Case study · Pilot

HealthLabs Pharm: 23% basket-value lift through in-call upsell

Pilot 01-03/2026, supplements/wellness category. How we built the upsell map, what we'd never propose, and when the AI says "not today". With transcripts.

Apr 29, 2026CallBotAgent Team~7 min readCase study
HealthLabs Pharm case study — in-call upsell for supplements
The challenge

The challenge: scaling upsell without crossing the line

HealthLabs Pharm is a supplements and cosmetics store in the Beauty & Health category. Pre-pilot, a 3-person team manually confirmed orders, and in-call upsell attempts had roughly 12% conversion — operators didn't have time for thoughtful pitches, and pitches were inconsistent across calls.

An additional problem: the supplements/wellness category has legal-strict limits. An operator cannot advise on dosing, evaluate ingredient interactions, or suggest supplementation regimens. Every upsell has to be based on what's already in the store's catalog and what fits the customer's order.

Product boundaries

What the AI proposes (and what it does NOT)

Principle: the AI is an operator of the store's script, not a health consultant. The upsell map is defined by the store in the panel — the AI only reads it.

Upsell decision boundary: allowed catalog-based proposals on the left, banned medical topics on the right

AI proposes

  • Larger pack of the same product

    1× vitamin D 1000 IU 30 caps. → 90-day pack 1× D 1000 IU 90 caps. Larger quantity, same product.

  • Higher-tier version from the catalog

    D 1000 IU → D 2000 IU, if the store has it in the catalog. The AI doesn't invent variants — it picks from the catalog.

  • Accessories from the store catalog

    Protein shaker, dropper for liquid vitamins, pill organizer. Related products defined by the store.

  • Seasonal promotional bundle

    The store defines bundles in the panel (e.g., "Vitamin C + zinc – autumn 2026") and the AI proposes them when the order matches the trigger.

AI never proposes

  • Evaluating ingredient interactions

    The AI will never say "vitamin X works well with mineral Y". That's a question for a pharmacist or doctor, not the AI.

  • Dosage recommendations

    The AI does not suggest a supplementation schedule or compare doses across products in terms of efficacy.

  • Products outside the catalog

    The AI doesn't know and doesn't invent products. Whatever it proposes must be in the store's catalog.

  • Medical questions

    When a customer asks "can I take X with my blood-pressure medication", the AI deflects: the question goes to the sales department and someone will call back.

  • Cross-sell as a "health synergy"

    A "morning energy" or "autumn-winter immunity" bundle has to be defined as a store bundle, not built ad-hoc by the AI.

Upsell map

Upsell map — what worked in the pilot

4 upsell patterns with the highest acceptance rate during 01-03/2026.

Upsell map: 4 product-pairing patterns with acceptance rates from 24% to 38%
Specific product names generalized, data anonymized. Numbers from a sample of ~600 calls.
Base productProposalAcceptance
Vitamin D 30 caps. (1× monthly)90-day pack (3× monthly)~38%
Magnesium 60 caps.Bundle with vitamin B6 (defined by the store as a "regeneration duo")~29%
Whey protein 1 kg600 ml shaker (catalog accessory)~24%
Collagen sachets, 30-packLarger version: 90-pack~33%
Results

Pilot results (01-03/2026)

Measured on a sample of ~600 calls in the supplements/wellness category. All numbers are relative to the pre-pilot manual baseline.

+23%
AOV on orders with accepted upsell
vs orders without a pitch
~31%
acceptance rate on upsell pitches
vs ~12% pre-pilot manual
+35 s
average call length with upsell
1:50 → 2:25 — the production cost of upsell
0
complaints of "I ordered something I didn't want"
the AI always confirms the extra item explicitly and waits for acceptance

* Data from the HealthLabs Pharm pilot deployment (01-03/2026) · Shopify · supplements/wellness PL e-commerce category. Sample ~600 calls. ~7% of calls with the "urgent shipping" trigger were excluded (upsell deactivated). The specific result depends on the store's upsell map, customer retention, and category seasonality.

Transcripts

Call transcripts

3 real patterns from the pilot. Customer names changed, amounts generalized.

Successful upsell (larger pack)
AI

Good day. Your order #15-2026/03 is one pack of vitamin D 1000 IU, 30 capsules, for 39 zł. Is that correct?

Customer

Yes, that's correct.

AI

We also offer a 90-day pack of the same vitamin for 99 zł — that's a saving of 18 zł versus 3 separate packs. Add it to your order?

Customer

Add it.

AI

Excellent. Order updated: 90-day pack of vitamin D 1000 IU for 99 zł. Shipping tomorrow to a parcel locker, is that everything?

Customer

Yes, thank you.

Decline, AI doesn't push
AI

We also offer a 90-day pack of the same vitamin for 99 zł — that's a saving of 18 zł. Add it to your order?

Customer

No, thanks, this is enough for now.

AI

Understood. Your order remains unchanged: 1 pack of vitamin D 1000 IU for 39 zł, shipping tomorrow to a parcel locker. Have a great day.

Medical question, AI deflects
Customer

Can I take this vitamin together with blood-pressure medication?

AI

This question requires consultation with our specialist from the sales department. They will contact you today by 4 p.m. Your base order is confirmed; the consultant will answer the dosing question. Is that alright?

Customer

Yes, thank you.

Takeaways

What we took away from the pilot

1

The upsell map MUST come from the store. The AI is not a tool for "creative" recommendations in wellness — our role is to enforce the store's map, not invent our own.

2

1 pitch per call = highest ROI. A second attempt after a decline costs time, irritates the customer, and lowers store confidence.

3

Deflection to a human = zero risk + extra conversion. Medical questions handed to the sales team not only preserve compliance — in ~40% of cases the customer comes back with an answer and buys additional items.

4

The refill pattern for returning customers has 2× higher acceptance than cross-sell. "A bigger pack of the same" wins over "a different product alongside".

● 7-day trial · 1,000 trial credits · cancel before charge

AI from your industry.From the very first call.

1,000 trial credits (~20 calls) · full access to your chosen plan · cancel anytime. Card required · no charge during trial · auto-activation on day 8.

Cancel anytimeEmail technical supportGDPR · EU serverCard required · 0 zł in trial