Exercise · Worked exercise

Agency call tracking case study: a simulated example

Content updated:

Short answer

This simulated case shows how call volume can hide differences in cost, handling and sales across channels. Its data and calculations are illustrative: they illustrate how to investigate opportunities, calls with unknown sources and handling problems, rather than reporting a customer result.

Key verification: Reproduce the calculations and distinguish hypotheses from verified causes before acting.

Sources and limitations

An agency can measure clicks and forms and still not know what happens on the phone. This example shows the questions that come up when campaigns, call handling and sales are joined together while the data is split between the agency and its client. All data is illustrative: it does not come from a CallsIQ client or from a provider test.

The business in the example provides home repairs. The cohort covers one hundred initial calls from one hundred different contacts over one month. To keep things simple, there are no repeat calls. Closed sales are observed up to 45 days after the end of the period; the figures do not represent all of the business’s sales.

Who holds each piece of data

The particular difficulty for an agency is that it does not control the whole journey. It runs the campaigns and often the call tracking, but call handling, qualification and sales happen on the client’s phones and in the client’s CRM. The report is only reliable if each party supplies its part with the same matching key: caller number, date and time.

Where each piece of data in the case comes fromThree sources on the left feed a sheet reconciled by contact: ads, with spend and campaign, supplied by the agency; call tracking, with call and source, supplied by the agency or the client; and the client’s CRM, with qualification and sale, supplied by the client. Access to the CRM requires a data processing agreement under article 28 of the GDPR. The reconciled sheet leads to a decision with a review date. Ads: spend and campaign supplied by agency Call tracking: call and source agency or client Client’s CRM: qualification and sale supplied by client Sheet reconciled by contact Decision with a review date Processing agreement GDPR, article 28
The agency can produce the ads part on its own and sometimes the calls part; the sales part depends on the client. Without the matching key and the processing agreement, the central sheet cannot be built.

If the agency accesses the CRM, recordings or end customers’ numbers, it is processing data on its client’s behalf. Article 28(3) of the GDPR requires that processing to be governed by a contract setting out its subject matter, duration, purpose, type of data and obligations. Signing it before requesting the first export avoids halting the analysis halfway. Separating access between clients is covered in the guide to separating clients and access in an agency (in Spanish).

The records before any decision

Agency call tracking case study: a simulated example: table 1
SourceCallsAnsweredQualifiedSalesAdvertising
Google Ads4032206€600
Organic201883Not assigned
Meta Ads10631€150
Direct201852Not assigned
Unknown10820Not assigned
Total100823812€750 known

The dataset assumes that the qualified opportunities and sales in the table come from answered calls. “Not assigned” does not mean zero cost: SEO work, call handling, brand and tools also cost money. The known advertising total is not the business’s full acquisition cost. All amounts exclude VAT, for both spend and sales.

First finding: cost per call hides differences

Google Ads and Meta Ads both cost €15 per call: 600/40 and 150/10. That figure could lead you to treat them as equivalent channels. Looking at qualified opportunities, Google costs €30 per opportunity and Meta €50. Advertising cost per observed sale is €100 and €150 respectively.

Meta has only one sale in the example. A second sale would bring its cost down to €75. That is why an apparent difference does not justify an immediate cut: the evidence is thin and some opportunities may still be open. The report must state the observation period.

Second finding: a call-handling problem to investigate

80% of Google calls and 60% of Meta calls were answered. In total, eighteen attempts went unanswered. That suggests reviewing opening hours, the diversion destination, voicemail, capacity and callbacks. It does not prove that all eighteen calls were lost opportunities, nor does it justify assigning them hypothetical revenue.

Here the agency can only point to the symptom: call handling is the client’s responsibility. The useful step is to give the client the list of unanswered calls with date, time and source, and agree who reviews it, rather than an aggregate figure that nobody can act on.

Third finding: ten contacts have an unknown source

10% of calls are unattributed in the example. The agency should investigate why: a shared number, no link to the website visit or measurement not available. It should not spread those ten contacts across campaigns to make the report look complete.

It would also keep observed source separate from commercial credit. If someone arrived through several channels, a first-source rule is an analytical choice, not proof of causation.

