The agent prepares offers in seconds. Then every fourth offer waits for the same specialist. By the afternoon, the proposals are neatly sorted in the system. Customers are still waiting.
A fast preparation step can feed a slower review process more work than it can handle. Before introducing an agent, calculate capacity at the end of the workflow, particularly when approvals or complex exceptions depend on a small number of people.
Count resolved cases
A fictional specialist merchant receives 1,000 requests for standard quotes each week. Each request currently takes an average of six minutes to process, adding up to 100 working hours. A new agent will collect data, prepare offers and complete straightforward cases within defined rules.
The following model assumptions make the capacity calculation tangible.
In this model, 75 percent of requests can be completed without further work. The remaining 25 percent need a specialist review averaging twelve minutes. These harder cases are not equivalent to the previous average request. Add eight hours a week for sampling, supervision and error analysis.
| Component | Calculation | Weekly effort |
|---|---|---|
| Previous workflow | 1,000 × 6 minutes | 100 hours |
| New specialist reviews | 250 × 12 minutes | 50 hours |
| Supervision and quality assurance | Model assumption | 8 hours |
| New total effort | 50 + 8 hours | 58 hours |
On paper, 42 hours become available. One crucial question remains: Are those 50 specialist hours actually available?
Calculate total time and suitable capacity separately
Suppose the authorised specialists can reserve only 30 hours a week between them for review. At twelve minutes per case, they can review 150 cases. However, 250 new cases arrive. Without further action, the backlog grows by 100 cases a week.
After three such weeks, the model produces 300 additional waiting cases, assuming no initial backlog and no cases resolved another way. Saving time overall does not automatically solve this problem. The saved hours may be in general customer service, while only two specialists can approve prices.
The first management question is therefore: Whose work disappears, and whose workload grows? Then decide whether training, different approval limits or a smaller initial scope will help. Permissions should change through a deliberate decision, rather than being silently relaxed by a full queue.
Determine the exception volume you can handle
A simple calculation gives an initial capacity estimate:
Cases that can be reviewed = available specialist minutes ÷ average review minutes per case.
This calculates capacity, not an individual customer's waiting time.
Here, 1,800 ÷ 12 = 150 cases a week. At 1,000 requests, that corresponds to a maximum calculated exception rate of 15 percent. This limit already uses all the available review time. It leaves no buffer for variation, illness or unusually long cases.
Do not treat 15 percent as a comfortable operating target. Also establish when requests arrive and how much review time varies. A calm weekly average does little for a customer whose urgent quote is caught in a Friday peak.
Three triggers for reducing automation
First, unresolved cases grow beyond the agreed threshold. Limit new automatic transactions in the affected group and transparently route additional requests into a supervised process. Existing customer promises still need an owner and a resolution.
Second, the oldest unresolved case exceeds the promised turnaround time. Its age then matters more than a healthy average. The team prioritises by customer impact and establishes which promises can still be met.
Third, a systematic error keeps creating exceptions. Fixing its cause comes first. Assigning more reviewers to the same incorrect compatibility rule becomes an expensive way to maintain data.
Set volume and time thresholds according to your service promise. General industry figures would imply a precision that each business needs to establish for itself.
Put four measures side by side
Show incoming cases, correctly completed cases, reviews still open and the age of the oldest open case. Add specialist time and error causes. Distinguish justified questions from avoidable rework.
Calculate cost across the complete workflow. Alongside specialist and supervision time, include model and system costs, implementation and observed correction costs. Released time can provide capacity for better advice; the calculation alone does not create an equivalent reduction in staff expenditure.
Chapter 14: The economics behind the agent provides the broader basis for assessing the additional economic contribution.
Start within a manageable scope and define how to resume
Begin with a group of cases whose exceptions the team can handle. In a supervised trial, measure the specialist minutes actually required and establish which tasks need to continue during an interruption.
After reducing automation, expand again only when the backlog has been handled, the cause understood and sufficient capacity demonstrated. Assign someone to make that decision. A counter reaching zero should not replace business approval.
Connect this introduction to Chapter 19: The first 90 days and the choice of partners. For a practical Monday start, measure the complete work needed to resolve twenty suitable cases, especially specialist review. The next useful step may be a better form or a clearer rule. That can be a worthwhile outcome of an agent project too.