Action Center & opportunity radar
In plain English
This screen has two halves.
The top half: five jobs to do today. The app checks nine things about your shop — are there campaigns waiting for approval? Any products about to run out? Any subscriptions failing to charge? — and shows you the five most worthwhile.
They're ordered by how much money they're worth divided by how much work they are. So a £2,000 job that takes five minutes beats a £3,000 job that takes two days. That's deliberate: you'll actually finish the first one.
The bottom half: money you're not collecting. Not jobs exactly — more like five pots of money sitting there. Abandoned baskets. Customers who've drifted away. Subscriptions failing on payment.
Each pot has a confidence label, and this is the most useful thing on the screen:
- High — the app counted actual things. Trust this number.
- Medium — based on a reasonable average. Roughly right.
- Low — an educated guess. Worth testing, not worth planning around.
A "high confidence £8,000" is worth chasing before a "low confidence £14,000." Most people get this backwards and go for the big number first.
How to use it
- Click "Intelligence" in the left-hand menu. This is the screen it opens on.
- Read the five job cards from top to bottom.
- Each card tells you what to do and why. Read the "reason" line — it takes five seconds and catches mistakes.
- Click the button on the card to go and do it.
- Do anything that takes under five minutes right now. Don't leave it for tomorrow.
- Scroll down to the money pots. Sort by confidence, not by amount.
- Start with the "high confidence" ones — they're the quickest, surest wins.
⏱ ~10 min/day · 💳 Pro+ · 🎯 The five best moves available to you today
Why this matters for your business
There's a specific failure mode that hits every growing store. You know roughly what's wrong — you always have five or six things half-noticed in the back of your mind. What you don't know is which one is worth doing first, because they're incomparable. Reordering stock, sending a winback, approving a campaign, fixing a failing subscription — different teams, different timescales, different units.
So the choice gets made by whatever's most salient: the thing someone emailed you about, the thing that annoyed you yesterday, the thing that's easiest. Salience is a terrible prioritisation function, and over a quarter it costs real money.
The Action Center makes them comparable. Every candidate is reduced to two numbers — estimated dollar impact, and effort — and ranked on the ratio. Five items, because five is what a small team finishes. The result is that the easy-but-worthless task stops beating the awkward-but-valuable one.
The opportunity radar handles the other half: money that isn't a task. Nobody is going to "do" cart recovery today, but $3,100 of unrecovered carts is a fact worth knowing when you're deciding where the next project goes.
What this typically unlocks
| Outcome | Typical result |
|---|---|
| Daily prioritisation time | ~0 — the ranking is the decision |
| Highest-impact task actually done first | Consistently, rather than by salience |
| Money categories you're not capturing | Five, quantified with confidence |
| Trust in the recommendation | High — every rule is inspectable |
| Actions requiring a separate tool to execute | 0 — every item deep-links |
What you actually get
The five ranked actions
Nine rules run on every load. Each checks one condition and emits at most one candidate. Candidates are scored and the top five shown.
| Rule | Fires when | Effort | Typical impact basis |
|---|---|---|---|
| Approve pending campaigns | Campaigns awaiting approval | Low | Expected campaign revenue |
| Enable cart recovery | Open carts with no recovery running | Low | Open carts × recovery rate × cart value |
| Send winback to at-risk customers | Customers flagged at churn risk | Low | At-risk count × LTV fraction |
| Reorder SKUs at stockout risk | Velocity vs stock predicts a stockout | Medium | Projected lost sales |
| Apply price-test winners | A price test has a winner not yet applied | Low | Measured lift × volume |
| Bundle cross-sell pairs | High-confidence affinity pairs not bundled | Medium | AOV uplift × reachable orders |
| Review critical anomalies | Unreviewed critical anomalies exist | Low | Severity-weighted |
| Reactivate dormant customers | Dormant segment above threshold | Low | Dormant count × reactivation × LTV |
| Recover failing subscriptions | Subscription payments failing | Medium | Failed contracts × monthly value |
Each card carries:
| Field | Purpose |
|---|---|
| Title | An imperative — "Reorder 4 SKUs at stockout risk" |
| Reason | Why this fired, in plain language |
| Estimated impact | Dollar figure, when computable |
| Effort | Low / Medium / High |
| Action button | Deep-link to where you do it |
Scoring
score = estimated impact ÷ effort weight
| Effort | Weight |
|---|---|
| Low | 1 |
| Medium | 2 |
| High | 4 |
A low-effort $2,000 action scores 2,000. A medium-effort $3,000 action scores 1,500. The smaller number wins, because you'll finish it — and probably finish something else after it.
When a rule can't compute a dollar impact, a base impact stands in so the item still ranks rather than disappearing.
