Why a composite case study (and not a named customer)
This page is a composite case study. It is not a named-customer story, it is not a polished single-brand reference, and it does not pretend to be either of those things. We want that statement to be the first thing a reader sees, because the alternative (presenting an invented operator as if they were a real signed customer) would damage trust faster than any short-term SEO benefit could justify.
Here is the plain version. ChatRaj launched in early 2026. As of the publication date of this analysis we do not yet have a signed reference-customer agreement in the beauty DTC vertical that lets us publish an individual brand's logo, founder photo, and dashboard screenshots. A handful of operators have offered to be referenced once we ship the logo-permission workflow later in the year. Until then we have a choice: stay silent on beauty case studies, or publish illustrative analyses drawn from the patterns we see across many operators. We chose the second path because the patterns themselves are useful information for buyers, and because composite analysis is a long-standing convention in industry writing (McKinsey, BCG, HBR all use composites) provided the framing is labelled clearly and repeatedly.
The substantive basis for this analysis is feedback from roughly a dozen and a half indie and mid-market beauty and skincare DTC brands on Shopify who deployed ChatRaj between January and May 2026, plus the aggregate analytics signals from their bots (message volume, deflection rate, Unanswered tab activity, lead capture counts, shade-match question patterns) that we see in our admin tooling. No individual operator is identified. Every metric in this page is presented as a typical range, never as a single-customer figure, and every range is calibrated to the spread we actually observe.
If you are looking for a named reference for procurement purposes, contact us through the pricing page and we will introduce you to operators willing to take a reference call directly. This page is the public, illustrative composite. The named references are private and one-to-one.
The persona: Nia and Velora Botanicals (fictional)
For this composite analysis we use a fictional persona named Nia Adeyemi and a fictional brand named Velora Botanicals. Neither is real. The persona and brand are constructed to give the analysis a concrete shape readers can hold in their heads, but the underlying numbers and patterns are drawn from the operator cohort described above. We will use the persona's name throughout the page; please read every mention of Nia or Velora as illustrative shorthand rather than as a description of a real customer.
In this illustrative setup, Nia runs Velora Botanicals from a small studio outside Toronto. Velora sells a tight catalogue of shade-matched complexion products (a serum foundation in twenty-eight shades, a tinted balm in twelve shades, a cream blush in nine shades) plus a small skincare regimen line (a hydrating serum, a barrier cream, a gentle cleanser, a vitamin C ampoule). The brand voice is warm, science-literate, and explicit about ingredient sourcing. The catalogue is around fifty SKUs total. Nia handles support, content, and most of the photography herself; a part-time assistant handles fulfilment.
The traffic shape, in this composite, is modest and stable. Velora sees roughly 9,500 unique monthly visitors. Average order value sits near 74 Canadian dollars. Conversion hovers around 1.6 percent. The store has been live for about three years and the brand has grown through TikTok GRWM content, careful press placements, and a tightly run email list.
The illustrative pain point that drives this composite case study is the volume and complexity of pre-purchase questions. Velora, in this composite, gets roughly 140 support touchpoints per week before deploying ChatRaj. Around 60 percent are pre-purchase. The bulk fall into four buckets: shade-match questions ("I am NC30 in MAC, which Velora shade is closest"), skin-concern questions ("which serum is safe for rosacea-prone skin"), ingredient questions ("is the vitamin C ampoule fragrance-free, is the balm reef-safe, is anything tested on animals"), and shipping and returns questions for the Canadian and US split. Every answer already exists somewhere on the site. Visitors are not finding it. Nia is retyping the same shade-match guidance at half past six in the morning, before she opens her batch-formulation notes for the day.
We chose this persona shape because it matches the modal indie-to-mid-market beauty DTC operator we see in our composite cohort: tight catalogue, complex shade and ingredient questions, brand voice that has to feel warm and expert at the same time, founder time being eaten by repeat questions that the site already answers in writing.
