The backlog is bigger than the calendar
The testing program is working — seven tests read out, five clear wins, and roughly $4.8M in annualized impact traced back to them. The constraint now isn't whether the ideas are good. It's that there are far more of them than there are slots to run them in.
Unique concepts in the backlog — 8 in review or build, 20 queued across optimization and learning.
Tests read out so far. Each one consumed design time, dev time, traffic, and weeks of calendar.
Of those seven returned flat or partial results — the cost of learning something the hard way.
The gap: a concept only gets meaningfully evaluated once it's already built and live. By then you've spent the slot. Deciding what to test next is still largely a judgment call made in a room — and the ideas that lose are indistinguishable from the ideas that win until the results come in.
One step, upstream of the build
Nothing about the current process changes. A review step slots in between deciding what to test and committing engineering to it.
Ideation
Concepts come out of analytics, CJA friction, and stakeholder priorities.
Persona review
Each concept is scored by three audiences. Ship, refine, or kill — with reasons.
Target build
Only concepts that earned the slot go into development.
Live & read out
Real users decide. Results land against a documented rationale.
It pays for itself out of waste that's already there. Catch two or three weak concepts before they're built and you've recovered the cost of the layer in slots, dev hours, and calendar you didn't spend finding out the expensive way.
A read on each concept, from inside each audience
Every proposed test is evaluated through detailed, behavior-based profiles of the people OnDemand actually serves. Each persona has their own device, their own reason for being there, their own anxieties and objections — and each scores every variant on the same rubric, every time.
It isn't a general UX opinion, and it isn't a prediction. It's a specific, scored answer to one question: does this concept work for this person, and why?
Intake
The concept, the variants, the hypothesis, and the primary KPI. Missing a hypothesis or KPI pauses the review rather than guessing.
Review
Each persona works through every variant as themselves and scores it against the rubric.
Synthesis
Scores are cross-read to find friction that repeats across audiences — the highest-signal problems, ranked.
Verdict
Ship, refine, or kill — with specific, buildable changes if it needs work before it earns a slot.
Signal is consistent across audiences. The concept is ready for a Target slot as designed.
The idea holds but something specific breaks for one audience. Fix it first, then test it.
The concept doesn't do what the hypothesis assumes. The slot goes to something better.
Three OnDemand audiences, drawn from your existing framework
These come straight from the audience work already in place. Building them out into full behavioral profiles is the first piece of this — the level of detail a review needs to produce specific findings rather than generic ones.
Maya
Something is wrong and she wants an answer. Arrives from a symptom search, usually on mobile, usually anxious. Thinks in symptoms, not clinical category names.
David
Tracking his own baselines over time. Comparison-driven, returns often, wants detail and data. The repeat visitor most likely to notice a change that helps first-timers.
Priya
Proactive but new to ordering her own labs. Needs to understand what a test is, what happens next, and why to trust the process before she'll buy.
Every persona scores every variant on these six dimensions, 0–10, with written rationale. Same rubric every time, so reviews stay comparable across concepts and over time.
A durable asset, plus a review per concept
The personas themselves are the deliverable that lasts. Built to review-grade depth and signed off by your team, they belong to Labcorp — reusable on any concept, page, flow, or campaign, by anyone. They're built once. Every review after that is faster because this part is already done.
Each review then produces four layers of output:
Scorecards
Every persona against every variant across all six dimensions. Weak spots and clear winners are visible at a glance.
First-person narratives
Each persona describes the concept in their own voice — what worked, what confused them, where trust broke, the exact moment they'd leave.
Cross-audience synthesis
Where the decision gets made. Friction appearing for more than one persona is flagged as high-signal and ranked by severity.
Verdict & refinements
Ship, refine, or kill, with a signal-consistency note and a scoped list of changes to make before it earns a slot.
The tool isn't the product
These reviews are run with AI, and that's deliberate — it's the reason this can happen before a test ships rather than as a post-mortem afterward. Traditional multi-persona usability research runs into five figures and takes weeks, which is exactly why it never gets run against a test backlog.
What the AI does is make the evaluation fast and cheap enough to apply to every concept. What makes the output worth acting on is everything around it: personas built and validated with your team, every finding checked against the actual design, patterns synthesized across audiences, issues prioritized by severity, and refinements made specific enough to hand to a designer or developer.
Put plainly: the method is what keeps this fast and affordable. The judgment layer on top is what makes it useful.
What this looked like for another healthcare client
A healthcare organization ran this review on a key page and scored 4.3 — the personas surfaced friction the team hadn't seen, because each issue only mattered to one specific audience. The recommendations were implemented and the page was re-reviewed at 7.5. The engagement started as a single review and grew into a multi-phase program.
What this is, and what it isn't
What it is
- A screening layer that raises hypothesis quality and keeps weak concepts out of the queue
- A consistent rubric applied the same way every time, so concepts are comparable against each other
- A documented record of why each test earned its slot — useful when results come in later
- Fast enough to run on every concept without slowing the program down
What it isn't
- A conversion forecast — it evaluates concepts, it doesn't predict lift
- A replacement for live testing; real users still decide, this makes the tests worth running
- A substitute for analytics, session replay, or primary user research
- Design work — it evaluates what the team provides and says what to change
Build the personas once, then apply them anywhere
The personas come first
Everything else depends on them. Building the three OnDemand audiences out to review-grade depth and getting them approved is the one-time piece — and it's what makes every review after it fast, consistent, and comparable.
From there the natural first application is the Target backlog, where there are the most decisions to make and the most slots to protect. The same personas apply just as well to Context Hub personalization variants, PDP and PLP work, or anything else on the roadmap — without rebuilding anything.