 78 percent of businesses pay for software, about 26 percent for an office, three times as many. And software leads on money too: about 11 percent of all business expenses go to software and subscriptions, about 2 percent to office rent. Measured against total spending, the software bill is more than four times the rent.

One detail sensibly qualifies the picture: those who do have an office pay considerably more for it than for software. The median office rent is roughly €1,060 per quarter, the median software bill roughly €330. So the office has not become cheap: it has become rare. The workshop of the self-employed is mostly no longer a floor plan but a stack of subscriptions.

Where the money actually goes

Download CSV

Counting euros rather than businesses reshuffles the ranking. The largest single block of spending is subcontractors and freelancers: about 17 percent of all business expenses. Among businesses that hire subcontractors at all, the median is about €1,370 per quarter, and about two in three of them do.

This is the real structural finding of the analysis: solo self-employment is rarely solo. It is a network in which the self-employed hire each other. The largest cost block of one-person businesses is the income of other one-person businesses.

Travel follows at about 12 percent and software at about 11 percent, then taxes at about 10 and insurance at about 6 percent. Office rent lands in the lower field at about 2 percent, level with legal and accounting fees.

What changes across four quarters

For this question we switch the basis: instead of all businesses, only the fixed comparison group counts, the same firms across all four quarters. Reach figures are naturally higher there, because those firms are continuously active; what matters here is not the level but the movement.

And the short answer is: remarkably little. Across four quarters the ranking of cost blocks stays practically unchanged. Software is at the top throughout, subcontractors remain the largest block of money, office rent stays at the bottom. Anyone expecting a dramatic shift will not find one in this data.

What is visible instead is a slight rise across almost every category, on the order of two to six percentage points. We read that cautiously on purpose: over the same period, the number of categories a business touches per quarter also rose, by about five percent. Part of those broad small gains is therefore more complete categorization rather than new behavior. So we name the movements without selling each as a trend: at the upper end insurance and marketing, at the lower end the reach of subcontractor spending, the only one to give way noticeably.

Two movements sit clearly above that measurement threshold. The first is the size of the software bill: within the comparison group, from roughly €437 to roughly €525 per quarter at the median, a gain of about a fifth. Because that counts amounts rather than categories, it is unaffected by the capture question. The second follows in the next section.

For travel, our data show a marked drop in the final quarter; because travel costs swing seasonally and travel receipts are often captured late, four quarters are not enough to separate season, late capture, and a genuine trend. We therefore do not report it as a finding, but keep watching it.

What is actually inside the categories?

Download CSV

A category is not yet an answer. "Software" can mean Microsoft Office or an AI subscription. So we looked inside the largest categories, grouping transaction texts into vendors, only ever in aggregate across many businesses.

In the software block Apple sits at the top, with a caveat worth knowing: the App Store also bills other vendors' subscriptions, so Apple collects other companies' revenue here. Below that it gets unambiguous. OpenAI and Anthropic reach about a quarter and a good fifth of software payers respectively, putting them well ahead of Microsoft and Adobe at about 10 and about 9 percent. In sheer reach, the two AI vendors are roughly two and a half times as present as the classic heavyweights of office software. In between sit Amazon and AWS, the hosting provider IONOS, Google Workspace, and the professional publisher Haufe.

For the other categories the answer is even more revealing, particularly where it comes out negative.

Download CSV

Where the self-employed buy products, they buy from a handful of global brands. About two thirds of the phone bill goes to Telekom, Vodafone, or O2. The ad budget goes to Meta and Google. Office supplies mostly means Amazon: about half of the businesses with office-supply spending buy there. Equipment comes from MediaMarkt, IKEA, Amazon, and Apple; transportation from Deutsche Bahn, Aral, Shell, and Uber; and where meals are expensed, McDonald's appears more often than any other brand.

Where they buy work and space, there are effectively no brands at all. For subcontractors, only about 2 percent of transactions can be attributed to a known vendor; for legal and accounting advice, about 4 percent. The largest cost block of self-employment therefore flows to people and small firms that appear in no ranking, because each name occurs only once or twice.

The finding is sharpest for office rent, and it answers the obvious question: no, the self-employed are not sitting in WeWork. Across the entire analysis, not a single coworking brand reaches our publication threshold: not WeWork, Regus, Design Offices, Mindspace, or Impact Hub. About 87 percent of rent transactions carry no recognizable vendor name at all. Those who have an office rent it from a private landlord or a local property manager, not from a chain.

The fastest-growing item: AI

Download CSV

Within the software block, one item is growing faster than anything else: AI tools. In the third quarter of 2025, just over a quarter of businesses paid for at least one AI subscription; a year later, nearly four in ten. That is a gain of about 11 percentage points in four quarters, in relative terms almost 40 percent, and the only movement anywhere in the cost structure that sits far above anything more complete categorization could explain.