The opportunity radar
Five estimators, each answering "how much money is sitting here that we aren't capturing?"
| Opportunity | How it's estimated | Confidence |
|---|---|---|
| Cart recovery | Open carts × 10% × average cart value | High |
| Dormant winback | Dormant customers × 8% × (average LTV ÷ 12) | Medium |
| Inelastic price increase | SKUs with elasticity |e| < 1 × monthly volume × 5% × AOV | Medium |
| Cross-sell | High-confidence pairs × reachable orders × 15% AOV uplift | Low |
| Subscription dunning | Failed contracts × monthly subscription value | High |
Each shows the amount, the unit count behind it, and a deep-link to act.
Why confidence varies so much
This is the most useful column on the radar, and the one most people ignore.
- Dunning is high confidence because it's arithmetic. These specific contracts failed, at these specific prices. There's no behavioural assumption — the customer already committed.
- Cart recovery is high confidence because 10% is a well-established recovery rate and the cart values are actual observed carts.
- Winback and price-up are medium. Both rest on a behavioural rate (8% reactivation) or an economic model (elasticity) that's measured but noisy.
- Cross-sell is low. It stacks an affinity model, an uplift assumption, and a reach assumption. Directionally useful, numerically soft.
Sorting by confidence rather than amount is almost always the better read. A high-confidence $8,400 beats a low-confidence $14,000 for deciding where to spend next week.
The cross-sell cap
Cross-sell reach is computed as 2% of orders per high-confidence pair, capped at 30% of all orders.
Without the cap, a shop with 40 pairs would show an estimate implying cross-sell touches 80% of the order base — a number that's obviously wrong to anyone who thinks about it for a moment, and quietly corrosive to trust in every other figure on the page. The cap makes the estimate smaller and defensible.
How it works (without the technical bits)
Rules, not a model — and why
A learned ranker would probably order these marginally better. It was rejected deliberately.
This is the surface that tells you what to do now. If you can't see why an item is on the list, you won't act on it — and if you act on it once and it's wrong, you'll never trust the screen again. Nine legible rules give you something a model can't: when the tool is wrong, you can see exactly which condition misfired, and you know whether to distrust that one rule or the whole list.
Legibility is worth more than a small accuracy gain on a surface whose entire value depends on being believed.
Real merchant scenarios
Scenario A — Where salience was costing $40k a year
Setup. Kitchenware merchant, one-person marketing team. Daily priorities came from whatever was in the inbox.
First Action Center load:
| # | Action | Impact | Effort | Score |
|---|---|---|---|---|
| 1 | Recover 12 failing subscriptions | $4,100 | Medium | 2,050 |
| 2 | Enable cart recovery — 340 carts unrecovered | $1,900 | Low | 1,900 |
| 3 | Apply 2 price-test winners | $1,400 | Low | 1,400 |
| 4 | Reorder 6 SKUs at stockout risk | $2,600 | Medium | 1,300 |
| 5 | Send winback to 84 at-risk customers | $980 | Low | 980 |
The revelation. Items 1 and 3 had been available for over two months. Nobody had emailed about them, so they'd never been salient. The two price-test winners had been sitting in a results tab, already proven, unapplied.
Twelve months of that pattern, extrapolated from the first quarter's recovery, was worth roughly $40,000.
Scenario B — The low-effort item that outranked a bigger one
Setup. Two candidates in the same load:
Reorder 3 SKUs at stockout risk $3,000 Medium (÷2) → 1,500
Approve 4 pending campaigns $2,200 Low (÷1) → 2,200
Merchant's instinct was to do the reorder — bigger number, more obviously urgent.
What the ranking knew. Approving campaigns takes four minutes. The reorder needed supplier confirmation, a purchase order, and budget sign-off — realistically two days.
What actually happened. They approved the campaigns at 9:04am, then started the reorder. Both were done by Thursday. Had they started with the reorder, the campaigns would have slipped another week and missed their send window entirely — a real $2,200, permanently gone.
Scenario C — Dunning, discovered
Setup. Meal-kit subscription business. Churn had crept from 4% to 7% monthly over a quarter. The team was building a retention campaign.
Opportunity radar:
Subscription dunning recovery $11,200 High confidence 68 contracts
Cause. 68 subscriptions were failing on payment — expired cards, mostly — and being auto-cancelled with no retry. They'd been counted as churn in every report.
Why high confidence mattered. The team had been about to spend six weeks on a retention campaign targeting a churn problem that was, in large part, a billing problem. Because the estimate was arithmetic rather than modelled, they trusted it enough to change plan immediately.
Result. Dunning sequence live in nine days. Monthly churn back to 4.4% within two cycles. The retention campaign shipped a month later, against the real, smaller problem.
Scenario D — Reading the confidence column correctly
Setup. Merchant sorted the radar by amount:
| Opportunity | Amount | Confidence |
|---|---|---|
| Cross-sell | $14,600 | Low |
| Inelastic price increase | $9,200 | Medium |
| Cart recovery | $4,800 | High |
| Dunning | $2,100 | High |
They started with cross-sell. Three weeks building bundles. Measured uplift: about $2,900 — roughly a fifth of the estimate.
