Every Monday, Lowveld Ledger's sales and marketing team looked at the same dashboard. It said that the team had sent 3,200 emails the week before, made 410 calls and sent about 150 LinkedIn messages. The numbers went up and down a little. Nobody could say what they had learnt from them in the last quarter.
That dashboard is common, and it is not useless. Activity tells you whether the work was done. What it cannot tell you is whether the work was the right work, and when it is the main thing a team reviews, it quietly becomes the goal. More emails start to feel like progress.
CloudNala's view is that a go-to-market system should treat activity as an input and learning as the unit of progress: a hypothesis tested, a result compared with a holdout, and a decision recorded. Revenue is the outcome you want. Learning is how you find out which of your actions actually move it.
This is Part 9 of the series. Earlier parts set up plays with kill criteria and signals stated as hypotheses. This part is about measuring whether any of it is working.
Lowveld Ledger is fictional, and its systems and figures are illustrative. It is a Mbombela accounting and payroll software company for SMMEs with about 2,300 paying customers, a nine-person sales and marketing team and a GTM engineer.
Four layers of measurement
It helps to separate four layers that dashboards tend to mix together.
Activity is what the team did: emails sent, calls made, events run. It answers "did we do the work?" and it is entirely within the team's control, which is why it is so popular to report.
Output is how the market responded in the short term: replies, meetings booked, leads accepted by sales. It answers "did anyone respond?"
Outcome is what changed for the business: paying customers, revenue, retention at 90 days. It answers "did revenue change?" It is what everyone cares about, and it arrives slowly and with many causes mixed in.
Learning is what the team now knows that it did not know before: which hypotheses were confirmed, which were refuted and what was decided as a result. It answers "what do we now know?" It is the only layer that tells you what to do differently next month.
None of these layers replaces the others. The mistake is to review only the first one, or to assume that improvements in the first will flow through to the third. They often do not. An SDR team can double its call volume while sales-accepted leads stay flat, because the extra calls go to accounts that were never going to buy.
The experiment log
Learning has to be written down or it evaporates. Lowveld keeps an experiment log: one row per test, whether the test is a play, a change to a signal, a scoring adjustment or a pricing-page experiment. Each row has nine fields.
| Field | Example |
|---|---|
| Hypothesis | "Trials that start a test payroll and stall will convert more often if offered a set-up call" |
| Play or signal | "Trial payroll stalled" play |
| Metric | Trial-to-paid within 30 days |
| Holdout | 20% of eligible trials, randomly assigned |
| Sample | 200 trials |
| Window | 8 weeks |
| Result | Higher conversion than the holdout; in-app prompt drove almost all bookings |
| Decision | Revise and re-pilot without the AE follow-up email; scale if the result holds |
| Date | Review date, and who was in the room |
The decision field is the one that makes the log a management tool rather than a research archive. "Interesting" is not a decision. "Scale", "revise and re-pilot", "retire" and "inconclusive, extend to 400 trials" are.
The log also records failures, and these are often the most valuable rows. At Lowveld, a row that says "industry-specific outbound to construction firms: replies but no trials, because their payroll is mostly weekly" stops someone repeating the same experiment in a year's time.
Leading and lagging indicators
The difficulty with outcomes is that they lag. At Lowveld, a trial lasts 30 days. A customer is not counted as retained until they have paid for 90 days. Anything that changes at the top of the funnel today will not show up in retention for four months.
The answer is to pair every lagging indicator with a leading indicator that the team has evidence for, and to treat the link between them as a hypothesis in its own right. Lowveld uses three pairs:
- Test payroll completed during the trial leads trial-to-paid conversion. In Lowveld's illustrative data, trials that complete a test payroll convert far more often than those that do not.
- Sales-accepted leads lead opportunities created, which is why the accept-or-reject step from Part 8 matters for measurement as well as for handoffs.
- First live payroll within 14 days of signing leads 90-day retention.
