The proposal is ready. It looks good. The packages are clearly explained, the customer’s requirements appear in the right places, and the next step is waiting in Microsoft Teams.
Now somebody has to check it.
Consider a hypothetical healthcare provider preparing an employee benefits offer for a corporate customer. A Digital Employee has assembled the proposal from approved service information and the customer brief. The sales manager opens it and finds that a regional coverage assumption needs confirmation. An exception to the standard package requires a commercial decision. The promised start date depends on onboarding capacity.
The document took minutes to produce. The business still has work to do before it can use it.
That remaining work belongs in the economics of the role. It includes the review that was always necessary and the additional effort introduced by the new method. Unless both are visible, the company cannot tell whether it has gained capacity or moved work into a less visible part of the organisation.
Define the result before counting it
A prepared offer, an approved offer and a signed agreement are different results. Each has a different owner and acceptance condition.
For this sales role, a useful result might be an offer ready for an authorised commercial decision: the correct customer entity, a traceable service configuration, current prices, and explicit unresolved exceptions. The Digital Employee does not need to make every decision to complete that responsibility. It does need to leave the decision-maker with a bounded question and the evidence to answer it.
If a missing fact is honestly identified and routed to the right person, that can be successful work. If the system quietly assumes the answer, a polished proposal may conceal an incomplete job.
This distinction also matters when suppliers charge for outcomes. HubSpot’s April 2026 announcement ties agent charges to specified events, including resolved conversations and recommended leads. Such definitions make the billing unit more concrete. A buyer must still decide how that event relates to acceptance within its own operation. HubSpot’s announcement makes the commercial definitions explicit.
An invoice can tell you what the supplier counted. Your operating standard must tell you what the business received.
Include the unsuccessful work
Suppose a company measures only the offers its sales manager approves. It records the software charge and a little review time. Offers abandoned halfway through, duplicate attempts and cases sent back for correction disappear from the calculation.
The resulting number rewards an incomplete account of the work.
A more useful measure is cost per accepted outcome: the total operating cost for a defined cohort of cases, divided by the outcomes from that cohort that meet the acceptance standard at a stated cut-off.
The numerator should include the cost of every attempt in that cohort. Add the allocated service cost, human preparation and supervision, correction, and attributable downstream repair. Identify one-off setup costs separately and show how they are allocated. If a case has no accepted result, its cost does not vanish.
The denominator needs equal care. Count one accepted result once. Record unresolved cases and their age alongside it. An early readout remains provisional until the reopening window closes; restate the result when attributable defects change acceptance or cost. Define how long a result can be reopened and how a later defect will be attributed. Otherwise, accepting everything on Friday and discovering the problems on Monday becomes an accounting advantage.
This measure is a proposed operating discipline, not a universal accounting standard. Its usefulness depends on definitions that people can audit and keep stable.
A worksheet for one responsibility
The following example is hypothetical. It defines a way to measure the proposal role without claiming a customer result.
| Measurement choice | Definition for the healthcare proposal role |
|---|---|
| Unit of work | One requested corporate offer for one customer entity and agreed scope. Revisions remain attached to the original case. |
| Accepted result | A decision-ready proposal with source references, current commercial terms and clearly routed exceptions. |
| Acceptance owner | The authorised sales manager, using the same checklist for assisted and existing work. |
| Cohort and cut-off | Requests opened in a four-week window, assessed after a further two weeks. Open cases are reported separately. |
| Reopening rule | Defects found within 30 days are attached to the original case. Later defects are logged separately and trigger a review of the window. |
| Costs included | All attempts, allocated service and integration costs, human preparation, review, correction and attributable downstream repair. |
| Comparison | Similar offer complexity, service regions and exception rates under the existing process. |
| Companion measures | Acceptance rate, elapsed time, unresolved case age, serious defects and human minutes per accepted result. |
The worksheet should also name the person responsible for collecting the data. Review time that nobody records will be estimated differently by every advocate in the room.
For a simple illustration, imagine a cohort costing €3,600 in total and producing 60 accepted outcomes at the agreed cut-off. The cost is €60 per accepted outcome. That arithmetic says nothing yet about whether the system is good value. The comparison could involve a cheaper existing process, a faster alternative, or work that previously remained undone. These figures are illustrative, not Outcome1.AI performance data.
Human attention has an opportunity cost
A sales manager may reasonably spend time deciding an unusual price. That is part of the role. Spending the same time reconstructing which price list the system used is a different burden.
Both consume minutes, but they require different interventions. The first calls for clear decision rights. The second may call for better source handling, a narrower assignment or a change to the product.
Measure the reasons for human involvement as well as the duration. A single “human review” total can conceal the difference between valuable judgement and avoidable investigation.
Do not convert every minute released into a salary saving. If the manager remains employed and uses the time to work on customer relationships, the benefit should be tested through that additional capacity. If no useful work fills the time, the financial case is weaker. The business consequence has to follow the evidence.
Give the old process a fair hearing
There is a serious argument for accepting substantial review during an early deployment. Familiarity takes time, and a team may deliberately inspect every case while it learns the failure modes. A narrow trial can therefore look expensive before the operating model settles.
That is a reason to distinguish learning costs from recurring costs. It is not a reason to stop counting.
Set a review point and a condition for changing the level of supervision. The condition should depend on observed performance within the role’s permitted scope. A promising demonstration cannot justify expanding authority into cases that were never evaluated.
The existing process deserves the same scrutiny. People also make mistakes, ask colleagues for help and leave work unfinished. Comparing a fully costed Digital Employee with an imaginary error-free employee will be misleading. So will comparing a carefully selected AI trial with the messiest month in the team’s history.
Use the same acceptance standard on both sides. Preserve the differences that matter, including whether one method can serve demand the other cannot reach.
Capacity becomes credible at acceptance
Outcome1.AI’s Digital Employee is defined around a business role. That makes acceptance a natural unit of accountability: a bounded responsibility, a result the business can use, and a visible route for what remains unresolved.
The standard must apply to Outcome1.AI as much as to any other supplier. The useful conversation starts when the buyer can see the cost of the whole responsibility, including its awkward cases.
Return to the sales manager in Teams. The offer is ready, the regional assumption has been resolved, and the pricing exception is presented with its commercial implications. The manager can decide without reconstructing the work.
That is where the capacity claim begins to earn its place in the budget.