Loading…
The 3-click model is not a marketing concept — it is a design constraint that forces clarity about what AI should do, what the human should review, and what constitutes an exception. Here is what it means in practice.
The phrase "3-click procurement" appears on a lot of vendor slides. It is usually presented as an outcome — "reduce your process to 3 clicks" — without specifying which 3 clicks, what happens between them, or what the process looks like when something goes wrong. Used this way, it is marketing language without operational content.
Used correctly, the 3-click model is a design constraint, not a metric. It forces a specific set of questions about AI, human judgement, and exception handling that most procurement process designs never ask.
In a well-designed 3-click procurement process, the three clicks are:
Between those three clicks, a significant amount of work happens automatically: document ingestion, data extraction, validation against master data, PO matching, compliance checking, audit logging. The human does not see this work because it does not require human judgement. That is the point.
The value of the 3-click model is that it forces three uncomfortable questions during process design:
To design a 3-click process, you must be explicit about which tasks the AI can perform without human oversight in the loop. This is not a philosophical question — it is an empirical one. For invoice processing, AI can reliably extract invoice number, date, supplier, line items, and amounts from structured PDF invoices. It can match an invoice to a PO when line items align within tolerance. It can validate VAT numbers against HMRC or similar registries.
What AI cannot reliably do without human review is make contextual judgements: whether a rate change is commercially acceptable, whether a new supplier is strategically appropriate, whether a contract clause represents an acceptable risk position. Those judgements belong in the "review AI output" click, surfaced clearly so the human can make them efficiently.
This is where most AI procurement implementations fail. They surface everything to the human — every extracted field, every matching result, every flag — and call it "AI assistance." What actually happens is that the human spends as much time reviewing the AI output as they would have spent doing the task manually.
A genuine 3-click design surfaces only what the human needs to make the approval decision. For a standard invoice that matches a PO within tolerance, this might be: supplier name, invoice number, total amount, PO reference, and a green "matched" status. The human verifies these five data points and clicks approve. Anything else — the extracted line items, the matching algorithm confidence score, the individual field validations — is available on request but not presented by default.
The 3-click model only works for standard cases. Every real process has exceptions: invoices that do not match a PO, suppliers whose compliance certificates have expired, purchase requests that exceed budget. The design question is: what is the exception path, and does it break the 3-click model or route to a different process?
The correct answer is that exceptions route to a separate workflow with appropriate escalation — they do not add clicks to the standard path. The standard path stays at 3 clicks. The exception path has its own design with its own click count, which may be higher because more judgement is required.
For a procurement team, the 3-click model changes the nature of the role. Instead of spending 70% of time on transaction processing — data entry, matching, chasing approvals — the team spends 70% of time on the review function: exercising judgement on AI outputs, managing exceptions, and overseeing the AI's performance on the standard cases.
This is a real productivity shift, but it requires the team to trust the AI on standard cases. That trust is built through transparency — the AI must show its reasoning clearly enough that the human can verify it quickly, not just accept or reject a black-box recommendation.
A 3-click process that the team does not trust defaults to a 30-click process, because every click generates a manual verification. Trust is built by design — through clear AI reasoning display, accurate matching, and visible exception routing — not by asking the team to trust the system.
The most tractable place to start is invoice processing, because the standard case is highly defined (invoice matches PO within tolerance, supplier is active, total matches), the exception cases are clear (no PO reference, tolerance exceeded, blocked supplier), and the SAP posting step (MIRO) is well-understood.
For a new implementation, the sequence is:
Invoice processing is the entry point because it is well-defined. But the same 3-click design principle applies to sourcing event initiation, supplier onboarding, contract renewal, and catalog management. In each case, the design question is the same: what can AI do reliably, what does the human need to review, and what is an exception?
The procurement teams that benefit most from AI are not those with the most sophisticated AI — they are those who have been most rigorous about answering those three design questions for each of their core processes. The 3-click model is the discipline that forces that rigour.
We work with procurement and IT teams on Ariba implementation, BTP integration, and AI-assisted invoice processing. If this article raised a question specific to your landscape, get in touch.