The vendor landscape also shifted over the year: the share of businesses with spending at Anthropic rose from about 6 to about 26 percent, while OpenAI's share stayed within a narrow band between 22 and 26 percent across the whole period. One qualification belongs with that: subscriptions billed via app stores or payment processors appear on the bank account under the name of the billing service. That blur can hide spending, but it cannot invent any.

And it does not stop at more payers. Within the same comparison group the bill itself grows: the median AI spend per business with an AI subscription rises from roughly €61 to roughly €107 per quarter, a gain of about three quarters. More businesses pay for AI, and they pay more.

Download CSV

What it means

Three things follow from a year of business expenses.

First, the fixed costs of self-employment have gone digital. The base load is not floor space but subscriptions, and that base load is growing, in reach and in size. Anyone wanting to cut costs no longer renegotiates a lease; they clean up their subscription stack.

Second, the largest cost block is work that goes to other self-employed people. That makes this group far more economically interwoven than statistics about "solo self-employed" suggest: a downturn in orders does not hit one business, it hits a chain.

Third, the bookkeeping gets harder, not easier. Many small subscriptions from many vendors abroad mean more receipts, more reverse-charge cases, and more checking than a single monthly rent payment.

And running through all of it is a split: what is a product comes from a few large brands; what is work or space comes from nameless individuals. Both halves are growing, but only one of them can be negotiated, switched, or cancelled.

Part of being credible is naming the limits. Anyone who runs their books digitally and links their bank account via an interface works more digitally than the average of German self-employment; the software share measured here should therefore be read as an upper bound and the office share as more of a lower bound. The fixed comparison group also consists of firms that were continuously active for four quarters; very young and very short-lived businesses are not in it.

This analysis will be updated annually at this address; the changelog at the bottom of the page documents every revision. All analyses in this section: Data & Research.

Methodology

The basis is bank transactions recorded as expenses by self-employed people and small businesses in Germany. Time series follow the value date of the payment, not the moment it was captured: anyone who links a bank account also imports their history, so the capture date reflects product usage while the value date reflects when the payment actually happened. The most recent quarter may still change slightly as bank synchronization catches up.

We deliberately use two bases. Reach and spending shares refer to all businesses with business expenses between July 2025 and June 2026: the broadest possible basis, including those that joined partway through the year. Change over time refers to a fixed comparison group of businesses that recorded expenses in all four quarters; only there is a change a change in behavior rather than in who is counted. The two bases are not convertible into one another: within the fixed group every reach figure is markedly higher, because those firms are continuously active and observed across a full year. We therefore state which basis applies to each claim and use the more conservative, broader figure for reach. "Share of businesses" means the share with at least one expense in that category during the period; businesses observed for only part of the period lower that share, so the values are lower bounds rather than upper ones. Amounts and spending shares are evaluated only for payments in euros. Categories that are not business expenses (private withdrawals, transfers between a business's own accounts, VAT prepayments) are excluded. A catch-all category for other deductible expenses reaches a reach similar to software and about 9 percent of volume; it is not shown in the charts because it carries no substantive meaning, but it is included in the spending shares. Median figures refer to businesses with expenses in the respective category.

Spending shares for individual quarters swing considerably due to a few large single payments; we therefore publish them only for the full period and show change over time via reach and median amounts. Activity within the comparison group is stable across quarters: the number of transactions and the spending volume per business are nearly level at the start and end of the period, so the changes do not stem from unevenly dense data. The number of categories touched per quarter does rise by about five percent, however; gains in reach on the order of two to three percentage points therefore fall within what more complete categorization can explain, and we do not report them as a trend. Vendors are detected from the transaction text and consolidated into brands (for example the various Amazon and petrol-station spellings). Only vendors appearing across many mutually independent businesses are published; that structurally excludes private individuals, since a private contractor never appears across dozens of unconnected businesses. Payments to Norman itself and the bank's own account fees are excluded from all vendor rankings. Not every transaction carries a recognizable name: purchases billed through app stores, PayPal, Klarna, or Paddle appear under the billing service rather than the actual vendor, which makes per-vendor shares lower bounds. The share of attributable transactions is itself a reported figure and ranges by category from about 2 to about 66 percent. Internal and test accounts are excluded.

All values are shares or median amounts per business. We do not publish absolute user or company numbers. Individual businesses are never identifiable. How we work with data in general is documented in Methodology & data basis.

About this data

Norman is an accounting platform for self-employed people and small businesses in Germany; the anonymized expense data analyzed here originates there. Norman offers AI-powered features itself and therefore has a commercial interest in the software and AI market; payments to Norman are not part of the analysis. The category figures are available as a CSV download above.

Charts and figures on this page may be used freely with attribution to "Norman". Press inquiries and data requests: press@norman.finance.