What would your business cost to operate if you designed it today?
It is the last week of the month. Supplier invoices are waiting for a match. Approvers are chasing missing information. The finance team is working late, and the controller is being asked whether another hire or an AI tool would solve the problem.
Both might help. But the first question is what is creating the work.
Perhaps invoices arrive in several inboxes. Perhaps purchase orders are incomplete. Perhaps a capable employee spends hours moving information between systems because nobody has changed the process around them.
“What would your business cost to operate if you designed it today?” is a useful way to examine those choices. It asks which work is necessary, how it should be done, and where people need to apply judgment. The answer should account for the service your customers and suppliers expect, the controls you need and the changes your team can realistically absorb.
Look at the work before choosing the tool
Take one process from beginning to end. For supplier invoices, follow the work from arrival through matching, approval and preparation for payment. Find out where someone waits, re-enters information, asks a question or corrects an error.
Then separate three kinds of work:
- Work you can remove: duplicate entry, unnecessary handoffs or checks that repeat what another system already does.
- Work you can make easier: collecting documents, matching information or routing a request to the right person.
- Work that needs accountable judgment: resolving a disputed invoice, approving an exception or deciding whether a payment should proceed.
This makes the options easier to compare. A clearer purchasing rule might prevent an invoice exception. A feature already in your accounting system might remove duplicate entry. AI might help interpret varied documents. Another person might be the right answer when the process is sound but the volume has outgrown the team.
A hypothetical accounts payable example
The following numbers are invented for illustration. They are not a Safebox client result, a supplier quote or an industry benchmark.
Assume a company handles 3,000 supplier invoices each month. Its staff spend an average of eight minutes per invoice on the tasks being reviewed, including routine work, corrections and exceptions. That is 400 hours a month. At an assumed loaded labor cost of $40 an hour, the work uses $16,000 of staff capacity.
The controller considers two alternatives. A process change could simplify intake and approval routing using existing software. An AI-assisted approach could extract and match information, with people reviewing exceptions and retaining payment authority.
| Monthly assumption | Current process | Simpler process change | AI-assisted process |
|---|---|---|---|
| Invoice work, including human review | 400 hours | 320 hours | 240 hours |
| Value of that staff time at $40/hour | $16,000 | $12,800 | $9,600 |
| Additional software and usage expense | $0 | $0 | $2,500 |
| Additional maintenance, monitoring and support expense | $0 | $500 | $800 |
| Total monthly resource cost in this comparison | $16,000 | $13,300 | $12,900 |
| Separate one-time implementation cash expense | $0 | $4,000 | $18,000 |
These assumptions hold invoice volume, service quality and control requirements constant. Existing costs that would not change are excluded from all three columns. The additional support expense is assumed to cover external maintenance and monitoring; internal invoice-review time is already included in the hours above. The implementation amounts are hypothetical cash expenses. Any internal time needed for design, training, testing and transition must be budgeted separately.
On those assumptions, AI has the lowest monthly resource cost, but only by $400 compared with the simpler change. Its implementation expense is $14,000 higher. Even before considering transition effort, that difference makes the simpler option worth testing. The real decision depends on whether either approach can deliver the assumed accuracy, service and capacity.
Freed time and cash savings are different results
The AI option releases 160 hours of monthly invoice work. Valuing those hours at $40 gives $6,400 of capacity. After $3,300 in additional monthly expense, the calculated resource-cost improvement is $3,100.
That does not mean the company’s bank balance improves by $3,100 a month.
If the employees stay and payroll does not change, the company spends an additional $3,300 while gaining time for other work. That may be valuable if it reduces a backlog, improves supplier service or supports growth. The business case should name that use of capacity and measure it.
Suppose, instead, the company can demonstrate that 100 of the released hours eliminate $4,000 of overtime or temporary labor expense each month. The monthly cash benefit would then be $700: $4,000 avoided expense less $3,300 of new expense.
At that steady rate, recovering the hypothetical $18,000 implementation payment would take about 26 months. A phased rollout, extra transition costs or lower savings would extend that period. If the avoided expense does not materialize, that cash-payback calculation does not apply.
The same discipline belongs in the simpler-process option. Compare actual changes in spending, useful capacity and service—not a labor-value estimate for one option against a cash budget for another.
Test the assumptions that could change the decision
A pilot should use a representative mix of work: easy invoices, incomplete documents, mismatches and unusual cases. A demonstration built around clean examples will not show how much review the team will need in daily operations.
Measure the total staff time, including corrections. Check whether invoices move faster, whether errors increase, and whether the new process creates extra work elsewhere. Record software usage and support time as well as the work completed.
Keep an owner for exceptions and a clear route back to the existing process when the new approach cannot handle a case. Before expanding, agree what would justify continuing, changing the approach or stopping.
Review the economics again after the process is in regular use. Invoice volumes change. Suppliers send different documents. Software prices and support needs can change. A business case is useful only while it reflects the work people are actually doing.
Redesign one part of the business, then learn
Designing the business as if you were starting today does not require replacing everything at once. It gives you a way to question accumulated effort while respecting the people, systems and commitments already in place.
Begin with a process where the problem is visible. Define the result you need. Compare a few credible ways of achieving it, including improving what you already have. Test the smallest change that can establish whether the economics hold.
The result may be AI, a better use of existing software, additional capacity or a combination. What matters is whether the business can deliver the required work more reliably, at a cost it understands and can sustain.
Safebox helps finance, operations and technology leaders assess a specific business process, compare the available approaches and define a practical next step.