Order completion rate at 1mg has plateaued. How would you improve it?

Before I reach for a lever, I need to know what actually dropped, and what data I'm working with.

Raw "order completed" is the wrong target here, this is medicine, so a completed-but-wrong order is worse than no order at all.

Stage-level drop-off is uniform, so that axis is a dead end, the same signal I'd read in any funnel: if one step were broken I'd expect a spike there relative to the rest, and there isn't one. With thin instrumentation I shouldn't expect a re-ask to suddenly produce a clean breakdown either. So I'd pivot the axis, not where in the funnel, but what kind of order, and then who.

Given the signal points at first-time, new-condition Rx buyers, the causes bucket into three:

Uncertainty about correctness. Is this the right medicine and dosage, is a generic substitute safe, will it even be approved. A first-timer has no prior experience to anchor on.

Invisible verification. The order sits in pharmacist review with no visibility into status or timeline, which reads as limbo, not progress, to someone who's never been through this flow before.

Proxy-buying complexity. A meaningful share of new prescriptions are uploaded on behalf of a parent or dependent, where the buyer doesn't know all the details themselves and hesitates to commit without confirming.

Impact against effort, all targeted at the first-time-Rx segment specifically:

Do first (cheap, content or policy only, ships fastest):

1. Live verification status. "Your prescription is being reviewed, typically 20 minutes" instead of silence.
2. A "what happens next" timeline shown right after upload. Review, then substitution check if needed, then confirm, then payment. Distinct from live status, this orients a first-timer who's never seen the flow before, rather than updating them mid-wait.
3. Real-time upload-quality check. Flag a blurry photo, a missing doctor signature, or unclear dosage at the moment of upload, not after a wait that ends in rejection. This kills the worst version of the problem, waiting through verification just to be told to start over.

Do next (light ops or backend, still segment-targeted):

4. Generic-versus-branded explainer, shown only on first-time uploads.
5. A one-tap confirm-and-reassure step before payment. "Approved, matches your prescription exactly" or "Substituted with a same-salt generic, here's why," right at the moment uncertainty peaks, just before money is committed.
6. An explicit "we'll call you before substituting or rejecting" promise. Cheap to state, but needs a real process behind it so it's actually true. Targets the "will this silently be the wrong thing" fear directly.

Later, higher-effort (ops-heavy, highest-touch for the hardest cases):

7. Proactive pharmacist chat prompt, triggered specifically on first-time uploads, offering to confirm dosage or substitution before the customer has to wonder and abandon.
8. Assisted WhatsApp or call upload, specifically for proxy buyers uploading a parent's or dependent's prescription, who often don't know the full details themselves.

Compounding, not a fix for this order:

9. Save a verified prescription for effortless future refills. Doesn't move this order's completion, but converts a first-timer into a repeat customer faster on their next order, shrinking the problem segment over time. Worth doing, wrong to count as this metric's fix.

What I'd explicitly deprioritise: silent auto-substitution with no visible confirmation. It's tempting, it removes a step and speeds completion, but it strips out exactly the moment that builds trust for an uncertain first-time buyer, and it's a direct risk to the pharmacist-mismatch guardrail. I'd also skip a broad checkout redesign and blanket discounts, the data says the funnel isn't broken at a visible step and repeat-refill customers already convert fine, so neither would move the number that actually leaks.

Sequence: ship the first-time-prescription tag this week, cheap instrumentation that confirms or kills the read fast. In parallel, ship the live status, the "what happens next" timeline, and the upload-quality check, all safe regardless of how the tag lands. Next, the substitution explainer and the pre-payment confirm step. Later, the pharmacist chat prompt and assisted upload once ops can support them, and the save-prescription feature as a compounding, not urgent, follow-up.

Measured in three layers:

• Leading indicators: completion rate specifically among tagged first-time-prescription orders, and time-to-verification.
• Lagging headline: overall medicine order completion rate.
• Guardrails: pharmacist-rejection and cancellation rate, CS ticket volume.

If the tagged first-time cohort's completion doesn't move, the segment read was wrong, and I'd re-cut rather than keep shipping further down this list.

Proactive pharmacist chat could overload verification staff. De-risk: make it async, not a mandatory call, and cap volume until it's proven.

The segmentation guess could be wrong. De-risk: this is exactly why the first tier of fixes is cheap and harmless to repeat-refill customers, being wrong costs almost nothing.

A "we'll call before substituting" promise that isn't operationally true erodes trust worse than saying nothing. De-risk: only ship that messaging once the process behind it is real.

So: a uniformly flat funnel with thin data ruled out a broken step, but a category cut the blended average was hiding, Rx versus OTC, and a user-type cut inside Rx, first-time versus repeat, found where the real leak lives. I'd ship cheap, segment-targeted fixes for first-time prescription uncertainty first, confirm the read with a one-field tag rather than waiting on data that doesn't exist, and protect pharmacist-safety guardrails throughout.