The starting picture
A typical pre-deployment Tuesday for the illustrative Nia goes like this. She wakes up to twenty-six unread support touchpoints across email, Instagram DMs, and TikTok comments. Fourteen are pre-purchase. Eight of those are shade-match questions where a buyer has named the shade they wear in another brand and is asking what to order from Velora. Three are skin-concern questions where a buyer with rosacea, melasma, or barrier-compromise wants to know which product to start with. Two are ingredient questions where a buyer has flagged a sensitivity (fragrance, niacinamide tingling, salicylic acid). One is a shipping question from a US buyer asking about cross-border duty.
Nia answers each of these by hand. The shade-match answers reference the existing on-site shade guide. The skin-concern answers reference the on-site regimen guide and a couple of blog posts she wrote eighteen months ago. The ingredient answers reference the product description and the INCI listing on the product page. The shipping answer references the shipping policy page. In every case the answer is already published. The visitor either did not find it or did not trust that they had found the right answer for their specific case.
By the time Nia reaches the four touchpoints that genuinely need her judgement (a press request, a wholesale enquiry from a Vancouver concept store, a damaged shipment, a returning customer asking about restock timing), it is already half past eleven. The morning is gone. Across a week, that pattern eats roughly eighteen hours of founder time. The second loss, the larger one, is invisible to her dashboard. For every visitor who asks a shade-match question over email or DM, there are roughly six or seven who close the tab when they cannot find the answer. They wanted to know which serum foundation matched their tone. They did not want to email a brand to ask. They are gone.
What ChatRaj was trained on
The composite deployment indexes the full product catalogue (every shade page, every regimen product page, every gift set), the shade guide, the regimen guide, the ingredient glossary, the shipping and returns policies, the about page, and the blog archive that contains the longer-form skin-concern posts. For the composite Velora that is roughly 180 indexed URLs. The crawl takes around 25 minutes.
Two indexing notes matter for beauty DTC operators specifically. First, shade guides that are images-only (a JPEG swatch chart with no alt text and no machine-readable shade map) cannot be indexed by a content-grounded bot. The composite fix is to publish the shade map as text on the same page (shade name, undertone, depth level, comparable shades in two or three reference brands) and let the image sit alongside as visual reinforcement. Second, ingredient INCI listings on product pages should be plain text rather than embedded inside a PDF datasheet. Operators who ship the ingredient list as a downloadable PDF see the bot miss ingredient questions until the list is republished as on-page text.
The 4 conversation patterns that drove ROI
Across the composite cohort, four conversation patterns account for the bulk of the operational lift. We describe each pattern in turn because the pattern shapes determine whether a beauty DTC operator can expect the composite ranges to apply to their own deployment.
The first pattern is shade matching. A buyer arrives on a product page or the homepage, opens the chat, and asks "I wear shade X in brand Y, what should I order from you." The bot reads the on-site shade guide, finds the comparable Velora shade, and recommends it with a one-sentence justification ("Velora 14W is the closest match: warm undertone, medium depth, comparable to MAC NC30 and to Fenty 240"). For composite operators whose shade guide is published as on-page text rather than as an image, the bot deflects these conversations cleanly. For operators whose shade guide is image-only, the bot has to fall back to a generic answer until the text version is published.
The second pattern is skin-concern routing. A buyer arrives with a stated concern (rosacea, melasma, barrier compromise, acne-prone skin, pregnancy-safe routine) and asks which products are safe to start with. The bot reads the regimen guide and the relevant blog posts, recommends two or three products, and explicitly flags what to avoid. This pattern is the highest-stakes one for the bot's voice: the answers have to be warm, careful, non-prescriptive, and clear about the limits of remote advice ("we are not a substitute for a dermatologist for active conditions"). The composite operators who tune the system prompt for that voice in the first 48 hours see the cleanest deflection on this pattern.
