Reading your dashboard
A field-by-field guide to the Cancel Flows dashboard: what each number means, when to trust it, and how the comparisons are kept honest.
How to read this page
The Cancel Flows dashboard turns every cancellation moment into a small experiment. About 15% of your customers always see your fixed offer, and the rest see adaptive offers. By comparing the two groups over time you can see, in plain numbers, whether the smart offers are actually keeping more of your customers.
The page is laid out top to bottom:
- The headline cards. Offers Accepted, Save Rate, Value Boost, and Saved MRR Kept.
- Offer Performance. A heatmap of which discounts and durations customers accept most.
- Segments. The same numbers, broken down plan by plan.
- Recent Sessions. The latest cancel flows and what happened in them.
54
89 offers presented
61%
vs 44% fixed rate · significant
+39%
vs fixed rate
$3,180.00
monthly billings kept
Four cards sit at the top: Offers Accepted (a volume figure), Save Rate (quality: acceptances that actually stuck), Value Boost (how much better adaptive is than fixed), and Saved MRR Kept (the monthly billings you didn't lose).
Offers Accepted
The big number is how many customers applied an offer and kept their plan. The line underneath, offers presented, is how many cancel flows actually showed an offer. The gap between the two shows how often customers walked away anyway.
This card is a volume figure, not a quality figure. Two plans can accept the same number of offers while having very different save rates, which is why the next cards matter.
Save Rate
Save Rate is the share of customers who accepted an offer and are still subscribed roughly a month later. A discount only counts once it actually keeps the customer around, so this is the number that reflects real revenue staying put.
The card shows your adaptive save rate. The line underneath compares it to your fixed rate: the customers in the control group who always saw the plain fixed offer. A note like "data still rolling in" means there isn't enough data yet to trust the comparison, which is normal in the first weeks.
Value Boost
Value Boost is how much better the adaptive offers perform than your fixed offer:
| What you see | What it means |
|---|---|
| +22% (green) | Adaptive offers save noticeably more. The result is statistically significant. |
| ≈ 0% (gray) | No significant difference yet. Either offers are early in their learning or neither arm is winning. |
| −8% (red) | Discounts are too generous. Tighten your discount range in Settings, raise the floor or lower the ceiling, and let it keep learning. |
If acceptance rates are high but Value Boost is negative, the discounts are too generous: customers are taking offers that save them but cost you more than the fixed offer would. Tighten the range in Settings and let the bandit adjust.
The card stays gray and near zero until the difference is statistically significant, so early noise never gets mistaken for a win or a loss.
Saved MRR Kept
The extra monthly revenue those accepted offers represent. It is the discounted amount of each subscription that is still active, added together. Think of it as the monthly billings you would have lost to cancellations but did not.
It grows as offers are accepted and settles as the retention window passes: a customer who accepted but cancels anyway within about a month is removed from this number.
The control group
Every comparison on this page depends on a small, stable control group. A fixed share of your customers always receives your fixed offer instead of an adaptive one. Their results are the fixed rate you see quoted everywhere.
Because the split is by customer, a customer stays in the same group every time, and the comparison stays honest over time.
How the A/B split works
~15% of customers always see your fixed default offer. Their results are the fixed rate used for comparison.
The rest see offers chosen by the bandit: discounts and durations that learn from what customers accept.
Because the split is by customer, the same customer stays in the same group every time. That keeps the comparison honest over the long run.
Statistical significance
Numbers move around by chance, especially early. Before this page calls any difference real, it runs a statistical check and only labels a result once there is enough data:
- Significant. Confident at 99% that the difference is real. Shown in full color with the percentage.
- Approaching. Confident at 90% or better. Shown in amber with an ≈ so you know it is promising but not proven yet.
- No signal yet. There is not enough data, or no meaningful difference. Shown in gray.
A gray "no signal yet" is not bad news. It just means the honest answer right now is that there is nothing to report, and we are waiting for more data.
The significance ladder
Gray is not bad news. It just means the honest answer is "waiting for more data." The page refuses to call any difference real until there is enough evidence.
Offer Performance
The heatmap shows acceptance rate (not save rate) for every discount and duration combination the adaptive offers can pick. One cell is one offer: how often customers presented that offer took it.
This is where you watch the learning happen. As data builds, the cells that customers actually respond to stand out visually. A plan with little data starts from the middle of the grid and drifts toward the cells that win.
| Discount | 1mo | 2mo | 3mo |
|---|---|---|---|
| 50% | 39% | 43% | 55% |
| 40% | 50% | 53% | 63% |
| 30% | 46% | 52% | 67% |
| 20% | 37% | 42% | 45% |
Accepted / Presented · shaded by acceptance rate
Rows are discounts, columns are durations. Cells show acceptance rate: how often customers presented that offer took it. The combos that win stand out visually as the data builds.
Segments
The same comparison, plan by plan. Each row has:
- Save rate. With a small bar showing adaptive (brand color) against the fixed control (gray) for that plan.
- Uplift. How much better adaptive is than fixed for that plan, gated by the same significance rules as Value Boost.
- Best offers. The offers with the most presentations and their acceptance rates, so you can see what is working per plan at a glance.
Smaller plans naturally take longer to reach significance. Treat a segment that reads"no signal yet" as early rather than flat.
Recent Sessions
The latest cancel flows in the selected time window, in reverse order, with the session status, customer, subscription, the offer that was shown, and when the link was created. The status tells you where each customer ended up:
- Started: link created, not opened yet.
- Offer shown: the customer opened the page.
- Offer accepted: discount applied and the plan stays active.
- Offer declined: the customer turned the offer down.
- Canceled: the customer continued and the subscription is set to cancel at period end.
Where each session ends up
Started
link created, not opened
Offer shown
customer opened the page
Accepted
discount applied, plan stays
Declined
customer turned it down
Canceled
set to cancel at period end
How the numbers are kept honest
A few rules run in the background so the dashboard never oversells:
- Acceptances are reconciled against Stripe on a rolling 30-day window. A customer who accepts and then cancels anyway stops counting.
- Comparisons are always vs the control, never vs a guess, so the baseline is measured, not assumed.
- Cold start protection. Numbers only appear once enough offers have been presented for them to be meaningful.
- Every number respects the date window. Use the window picker at the top of the page (last 7 days, last 30 days, last 90 days, or All time) to look at a specific campaign or teardown run. Every metric, chart, and the offer heatmap re-scope to that window, and All time covers your entire history. Nothing is silently dropped.