Each link is reviewed every quarter. If the leading indicator improves and the lagging one does not follow, the link is weaker than assumed, and the team has learnt something important about its own model.
Cohorts beat monthly totals
Monthly totals hide almost everything a team needs to know. If Lowveld signs 55 customers in October, those customers came from leads that arrived in August, September and October, through different plays, at different points in the season. A change made in September is diluted by everything else.
A cohort view groups leads by the month they arrived and follows each group forward.
Reading across a row shows how one month's leads progressed. Reading down a column compares months at the same stage. In the illustrative figures, September's leads are the first full month exposed to the "Trial payroll stalled" play, and their day-30 paid figure is visible at the end of October, well before September's effect could be separated from the noise in a monthly total.
The "not yet" cells matter too. A cohort view makes it obvious which results are still maturing, which stops anyone from comparing September's 30-day figures with June's 90-day figures and concluding that something has gone wrong.
Attribution humility
The question every marketing team is asked is "which channel produced this customer?" The honest answer is usually "several, in an order we only partly observed".
Attribution models assign credit across touches: first touch, last touch, linear, time-decay and others. They are useful for spotting patterns. They are not measurements of cause, and different models applied to the same data can rank channels very differently. A company that moves budget based on one model's ranking is often moving it based on the model's assumptions.
Lowveld uses three habits instead of relying on any one model:
- Holdouts for anything automated. If a play has a holdout, its effect can be estimated directly, without attribution.
- Ask customers. A required "how did you first hear about us?" field at sign-up, in the customer's own words, is crude and surprisingly informative. It is the main reason Lowveld knows how much of its business starts with an accountant.
- Record the partner. Every partner-referred account carries the referring practice, so the partner channel's contribution never depends on whether a tracking link survived.
The attribution report is still produced. It is simply read as one input to a conversation, not as a verdict on budgets.
The monthly learning review
All of this comes together once a month in a 60-minute meeting. At Lowveld, the head of sales chairs it, and the GTM engineer, the marketing lead, the partner manager and the customer success lead attend. The agenda is fixed.
- Ten minutes: outcomes. Paying customers, revenue and 90-day retention against plan, shown as cohorts.
- Twenty minutes: experiments due for review. Each owner presents the result against the holdout and the kill criterion, and the group records the decision in the log.
- Ten minutes: leading indicators. Are the leading-lagging links still holding?
- Ten minutes: harm. Complaints, opt-outs, rejected handoffs and any governance exceptions.
- Ten minutes: next month. Which backlog plays start, which experiments extend, and what the team expects to learn.
Activity numbers are available in the pack but are not on the agenda. If they matter, they will come up as an explanation for an output or outcome. They are never the topic.
The meeting ends with one sentence written into the log: "This month we learnt that…". When that sentence is hard to write, the team is running too few real experiments, or running them without holdouts.
Start this week
- Relabel your current dashboard into the four layers. Count how many of its numbers are activity. If it is more than half, add at least one outcome shown as cohorts.
- Start an experiment log with the nine fields above, and backfill the last three things your team tried, including the ones that did not work.
- Book a 60-minute monthly learning review with a fixed agenda and a rule that every reviewed experiment ends with a recorded decision.
Next: Part 10, The GTM system needs governance too. All parts are on the series page.
How CloudNala can help
CloudNala helps go-to-market teams measure what their actions actually change. We restructure dashboards into activity, output, outcome and learning, build cohort views from your CRM, product and billing data, set up holdouts for automated plays and introduce an experiment log and monthly review that end in recorded decisions. We are candid about what attribution data can and cannot tell you.
Work with CloudNala
CloudNala helps organisations move from technology ambition to practical execution across cloud, AI, data, platform engineering and digital services.
Whether you are exploring AI, modernising your cloud environment, building a public-sector digital service, or turning an idea into a working MVP, we can help you shape the roadmap and deliver the next step.
Book a Go-to-Market Systems Review or write to us at consult@cloudnala.co.za