A Zanus AI business case should explain which work will improve, what it will cost and how the business will know. Start with one bounded workflow, such as booking enquiries or requested callbacks. Count completed customer requests and staff effort before assigning a financial benefit.

Record the current workflow first

Measure eligible incoming requests, the proportion completed, time spent handling them and time spent fixing errors. Include requests that arrive outside office hours and those that never receive a response. Use a period that represents normal operating conditions.

Keep the underlying service mix visible. Comparing simple booking requests during a quiet week with complex enquiries during a busy week can produce an apparent improvement that has little to do with the tool.

Use a complete cost ledger

CostEvidence
Subscription and usageCurrent written quote and pilot invoice
Setup and integrationsActual implementation effort
Ongoing reviewTime spent checking and correcting work
Exception handlingStaff time completing unresolved requests
Operational disruptionRecorded missed promises or duplicate work

Do not assume every minute removed from a call becomes cash saved. Explain whether staff capacity is being redeployed, overtime avoided or an actual expense removed.

Estimate benefit with conservative assumptions

For bookings, separate requests captured, appointments accepted and visits attended. Where a completed job has a financial value, use an appropriate contribution measure from your own business rather than crediting the full sale price as benefit.

Create a cautious case in which contact rates are lower and exception work is higher than expected. If the business case only works with perfect calls and no staff review, it is not ready to justify a broad rollout.

Make the pilot a decision, not a forecast

Agree in advance which result supports expansion, which calls for more testing and which stops the pilot. Include accuracy, customer promises and staff capacity as well as cost. Assign an owner to reconcile the evidence against the original assumptions.

After the pilot, update the ledger with actual invoices, completed outcomes and handling time. Label assumptions that remain untested. A transparent case with a modest measured benefit is more useful than a large ROI claim that the team cannot trace to completed work.

Questions and answers

Should every captured lead be counted as financial benefit?

No. Track the path to an accepted appointment or completed job and use your own defensible value assumptions, while allowing for leads that would have converted anyway.

Mapping the Full Journey of a Single Enquiry Before Measuring Anything

Before a business can count completed jobs or evaluate staff effort meaningfully, it needs a shared understanding of what one enquiry actually looks like from first contact through to a verified outcome. Without that shared map, different team members may count the same work differently, producing figures that appear to conflict rather than inform. The practical starting point is to walk one real enquiry type through every stage it touches and write down, in plain language, who does what, where decisions are made, and what conditions cause the path to branch.

Consider a fictional illustration: a caller contacts a trade services business to book a boiler service. The call arrives, someone captures the caller's name and address, availability is checked, a slot is offered, the appointment is confirmed verbally, a confirmation message is sent, the job is allocated to a technician, the technician attends, the work is recorded as complete, and an invoice is raised. That sequence looks linear, but in practice it may involve a receptionist, a scheduling tool, a job management system, and a technician app that do not automatically share information. Each handover point is a potential place where a job stalls, gets duplicated, or falls out of the count entirely.

Drawing this map reveals something a spreadsheet cannot: the difference between a request captured and a job completed is not a single event but a chain of small handovers, each requiring a human or system action. Evaluation criteria built around completed jobs must reflect that chain. If a new tool improves only the first stage, the business case should say so, rather than attributing downstream completions to that improvement.

Defining What a Completed Job Actually Means for Your Business

The phrase "completed job" carries different meanings depending on the type of business, and those differences matter when setting up an outcome metric. In a cleaning business, completion might mean the visit happened and the client confirmed satisfaction. In a legal practice, completion might mean a file is closed and a matter resolved, which could take months. In a restaurant, it might mean a seated party finished dining. None of these is a universal standard, and an evaluation framework that imports a definition from another industry may measure the wrong thing.

A useful working definition combines three observable elements: the service was delivered to the customer as promised, the business recorded it in its operational system, and any follow-up obligation was either discharged or logged. All three must be present for a job to count as genuinely complete. A visit that happened but was never recorded leaves the business with no evidence for invoicing or repeat contact. A record that exists but mismatches what the customer received creates a dispute risk. A job marked complete when follow-up is outstanding may inflate the count and mask an unresolved obligation.

When building evaluation criteria, write this definition down and share it with the people who will be collecting the data. Misalignment here is a common reason why pilot figures differ from expectations; two people counting the same week of activity can produce different totals if one includes partially completed jobs and the other does not. Agreeing the definition before data collection begins is a straightforward step that protects the integrity of the comparison.

