Customer Profiling from WhatsApp Conversations: What's Possible (and What's Ethical)
Every WhatsApp conversation a business has with a customer contains real signal — what they buy, when, how often, what they ask about before buying, what they complain about. The technical capability to turn that into a structured customer profile is not in question; it's straightforward to build. The actual question worth sitting with is narrower and more important: what should that profile be used for, and what crosses a line customers didn't agree to.
What's actually possible to build from conversational data
A WhatsApp-based ordering and support channel naturally accumulates, per customer:
- Purchase history — what they've bought, how often, and typical order value.
- Preferences stated directly — a size, a dietary requirement, a preferred delivery window, a product they asked about but didn't buy.
- Service patterns — how often they need help, what kind, and how quickly issues get resolved for them.
- Timing patterns — when they tend to order, which matters for scheduling promotions or stock around actual demand rather than guessing.
None of this requires anything invasive. It's largely just organizing what the customer has already told the business, across multiple conversations, instead of letting it evaporate after each chat ends.
Where this earns its keep: service, not just marketing
The most defensible, and often most valuable, use of a customer profile has nothing to do with targeting ads. It's continuity: a returning customer who doesn't have to re-explain their size, their delivery address, or their usual order. That's the difference between "what would you like to order?" starting from zero every time, and a system that already knows enough to offer "the usual?" A customer generally experiences this as a service improvement, not a privacy intrusion — because it's visibly using information they gave the business themselves, for a purpose they can see the benefit of immediately.
Where it gets uncomfortable: inference and surveillance
The line gets blurrier once a system starts inferring things the customer never stated — estimating income bracket from order value, flagging someone as a "high-value" or "low-value" target based on patterns they never consented to being analysed for, or using behavioural patterns to decide who sees a discount and who doesn't. This is where "customer profiling" starts to shade into something closer to surveillance, even when every individual data point involved was collected legitimately.
A useful practical test: would the customer be surprised, or feel a little betrayed, if they saw the full profile and the reasoning behind how it's used? If a customer would reasonably expect "you remembered my size" and would be uncomfortable with "we've categorised you as a low-price-sensitivity segment," that discomfort gap is the actual ethical line — not a legal technicality, a genuine signal about what the relationship can bear.
What POPIA actually requires, in practice
South Africa's Protection of Personal Information Act doesn't prohibit customer profiling. It requires it to be done with:
- A stated purpose — the business should be able to say, specifically, why it's collecting and holding a given piece of information, not just "it might be useful later."
- Consent or a lawful basis — using order history to speed up a customer's next order is reasonably implicit in the service relationship; using the same data to build a third-party marketing audience generally isn't, and needs explicit consent.
- Access and correction rights — a customer can ask what's held about them, and have it corrected or deleted.
- Proportionate retention — holding data indefinitely "just in case" is harder to justify than holding it for a defined, reasonable period tied to the actual purpose.
None of this is a reason to avoid profiling altogether. It's a reason to build it with a clear, defensible answer to "why do we have this, and what's it for" from day one, rather than retrofitting that answer after a customer asks.
A simple framework for deciding what to collect and use
Before adding a new data point to a customer profile, three questions are usually enough to sort the defensible uses from the ones that need a second look:
- Did the customer tell us this, or did we infer it? Stated preferences are lower-risk than inferred characteristics.
- Does using this make the customer's next interaction better, or just make our targeting more precise? The first is service; the second, alone, is closer to surveillance.
- Would we be comfortable telling the customer exactly how we're using this, in plain language? If the honest answer needs to be vague or technical to avoid sounding uncomfortable, that's worth revisiting before building it, not after a customer complains.
A practical rollout: how to introduce this without it feeling invasive
For a business building a customer-profiling layer for the first time, the order of operations matters as much as the features themselves:
- Start with what the customer explicitly gave you. Order history, stated preferences, and support interactions are the lowest-risk, highest-value starting point — and they're already enough to power "remember my usual order" and "notify me when this is back in stock" features that customers actively want.
- Make the value visible before asking for more. A customer who's experienced "you remembered my size" is far more receptive to a request like "can we use your purchase history to recommend other products?" than one who's never seen the profile do anything for them yet.
- Write the privacy notice in plain language, not legal boilerplate. A customer who reads "we use your order history to speed up your next order and suggest things you might like; we never sell this to third parties" understands the deal. A dense legal notice that technically discloses the same thing but reads like nobody expects it to be read doesn't build the same trust, even if it satisfies the letter of the law.
