Define the outcome. Then judge what changed.
Atlas defines the business result, identifies the evidence needed to measure it and checks what changed before recommending more investment. Clicks, inquiries, customers and contribution are different measures. The plan should make those differences visible.
A qualified outcome, agreed in advance.
The right measure depends on the business. It could be an attended appointment, an accepted sales opportunity, a first purchase or a repeat customer. We specify what qualifies, what is excluded, who owns the definition and which system records it.
Before work begins, establish the baseline, unit, observation window and target. Compare the same service, market and outcome definition. If the starting point is unknown, the first task is to establish it.
- Signals
- Impressions, clicks and page engagement show attention or behavior. They help explain a journey but do not establish revenue.
- Business outcomes
- Qualified leads, customers and repeat purchases require business rules and records beyond an advertising interface.
- Economics
- Acquisition cost and contribution require an explicit cost basis, customer definition and accounting period.
Use the smallest set of sources that answers the question.
An engagement may use the sources below where access and data quality permit. We confirm the available integration or export method during scoping. A list of systems is not a promise of a working connection to every account.
| Source | What it can establish |
|---|---|
| Search Console | Search impressions, clicks and reported indexing states within its scope; not closed sales. |
| Analytics | Recorded website events and journeys, subject to collection, consent and identity limitations. |
| Advertising platforms | Spend, delivery and platform-attributed actions; not automatically incremental customers. |
| CRM or booking records | Qualification, sales progress and customer outcomes when definitions and records are consistent. |
| Financial records | Recognized revenue and defined costs where the client can supply an appropriate reconciliation. |
A tag on the page is only a starting point.
For the agreed conversion, follow a test from the customer action to the receiving system. Confirm the event arrives once, carries the right non-sensitive context and can be reconciled with the business record. Document missing consent, duplicate events, delayed updates and unavailable joins.
If an inquiry is recorded but its qualification is unknown, report inquiries. Do not relabel them as qualified leads. If the link to sales is missing, fix that link before presenting customer acquisition cost.
What if the sources disagree?
Compare their definitions, time zones, attribution windows, filters and update delays. Keep separate source totals where they cannot be reconciled. Do not add platform-attributed conversions together and call the sum unique customers.
Use an acquisition ceiling the business can afford.
Contribution means revenue after the specified variable costs. It is not interchangeable with revenue, gross margin under another definition or lifetime value inferred from a short sample. The acquisition decision also needs the business’s cash constraints and capacity.
Synthetic example: $2,000 of media spend produces 20 qualified appointments and ten new customers in the defined window. That is $100 media spend per qualified appointment and $200 media spend per customer. Agency, production and other acquisition costs would increase the all-in figure. These invented figures illustrate arithmetic, not Atlas performance.
Without revenue, contribution and a valid comparison, that example says nothing about profit or incremental return. We state the cost scope beside the number.
Attributed is not the same as caused.
Attribution assigns credit under a defined rule. A before-and-after comparison describes a change. Neither alone establishes how many outcomes would have occurred without the intervention.
Where a causal answer is necessary and the scale supports it, the plan may require a randomized holdout or a carefully justified comparison. Seasonality, selection differences, other campaigns and changes in sales capacity belong in the interpretation.
When is a sophisticated model the wrong next step?
Media mix modelling needs sufficient reliable history and variation, with a defensible specification. A small or inconsistent dataset does not become trustworthy because the model is complex. Start with definitions and reconciled outcomes; scope advanced analysis only when the inputs can support it.
Can every channel be measured the same way?
No. Search demand capture, awareness activity and retention can play different roles and produce outcomes on different timelines. Agree the role, evidence and decision rule for each; label proxy measures instead of pretending they are direct sales.
Decide how you will learn before the result arrives.
A useful test names its opportunity, intervention, hypothesis, metric, baseline, target, observation window, source, owner and stop rule. The definitions remain attached to the result. If the plan changes, record the change rather than quietly moving the goalposts.
| Finding | Next decision |
|---|---|
| Comparable outcomes improve within agreed cost and capacity limits | Consider continuation or a bounded expansion; review alternative explanations. |
| Outcome quality falls or cost exceeds the agreed stop rule | Pause or revise the intervention; inspect the evidence before adding spend. |
| The source, sample or comparison is inadequate | Repair measurement or extend the observation window; keep impact unproven. |
A report should make the next decision easier.
The readout puts the agreed outcome, measurement period, source and change beside the work completed. It names the missing data and alternative explanations, then makes a recommendation: continue, revise, stop or investigate.
We agree reporting cadence and responsible people in the scope. Automated collection can reduce repetition; interpretation and changes to investment remain accountable decisions. The complete live connection from every lead to revenue is not assumed, and a sample report is not evidence of a client result.