The work that is left is judgement

Every other back-office tool now opens with the same pitch: upload your policies, your chart of accounts, your playbooks, and the model will do the rest. The pitch is honest about one thing. A model's answer is only as good as the context it has on the question, so whoever holds the context holds the account.

We ran into this from the other side. bonofici does not sell a tool. We take over a function, bookkeeping or claims or the service desk, and run it end to end. Doing that for a year taught us where the value actually sits, and it is not in the software.

Context is the product

What makes a back office run is not the ledger app or the ticketing system. It is the intelligence a tenured person carries: which supplier always invoices twice, what "reasonable" means for a €3,000 claim under this policy, when a refund request is really a chargeback threat. Companies know this. Every one of them wants to extract that intelligence from the people who have it and spread it around, so the next person does not have to relearn it.

Most tools try to do that by pulling knowledge into their own platform. The knowledge base becomes the Schelling point and keeping it in sync becomes everybody's job. Our discovery was that the useful unit is smaller and much more boring: a written standard. Scope, turnaround, thresholds, remedy. Four lines an operations lead can read in a minute and sign.

Once the standard exists, the software underneath is interchangeable. We have run the same bookkeeping standard through three accounting systems. The standard did not change. The judgement calls encoded in its thresholds did not change. Only the connectors did.

The tool learns the job, then the job disappears

Here is the uncomfortable part for anyone selling seats. A tool that starts by helping a person do a task gains, in the process, everything it needs to do the task without the person. The copilot becomes the autopilot. Invoice matching went this way. First response on tickets went this way. Renewal reminders went this way. Each was a feature someone paid a licence for, and each is now a routine case that runs without a human.

We think most workflow tools priced per seat are exposed to exactly this. If the work is automatable, the buyer will eventually pay for the work, not for the chair in front of it. That is why bonofici bills per completed job and not per seat or per hour. It is the only price that survives the job being automated.

What cannot be automated away

Two kinds of work resist this. Work that needs judgement, and work where regulation demands a named, accountable human. A disputed claim. A hire. A payment above a threshold. A settlement that will be audited. These are the exceptions, and the exceptions are where the money and the risk sit.

This is the part of the back office people will keep paying for, and it is the part we staff. Every bonofici engagement has a named operator who owns the exceptions and reviews the result against the standard. The automation does the routine ninety percent so that one person can give the other ten their full attention.

What we changed because of this

• We stopped selling access. Nothing to roll out, nothing to license. The client hands over the function and gets a ledger back.
• We put the standard first. No work starts until scope, turnaround, thresholds and remedy are written down and signed.
• We bill on the standard. A job bills when it meets it. A miss is redone and not billed.
• We keep the standard in plain text, editable by the client's own operations lead. The know-how should belong to the company, not to our tooling.