- Give an easy way to see and correct what's held. Even a simple "ask us what we have on file" process, handled by a human on request, goes a long way — it doesn't need to be a self-service dashboard on day one.
- Revisit what's being collected periodically, not just at launch. A field that seemed useful to collect a year ago but has never actually been used for anything is a liability sitting in a database for no active benefit — the kind of thing a POPIA audit, or a customer complaint, surfaces at the worst possible time.
What this looks like for a small business without a dedicated data team
None of the above requires a compliance department. In practice, for a single-location retailer or a small clinic, it comes down to three habits: only collect what's being actively used for something a customer would recognise as useful, say so plainly wherever that data is collected, and have a simple, known process for someone to ask what's held about them. A platform built for this can enforce the technical side (retention limits, access controls, consent tracking) automatically — but the judgment calls above, about what's actually fair to a customer, aren't something software can fully substitute for.
Where this applies the same way in healthcare
The same logic extends to patient profiling for care follow-up, with a higher bar: health information carries a reasonable expectation of greater confidentiality, and POPIA treats it accordingly. A patient profile used to follow up on a missed check-up or flag someone due for a routine screening is a service improvement in the same sense as a retail re-order reminder — it uses information the patient already shared, for a purpose directly tied to their care, disclosed and consented to as part of the relationship.
Questions worth asking before adding any new data point
Before a new field gets added to a customer or patient profile — a new inferred attribute, a new third-party data source, a new way of segmenting — it's worth running it through a short set of questions rather than adding it because it's technically easy:
- Who specifically benefits from this, the customer or just the business's targeting? Both can be true at once, which is fine — the problem is when only the second is true and the first is assumed.
- What's the plan if a customer asks us to delete this? If there's no clear answer, that's a sign the data point wasn't planned properly before being collected.
- Does this change what the customer sees, in a way they'd notice and potentially object to? Differential treatment based on an inferred profile — different prices, different stock shown, different offers — deserves more scrutiny than differential treatment based on something the customer explicitly chose, like a stated preference.
- Is there a real, current use for this, or a hypothetical future one? POPIA's proportionality principle leans toward collecting for an actual, near-term purpose rather than stockpiling data against a use case that may never materialise.
None of these questions have a universally right answer — they're meant to surface the trade-off explicitly, rather than letting a data point get added by default because the system happened to make it easy to collect.
Where Zentrya fits
Zentrya's platform builds customer and patient profiles from the service conversations a business already has — used for segmentation, targeted campaigns, and follow-up, within a POPIA-aligned framework rather than as an undisclosed surveillance layer. See the FMCG & Retail solution or the Healthcare AI solution for how this applies to your industry specifically.
Frequently asked questions
Is it legal to build a customer profile from WhatsApp conversations in South Africa?
Yes, when it's done in line with POPIA — which means the customer knows what's being collected, why, has given consent (often implicit in starting a service conversation, but explicit for marketing use), and can ask what's held about them or opt out. The data itself isn't the issue; undisclosed or unlimited use of it is.
What's the difference between profiling for service and profiling for surveillance?
Service profiling uses what a customer has told you to serve them better next time — remembering a size, a delivery address, or a past order so they don't have to repeat it. Surveillance-style profiling infers things the customer never told you and wouldn't expect you to know, purely to target them more precisely. The first improves the relationship; the second erodes trust even when it's technically legal.
Can this be used for healthcare patient profiling the same way?
The principle is the same — build a profile from what helps continuity of care, disclosed and consented to — but the bar is higher, since health data is treated as more sensitive under POPIA and patients have a reasonable expectation of stricter confidentiality than a retail customer does.
What should a business never do with this kind of data?
Never use it to make decisions that materially affect the customer without their knowledge (differential pricing based on inferred willingness to pay, for instance), never share it with third parties without clear consent, and never collect more than you'll actually use — a profile with no defined purpose is a liability, not an asset.
How transparent do I actually need to be about this?
Transparent enough that a customer wouldn't be surprised, or feel deceived, if they saw exactly what you'd stored about them and why. That's a reasonable practical test even before getting into the specifics of POPIA's notice and consent requirements.
See how Zentrya's Healthcare AI platform handles reminders, booking and live-agent handoff.
Explore the Healthcare AI solution