An AI tool can read a report, draft a client summary or reconcile a set of figures while someone watches. That demonstration proves the model can perform the selected task on the supplied material.
A weekly client process demands more. It has to start without someone opening the tool, move across the firm’s systems, stop safely when the facts are unclear and keep a record of the decisions. Signed documents provide a useful example because the individual task is simple and the surrounding operating work is easy to see.
One ordinary process, end to end
Engagement letters, declarations and returns go out through a signing platform. The signed copies return to an inbox at all hours and in whatever order clients complete them.
Handling one properly begins with detecting its arrival rather than waiting for the next inbox sweep. The process identifies the client, entity, document type and period, then files a confident match in the correct folder using the firm’s naming convention. An unfamiliar sender or ambiguous document goes to a person with enough context to resolve it. The practice-system status is updated where the interface allows; where it does not, a person completes the one remaining action with the rest staged. The run record captures what arrived, where it was filed, what was flagged and who intervened.
Reading the document and classifying it may take seconds in a demonstration. The operating requirements sit around that classification.
What the sequence demands
The sequence needs a durable place to run. Signed documents arrive outside business hours, and a quarter-end process cannot depend on one person’s laptop being awake. If a particular person must start it, that person still owns the recurring task.
The firm also needs ownership beyond the builder’s account. Runs should be consistent and visible to the people who depend on them. A process that works for a fortnight on one desk and stops when the builder gets busy has not entered the firm’s operating model.
Real work produces exceptions: a document from an unfamiliar address, a client with two entities and one surname, or a genuine judgement call. The workflow must recognise uncertainty and place the item on the right desk with its context. In the signed-document sequence, the exception route is part of the main design because unmatched items are expected.
People remain at the material checkpoints. A practice-system action that software cannot reach stays with a person, and an unmatched document waits for a decision. Human review belongs where consequence or judgement enters the process. Someone remains accountable for what reaches a client or regulator, while the system prepares the work and waits at the defined hold.
Finally, the sequence needs ongoing operation. Signing platforms change their notifications, folder structures are reorganised and new document types appear. Someone must monitor the runs, maintain the integrations and update the rules while preserving the approval points. Without that responsibility, the process will drift.
Strategy and preparation are still the minority
The industry numbers are consistent with this picture. Thomson Reuters’ 2025 research on tax and accounting firms found 14% had a defined AI strategy in place, and 25% had provided staff with generative AI training — training, note, not integration. Neither figure says AI has failed in firms; plenty of individual use is real and useful. What they show is that formal strategy and staff preparation remain minority practices — consistent with the operating gap the signed-document walk-through illustrates: tools in people’s hands before the processes, ownership and controls exist to run them as part of the firm.
In the signed-document example, classification is one part of a larger operating job. Durable execution, shared ownership, exception handling, human checkpoints and maintenance all have to work together. A technical and patient person may assemble several of those pieces during a quiet week. Sustaining the complete sequence for years is a different responsibility, particularly in a firm that does not employ engineers for that purpose.
What this means if you’re weighing it up
Before choosing a tool, map what the process needs in order to run as part of the firm. That exercise exposes ownership, exception, integration and approval requirements before money is spent. It also helps distinguish work with a useful operating case from work selected because it demonstrates well.
VonBolt Workforce provides that governed operating layer. A signed-document workflow executes on the platform, exceptions reach the right person with context, material checkpoints hold for review and the run record shows what happened. The firm’s people direct the work and retain sign-off. VonBolt operates the platform and keeps the workflow running.
Walk your firm’s version of the signed-document process from arrival to completed record. The missing operating requirements will become visible quickly, along with the parts already fit to run.
If you want to do that walk-through with us, get in touch.
Damien Zeller is the Principal of VonBolt Technology Group.