Separating Customer Identity from Individual Requests and Responsibility

A business that handles repeat customers needs to be careful about how it attributes completed work. A customer may submit multiple requests, have several jobs active simultaneously, or return after a gap. Counting at the customer level can obscure individual failures: if one of three requests from a single customer was never fulfilled, that gap disappears inside an aggregate figure that shows the customer as served.

Evaluation records should therefore operate at the request level, not only the customer level. Each inbound contact that generates a request should receive its own identifier, its own status, and its own resolution record. That structure allows a reviewer to ask: of all requests received in a given period, how many reached each stage of the process? How many stalled, and at which handover? How many required staff to intervene after an automated step?

Responsibility is a separate dimension from identity. A request may have been received by an automated channel, routed by a system, and fulfilled by a technician, with a manager handling a complaint arising from a scheduling error. The business case should be able to answer the question of who was responsible for each stage so that, if something goes wrong, the root cause can be traced rather than attributed vaguely to the process. This is especially relevant where staff capacity is being evaluated as a potential benefit; if responsibility is unclear, it is difficult to judge whether staff time genuinely changed or simply shifted from one task to another without being recorded.

A Worked Fictional Enquiry: Receipt Through Ownership to Completion

The following scenario is entirely fictional and is offered as a structured template for designing your own tracking, not as a description of actual system behaviour.

A property management company receives an after-hours maintenance request from a tenant reporting a faulty heating unit. The request arrives by phone and is logged with a timestamp, the caller's identity, the property address, the nature of the fault as described, and the channel through which it arrived. That record is the receipt point. An owner is assigned before the next business day opens; in this scenario, that is the facilities coordinator, who confirms the fault category and decides whether the job requires emergency attendance or can wait for a standard slot.

The coordinator books a contractor, records the scheduled visit time, and sends a confirmation to the tenant. That confirmation constitutes a promise: the business has committed to a specific outcome at a specific time. The evaluation framework should flag any job where a promise was made, because a later failure to attend becomes a measurable event, not an invisible one.

The contractor attends, completes the repair, and records the outcome in the job management system. The coordinator closes the request, logs the actual visit time against the promised time, and notes whether any follow-up is required. The tenant receives a closure message. At that point the job meets all three elements of the completion definition set earlier: delivered as promised, recorded in the operational system, and follow-up obligation discharged.

If the contractor had been unable to attend, the record would show a promise made and not kept, with a reason logged. If the fault turned out to be outside the scope of the original report, the record would show a variation. Both exceptions are useful data; they reveal where effort concentrates and where the original intake process may need refinement to capture more accurate fault descriptions at the point of first contact.

Designing Exception Handling as a Measurable Category, Not a Residual

A business case that accounts only for routine completions will underestimate the true operational picture. Exceptions, meaning requests that did not follow the expected path, absorb a disproportionate share of staff time and are frequently the point at which customer experience deteriorates. They deserve their own measurement category rather than being absorbed into a general figure for unresolved work.

Useful exception categories for counting purposes might include: requests where the caller could not be identified sufficiently to complete the booking; requests where the requested slot was unavailable and no alternative was accepted; requests that were escalated because the automated channel could not handle the enquiry type; and requests where the completed record mismatched what the customer had agreed. Each category produces different operational learning. A high volume of identity failures points to intake design. A high volume of slot refusals points to capacity planning. A high volume of escalations points to scope definition for the automated channel.

When reviewing a pilot period, exception categories should be reported alongside completion figures, not as a footnote. The ratio of exceptions to completions is a more honest signal of operational impact than the completion count alone. A tool that handles a high volume of straightforward requests efficiently while generating a large exception load for staff to resolve may produce a misleading completion headline. The staff cost associated with those exceptions belongs in the cost ledger, as this guide describes, and the resolution rate within each exception category should be tracked through the same pilot period.

Structuring Handover Arrangements Between Automated and Human Stages

Any process that combines automated handling with human follow-up requires explicit handover arrangements. Without them, requests sit in an ambiguous state where neither the system nor the staff member takes ownership, and the job does not progress. Handover arrangements are operational decisions that should be documented and tested before a broader rollout is considered.

A handover arrangement describes, at minimum: the trigger that moves a request from one stage to another; the person or role responsible for accepting the transfer; the expected response time; and what happens if the responsible person is unavailable. Each element is observable and can be tested during a pilot without waiting for a full production deployment.

