Track and iterate

The audit keeps its value if three things exist after the presentation: one scoreboard, an experiment log and a monthly review that can change the plan. Without them, the roadmap drifts within a couple of months and the team goes back to shipping whatever is loudest.

I’ll stay with the fictional B2B SaaS at 2M ARR from the six-month roadmap. Every figure about it below is illustrative.

Keep one scoreboard tied to the growth equation

Take the metrics from your growth equation, and stop at five to seven. Each one gets a single source and a single owner, so nobody argues about whose number is right.

Metric	Source	Owner	Baseline	Month 3 target	Month 6 target
Qualified leads per month	CRM	Head of growth	25	35	50
Signups with a known first-touch source	Acquisition analytics	Head of growth	40%	80%	90%
Visits to feature and pricing pages from the blog	Acquisition analytics	Content lead	800	1,500	2,500
Signup to activated	Product analytics	Head of growth	30%	33%	38%

Qualified leads sit at the top on purpose. Signups can rise while the people signing up are the wrong size for you, so use signups as an early signal and decide on qualified leads.

Read by cohort of signup month, because a total mixes old and new behavior. This query assumes a signups table with the dates of each stage:

SELECT date_trunc('month', s.created_at) AS signup_month,
       s.first_touch_source,
       count(*) AS signups,
       count(*) FILTER (WHERE s.activated_at IS NOT NULL) AS activated,
       count(*) FILTER (WHERE s.qualified_at IS NOT NULL) AS qualified
FROM signups s
WHERE s.created_at >= date_trunc('month', now()) - interval '6 months'
GROUP BY 1, 2
ORDER BY 1, 5 DESC;

The latest cohorts will look weak because their leads have had less time to qualify. Compare cohorts at the same age. And write down the date of every tracking change: when UTMs arrive on blog CTAs, “direct” will drop and “blog” will climb without anything having happened in the market.

Turn each action into a written hypothesis

Every roadmap line that can succeed or fail gets a card, written before launch. The decision rule matters most, because writing it after the result lets anyone read the result as they like.

Field	Example
ID	EXP-014
Hypothesis	Adding a CTA to the matching feature page after the first section of our 20 most-read articles raises qualified leads from the blog, because today’s CTAs sit at the bottom
Main metric	Qualified leads from the blog, by first touch
Guardrail	Time on page and signups from other sources stay stable
Method	Before and after by cohort, on those 20 articles
Duration	Eight weeks, set in advance
Decision rule	At least +30% over the prior eight weeks: roll out to all articles. Below +10%: revert and rework the offer
Result and decision	Filled in at the review

Tag each test so that it can be isolated in your reports:

?utm_source=blog&utm_medium=cta&utm_campaign=exp-014_feature-cta&utm_content=after-section-1

Log the failures as carefully as the wins. After a year, the file of what did not work is worth a lot to the next person who proposes the same idea.

Check what your volume can actually prove

A/B testing needs traffic that many B2B companies don’t have. Before launching one, estimate the sample with a common approximation, for a two-sided test at 5% significance and 80% power, where p is the baseline rate and d the absolute difference you want to detect:

n per variant = 16 * p * (1 - p) / d^2

Say the SaaS gets 400 signups a month and activates 30% of them.

Smallest lift to detect	Per variant	Total	Months at 400 signups
30% to 35%	1,344	2,688	6.7
30% to 40%	336	672	1.7

A five-point lift would eat the entire program for a single test, so this team only A/B tests changes big enough to move ten points. Everything else goes through weaker methods, stated as such: before and after by cohort, rollout to one country or segment first, or customer calls and session recordings. Set the duration in advance and don’t stop when the curve looks good, since checking repeatedly and stopping on a favorable moment inflates false positives.

Hold a review that can change the plan

Rhythm	Length	Who	Output
Weekly	30 minutes	Action owners	What shipped, what is blocked
Monthly	60 minutes	Owners and founder	Scoreboard read, decisions on finished experiments
End of month 3	Half a day	Owners and founder	Scores updated, plan for months 4 to 6 reworked
Month 6	One to two days	Whoever ran the audit	Light re-audit, next plan

The monthly review always ends with one of these decisions per finished experiment.

Result	Decision
Rule met, guardrails fine	Roll out and make it the new standard
Rule met, guardrail broken	Don’t roll out, look at the segment that lost
Flat	Stop. One retry at most, with a different hypothesis
Worse	Revert, and record why you think it failed
Inconclusive, too little volume	Decide on judgment and label it as judgment, or test something bigger

Read rising curves as critically as falling ones. On Derrick, the last 12 months brought 42.4K clicks and 19.2M impressions, which looks great until you see the 0.2% click-through rate: the site gets seen far more often than it gets chosen, and that gap is the next piece of work.

At month 6, run a light re-audit

Fix a bottleneck and it moves, so the verdict from six months ago is probably out of date. Replay the short version of the audit, and start with the data, since a new tool or a site change can have broken something along the way.

Check	Question to settle
Tracking	Does each conversion still have a source? Has the multi-touch model replaced last click in decisions?
Funnel	Which stage now has the worst rate, and for which segment?
Parked list	Which reopening conditions are met?
Experiment log	Which kinds of action worked here, and which never did?
Capacity	Did the team deliver what the plan assumed? If not, cut the next one

From this comes a new one-page verdict and a new six-month plan, built the same way as the first, with real results replacing most of the estimates. The first plan will have been too ambitious somewhere, and now you know where.

Your next step: create the scoreboard this week with the baselines you have today, put the monthly review in everyone’s calendar for the next six months, and write the card for the first experiment before it ships. A baseline can’t be reconstructed once the change is live.