Most contracts and invoices a mid-sized company creates are not unique. They follow a standard pattern: a frame that is 80 per cent identical, plus a variable part containing the specific parameters of the current case. Exactly this pattern is the ideal application for AI-supported pre-creation.
What “preparation” concretely means
AI does not produce a legally binding contract, and it does not send invoices. It prepares them — to a point where a human only needs a few minutes to finalise them.
A productive preparation delivers:
- Insertion of variable content: customer data, line items, prices, terms, conditions — extracted from CRM, ERP or order system and placed into the template frame.
- Consistency check: do the numbers add up (net, VAT, gross)? Do the terms fit the customer profile (payment terms, discount tier)?
- Highlighting deviations from the standard: if the proposed price is below the approved minimum margin or the term is unusual, it is clearly marked.
- Adaptable text blocks: suggested wording for specific situations — special cancellation, service extension, project-specific clauses — which the operator accepts or rejects.
The result is not a finished document but a draft that already contains 90 per cent of the work. The last 10 per cent — checking, adjusting, approving — stays human.
Where the lever is particularly large
Three operations are particularly fruitful in practice:
- Recurring invoicing (maintenance contracts, service subscriptions, monthly deliveries): here templates and variables are particularly well-defined. An experienced accountant creates a maintenance invoice in 30 seconds — that would otherwise be clicked together manually.
- Order confirmations with project-specific clauses: from offer, order and framework contract, the order confirmation is proposed with the correct wording, delivery dates and escalation paths.
- Reminders and payment notices with customer-appropriate tone: first reminder friendly, second clearer, third with concrete consequences. The tone is adjusted to the customer history — nobody addresses a 10-year repeat customer the way one addresses a delinquent first-time customer.
In all three cases, the template is standardised, the variable part clearly bounded and the risk manageable.
Which controls remain indispensable
The honest question is not whether AI can create contracts and invoices — but under which controls it works productively. Three are non-negotiable:
- Human approval before dispatch. Nothing goes out that a human has not seen and approved. Not even “automatic maintenance invoice on the 1st of the month” — the human looks at the list, ticks it off or corrects.
- Specification-based plausibility checking. Before approval, programmatic checks: is the price within the approved corridor? Are all mandatory fields filled? Is the tax logic correct (B2B-EU, Reverse Charge, etc.)?
- Audit trail. Every AI preparation, every human adjustment, every approval is recorded. In a dispute (internal or with a customer) it is traceable who decided what, when and how.
These three points are not built in out of mistrust against AI, but because commercial transactions have legal consequences — and those consequences are always carried by people, never by machines.
Common pitfalls
From the projects we have accompanied, three recurring mistakes:
- Automating across the board before templates are clean. If standard templates are out of date or contain contradictions, the AI inherits these problems. Template cleanup is a prerequisite, not a consequence.
- Special cases without a path. What happens with the unusual special order that does not fit the standard pattern? If no path is defined for it, a manual shadow process rebuilds next to the AI process.
- No feedback loop. When accountants make corrections to drafts, those corrections should improve the template system (or the AI template). Without that loop, they make the same corrections for months.
What remains in the end
Contracts and invoices are not mere paperwork — they are the formal traces of real economic transactions. They deserve care. AI does not take away the care — it takes away the mechanical preparation, so more care is available for the genuinely important points: are the prices right, do the terms fit, is the contracting party reliable, does the contract cover what both sides expect.
That is work that does not get less — it gets better, because the typing time stops.