A practical acceptance exercise might involve submitting a known test request during a period when the responsible person is unavailable, then observing whether the backup arrangement activates and within what timeframe the request is picked up. The observable evidence is the timestamp on the handover log and whether the request status updated correctly. That exercise does not require a large sample; it requires a deliberate test of the specific condition that matters.

Similarly, an enquiry that arrives outside office hours and triggers an automated acknowledgement should be tested to confirm that the acknowledgement accurately represents what the business can actually deliver. If the automated response promises contact by a specific time that the staffing arrangement cannot support, the business creates a promise it cannot keep, which becomes an exception requiring staff effort to manage. Reviewing the language of automated acknowledgements against the realistic capacity of the team handling follow-up is a straightforward evaluation step that often reveals adjustments needed before volume increases.

Questions That Change a Purchase Decision When Evaluating Outcome Metrics

Choosing how to measure completed jobs and staff work is not a neutral technical exercise; the measurement framework a business adopts shapes which tools appear to perform well and which appear to underperform. Several questions, when answered honestly, change whether a proposed tool represents good value for a specific operation.

First: can the tool's output be reconciled against your existing job completion records without manual transformation? If the answer requires a significant translation step each reporting period, the measurement overhead itself becomes a cost that belongs in the ledger. A reconciliation process that takes several hours of staff time each month across a pilot period adds up to a meaningful figure over a year.

Second: does the metric the tool reports match the completion definition your business agreed on earlier? Headline figures from a system may reflect its own internal logic, such as counting a confirmed booking rather than an attended visit, which may differ from the operational outcome the business actually values. Understanding that gap before committing to a reporting approach prevents the business from managing to a metric that does not reflect the work it cares about.

Third: how does the tool behave when a customer changes their request mid-process? A customer who books, then cancels, then rebooks creates a sequence of events that may be counted differently depending on how the system logs state changes. If re-bookings inflate the completion count, the business will overstate throughput. Asking to see how a specific scenario of that kind is recorded is a reasonable part of any evaluation demonstration.

Fourth: what does staff capacity actually look like after accounting for the exception handling and oversight work the tool requires? If the business case depends on a reduction in staff effort, the evaluation should include a record of how staff time was actually spent during the pilot, not an estimate made before it began. Time logged against review, correction, and exception resolution during the pilot period is the evidence base for that element of the case.

Building a Reviewable Sample Record for Each Stage of Your Pilot

A pilot generates value not only through its headline figures but through the underlying record of individual requests, each of which tells a specific story. Keeping a sample of individual records, covering both successful completions and exceptions, allows a reviewer to trace the path of a request end to end and identify where the process worked as designed and where it did not.

A reviewable sample record for a single request might include the channel and time of arrival, the intake data captured, the category assigned, the promise made if any, the handover points with timestamps, the staff actions taken, the outcome recorded, and whether any follow-up was triggered. That record does not need to cover every request in the pilot; it needs to cover enough variety to test the range of conditions the process will encounter in normal operation, including quiet and busy periods, straightforward and complex enquiries, and cases where the customer provided incomplete information at the outset.

After the pilot, the owner responsible for reconciling evidence against assumptions should be able to point to specific records that support or challenge each assumption in the original business case. A business case that rests on aggregate figures alone is harder to defend and harder to learn from. The sample records are the mechanism through which the team can identify whether an assumption held, failed at a specific type of request, or held only under particular conditions that may not persist at scale.

The discipline of maintaining and reviewing sample records also builds institutional knowledge about where the process is genuinely robust and where it depends on individual staff judgment that has not yet been made explicit. That knowledge is useful whether or not the tool moves beyond the pilot stage, because the map of the process itself is an asset the business retains.

Sources and next steps

This is an independent Sysevo planning guide, not a report of a hands-on Zanus AI test. Use the official Zanus AI Front Office overview to confirm current product scope and request a demonstration of the workflow described here. The scenarios and acceptance criteria above are proposed evaluation steps.

For the broader topic, read Zanus AI phone-agent cost comparison. If you are comparing providers, explore the Sysevo platform and request a Sysevo demo using the same scenarios. Sysevo publishes this guide and offers its own business software.

Related workflow guides

For current Sysevo subscriptions and usage options, compare plans and pricing.