Fourth finding: the verdict changes as the cohort matures

At 45 days, of the 38 qualified opportunities, 12 had been won, 14 lost and 12 were still open. Suppose the agency takes another cut at 90 days with the same calls and no new contacts added:

Agency call tracking case study: a simulated example: table 2
ChannelSales at 45 daysSales at 90 daysCost per sale at 45 daysCost per sale at 90 days
Google Ads68€100€75
Meta Ads12€150€75

With just one more sale, Meta costs the same per sale as Google. Had the agency cut Meta after the first cut, it would have acted on a difference that disappears once observation is complete. The practical rule is to fix in advance the cohort age you evaluate, as explained in the guide to conversion lag and cohorts, and never to compare channels of different ages.

What revenue adds and what it does not explain

If each sale in the scenario has a value of €600, defined excluding VAT and returns, attributed revenue at 45 days would be €3,600 for Google and €600 for Meta. Their ROAS would be 6 and 4. The twelve sales would total €7,200.

These ratios compare revenue with known advertising. They do not show profit: the cost of delivering the service and other expenses are missing. The margin-based calculation and cost per qualified lead are developed in the guide to cost per qualified lead and profitability. To see the same method applied by a business measuring itself, with time slots, Google Ads conversions and a business profile, read the home renovation company case study (in Spanish).

What the agency asks the client for before publishing the report

Agency call tracking case study: a simulated example: table 3
DataWho supplies itAcceptance check
Definition of qualifiedClient, in writingTwo people classify 10 calls the same way
Sales with number and dateClient, from the CRMEach sale links to a single call
Value of each saleClientExcluding VAT and returns
Spend per campaignAgencySame period as the calls
Test callsAgencyExcluded before calculating
Cut-off dateBothShown in the report alongside cohort age

With that list, the client call report can show every figure with its source. If the client wants to send sales back to Google Ads for bidding, the route is importing call conversions, which requires Google forwarding numbers and caller data.

Common mistakes and how to spot them

  • Spreading the unknowns. Symptom: the report adds up exactly to total calls with no unknown-source row. Check: ask for the rule used to assign each call with no source.
  • Cutting the cohort too early. Symptom: a channel with few sales and many open opportunities. Check: count open opportunities per channel before comparing cost per sale.
  • Mixing amounts with and without VAT. Symptom: ROAS changes between the CRM and the sheet. Check: note the basis of each column in its header.
  • Sales counted twice. Symptom: more sales than qualified contacts in some channel. Check: look for repeated phone numbers in the sales column.
  • Blaming the campaign for a call-handling problem. Symptom: the channel with the worst answer rate is cut. Check: compare its answer rate by time slot with the other channels at the same hours.

What this case still cannot decide

The example does not justify immediately switching off the campaign with fewer sales. Service mix, time to close or answering capacity may differ. The agency should keep the initial result and state a specific hypothesis: “In this time slot, unanswered calls prevent us from observing the commercial outcome.”

The next proposed test is to fix that coverage, record calls, opportunities and closed sales under the same rule, and allow the same observation period. Note any changes to ads, prices or staff that overlap with the test. If several things change at once, a later improvement does not show which one caused it.

The useful deliverable is a traceable decision with the question, the available data, the uncertainty, the action and a review date. The amounts and outcomes on this page remain simulated; they must not be cited as a return achieved for a CallsIQ client.

From the author's experience

After eleven years of supporting and implementing contact centre software, I always recommend that the client keep access to numbers, data and exports, with a hand-over clause written in for when the relationship ends. As a technical key to join a call with a sale, I prefer a unique call identifier propagated across systems; the phone number works as a supporting field, but the same person can call several times and an opportunity code can group different conversations together.

For unanswered calls, I would deliver a short summary by time slot and reason for non-answer, together with an actionable table showing time, queue and outcome: it is easier to fix a configuration or reinforce a shift when you know exactly when the problem occurred. For time to close, I would track cohorts at 30, 60 and 90 days and check the real timeframe before writing off a call that could still turn into a sale.

In reconciling telephony and sales, the piece of data that usually takes longest to arrive is a common reference and up-to-date statuses; I would ask for a stable identifier and an export with creation and closing dates, rather than matching by phone number alone. Nor would I attribute a specific campaign cut to an agency without having that case in front of me: the mistake I would actually check for is evaluating only sales closed within the same period as the calls, leaving out conversions that arrive weeks later.

Sources and limitations

Documentary review: . Content type: Worked exercise.

Scope: All data in this case is illustrative. They do not describe results from a client or a software test.

How to report a correction

How this guide was prepared

Official sources, explained calculations and clearly labelled examples. Read about our methodology and use of AI in writing.