Roopa — pilot model dashboard

What the model says, in one line

Roopa buys engagement. Engagement buys adherence. Adherence buys two different things that pay in two different currencies: a clinical outcome for the patient, and a return visit for the clinic. Everything below is that chain, made arithmetic.

reply rate → adherence Ā → { dose D → outcome ΔW, ΔHbA1c } and { return probability p → clinic contribution ₹ }

The four gauges

Like a car: two you watch every week because they move, two you check monthly because they lag.

Inputs

1 · Engagement (what Roopa controls)

A₀ — adherence in week 1 A∞ — where adherence settles τ — decay constant, weeks

Without any support, published behaviour-change work looks like A∞ ≈ 0.15–0.25 and τ ≈ 2–3. Roopa's entire product thesis is: raise A∞, stretch τ. If it can't move these two numbers, nothing downstream happens.

2 · Dose → outcome (biology)

k_W — % body weight lost per habit-week k_H — HbA1c drop per habit-week Weeks in the programme

Anchor: a fully adherent 12 weeks should produce roughly 5 % weight loss, so k_W ≈ 0.42. You do not get to estimate these from 10 patients. Take them from the PCOS lifestyle literature, then check your patients aren't wildly off.

3 · Clinic economics (who pays)

p₀ — month-3 return rate today β — return-rate lift per unit adherence M — clinic margin per return cycle, ₹ c — Roopa cost per patient per quarter, ₹ F — fixed cost per quarter, ₹

β is the number the whole business rests on and the one you have no data for. Everything else you can look up or ask ops.

Adherence over time

The exponential decay curve. The shaded area is the dose — the thing that actually produces an outcome. Two curves: with Roopa (your sliders) and the no-support baseline.

With RoopaNo support baseline (A∞ 0.20, τ 2.5)60 % floor

The chain, computed

Patient side

Clinic side

Doctor time — the constraint that kills pilots

Time per primary doctor per patient per week. The library hit rate h is what Roopa is really selling to the doctor.

q — questions per patient per week h — answered from the approved library t_r — minutes per escalated question

T = t_fixed + q · (1 − h) · t_r

Result

Target ≤ 5 min/patient/week. Above that, doctors stop opening the digest and the whole loop dies quietly.

What you can and cannot learn from 10 patients

Estimable at n = 10

A₀, A∞, τ	yes — 12 weekly points per patient
h (library hit rate)	yes — every question is logged
q, t_r, doctor minutes	yes
c (cost per patient)	yes
p₀ (baseline return rate)	from IDEA's records, not the pilot

Not estimable at n = 10

k_W, k_H — dose→outcome	needs hundreds; borrow from literature
β — adherence→return	needs a control arm; the honest pilot output is a range
Anything about cycles regularising	too few events in 12 weeks

So the pilot's job is not to prove the outcome. It is to prove the left half of the chain — that engagement holds and doctor time stays small — and to measure p₀ so the revenue case can be argued.

How to read this honestly

Every number here except your own operating costs is currently a guess. The model's value is not the output — it is that it tells you which guess to attack first. Rank the sliders by how much the break-even moves when you nudge them: that ordering is your research agenda. Right now it is almost certainly β, then A∞, then p₀.