What the confidence label was telling them. Cross-sell stacks three assumptions; each is reasonable, and multiplied together the error compounds. The label said "low" precisely because this outcome was likely.
What they did next. Cart recovery and dunning — $6,900 combined at high confidence — took four days and returned $6,400.
Recalibrated approach. Work high-confidence first, treat low-confidence estimates as experiments with a test budget rather than projects with a plan.
Scenario E — When an estimate was wrong, and it was fine
Setup. "Reorder 4 SKUs at stockout risk — est. $3,200."
The merchant checked the arithmetic and found one of the four SKUs was being discontinued. Its projected loss, roughly $1,400, wasn't real.
Because the rule was inspectable, this took two minutes rather than being an unexplained discrepancy later.
The outcome. They reordered the other three, archived the fourth so it stopped appearing, and kept trusting the surface. A black-box recommendation with the same error would have produced either blind compliance or a permanent loss of confidence in the tool.
Best practices
✅ Work the five top-down and don't re-sort them. The ratio is the point; overriding it with intuition reintroduces the salience problem.
✅ Read the reason line before acting. It takes five seconds and catches the discontinued-SKU class of error.
✅ Sort the radar by confidence. High-confidence money is faster, more certain, and builds the trust you need to attempt the speculative work later.
✅ Treat low-confidence opportunities as experiments. Budget a test, measure the actual, then decide.
✅ Check for repeat offenders. The same rule firing every day for a fortnight means a systemic problem — wire it into Automations instead of doing it by hand.
✅ Do the low-effort items immediately. Anything under five minutes should never survive to a second load.
❌ Don't sum the radar. The categories overlap — a dormant customer may also have an abandoned cart — and each is an upper bound on its own assumption.
❌ Don't ignore an item because it's small. Score already accounts for size; a $980 low-effort item outranking a $3,000 one is the system working.
❌ Don't treat estimates as commitments. They're conservative upper bounds on a stated assumption, not forecasts.
❌ Don't chase the biggest number on the radar first. Scenario D is the most common mistake made on this screen.
Plan tiers
| Capability | Free | Starter | Pro | Agency | Enterprise |
|---|---|---|---|---|---|
| Ranked daily actions | — | — | ✓ | ✓ | ✓ |
| All nine rules | — | — | ✓ | ✓ | ✓ |
| Opportunity radar | — | — | ✓ | ✓ | ✓ |
| Confidence labelling | — | — | ✓ | ✓ | ✓ |
| Deep-links to execution surfaces | — | — | ✓ | ✓ | ✓ |
| Subscription dunning opportunity | — | — | ✓ | ✓ | ✓ |
| Actions in the weekly digest | — | — | ✓ | ✓ | ✓ |
| Multi-store action roll-up | — | — | — | ✓ | ✓ |
Frequently asked
Why only five actions? Because five is what a small team completes in a day alongside normal work. A longer list gets triaged into "later."
Why did a rule not fire when I expected it to? Each rule has a specific trigger condition. If no campaigns are pending approval, that rule emits nothing — the list shows the next-highest candidate instead.
Are these the same as Daily Ops actions? No, and they can overlap. Daily Ops ranks findings emitted by every engine platform-wide. This ranks nine growth-specific rules computed here. Both are legitimate; this one is narrower and more analytical.
How is estimated impact calculated? Differently per rule, and each states its basis on the card. Cart recovery uses open carts × 10% × average cart value; dunning uses failed contracts × their actual monthly price.
Why is cross-sell always low confidence? It multiplies an affinity model by an uplift assumption by a reach assumption. Each is defensible; compounded, the uncertainty is large. The label is honest about that.
Can I dismiss an action? Actions are computed live from current conditions rather than stored — resolve the condition and the item stops appearing. To suppress a recurring one, fix the underlying state (archive the discontinued SKU, close the campaign queue).
Does the radar double-count with Revenue Leaks? They overlap on cart recovery by design — one is a prioritisation projection in the execution queue, the other an opportunity estimate here. Both apply the same conservative 10% rate, so the figures agree.
Why is my price-increase opportunity zero? Either no SKUs have measured elasticity below 1, or there isn't enough price-variation history to measure elasticity yet.
See also
- Growth Intelligence overview — the hub and its five sections
- Revenue command center — KPIs and revenue diagnosis
- Customer intelligence — the dormant and at-risk segments behind these rules
- Goals & alerts — targets and the weekly digest
- Daily Ops overview — the platform-wide execution queue
- Revenue Leaks — the leak view of recoverable money
- Automations — turning a repeating action into a rule
- Cart recovery & post-purchase — executing the recovery action