The third pattern is ingredient triage. A buyer asks whether a specific product contains a specific ingredient, or whether it is suitable for a stated sensitivity. The bot reads the INCI listing on the product page and answers directly. The lift on this pattern is modest in volume but high in trust: a buyer who can confirm the product is fragrance-free in the chat is a buyer who finishes the checkout.
The fourth pattern is the "build a regimen" flow, which is the highest-ROI pattern in the composite cohort and deserves its own section below.
Shade-matching conversations and how the bot handles them
Shade matching is the single most repeated pattern in the composite cohort. We will describe how the bot handles it because the mechanics matter for buyers evaluating whether a content-grounded bot can carry their shade conversations.
When the on-site shade guide is published as on-page text (shade name, undertone, depth level, two or three reference comparables per shade), the bot can answer "I wear shade X in brand Y, what do I order from you" in a single exchange. The bot reads the shade guide, finds the closest match by undertone and depth, names it, and offers a one-sentence justification. For the composite Velora with twenty-eight foundation shades and twelve balm shades, this pattern accounts for roughly 30 to 40 percent of pre-purchase deflection.
When the visitor adds a complicating detail (oily skin, dry skin, prefers more coverage, prefers dewy finish) the bot reads the product description and the regimen guide and offers a refined recommendation. The bot does not invent matches the shade guide does not support; it falls back to "the closest matches are A and B, and the regimen guide suggests A for oilier skin." This is the right behaviour and operators in the composite cohort consistently report that they prefer the bot to acknowledge uncertainty than to overcommit.
The failure mode is the visitor who asks a shade-match question for a reference brand the on-site shade guide does not name. The composite fix is to publish the shade map in a structured form that lists undertone (cool, neutral, warm) and depth (1 to 10) for each Velora shade, so the bot can match on undertone and depth even when the visitor names a brand the guide does not cover directly. Operators who add the structured shade map within the first two weeks see shade-match deflection rates climb meaningfully.
Lead capture via "build a regimen" flow
The "build a regimen" flow is the highest-ROI pattern in the composite cohort. Here is how it works. A visitor on the homepage or a category page asks something like "I have combination skin and dullness, where do I start." The bot opens a short structured conversation: skin type, primary concern, fragrance sensitivity, budget. After three or four exchanges the bot offers a personalised regimen recommendation (cleanser, serum, moisturiser, optional ampoule) and offers to email the recommendation along with a small first-order discount in exchange for the visitor's email.
Composite operators who implement this flow report a 2 to 4 times lift in captured emails on product and category pages relative to the prior passive pop-up. The captured emails flow into Klaviyo via the standard webhook and feed both the welcome series and a regimen-specific abandoned-cart sequence ("you started building a regimen, here is the saved recommendation"). The lift is highest when the discount is small (5 to 10 percent rather than 20 percent) and framed as a thank-you rather than as a bribe, which the cohort suggests is because the buyers self-select as already interested.
Two caveats matter. First, the regimen-recommendation copy has to be tuned for the brand voice or it reads as templated. The composite operators who spent an hour rewriting the recommendation phrasing in the first week report materially higher conversion on the captured emails. Second, the regimen flow has to be bounded: the bot has to refuse to give clinical recommendations for active conditions and has to point the visitor to a dermatologist for anything that crosses that line. The composite operators who set this limit explicitly in the system prompt do not see complaints; the one operator in the cohort who did not set it heard from a buyer who felt the bot had overpromised on a melasma routine.
Numbers: composite ranges seen across multiple DTC brands of this shape
This section is where the composite framing matters most. We will not give you a single set of "here is what Velora Botanicals achieved" numbers because Velora is fictional. Instead, every metric below is a typical range drawn from the composite cohort.
Pre-purchase support touchpoints, across operators in the composite sample, typically fall 40 to 55 percent in the first 60 days. The reduction is concentrated on shade-match, ingredient, and shipping questions. Skin-concern questions reduce too but less steeply, because the buyers who ask them often want a human follow-up regardless of how good the bot's first answer is.
Lead capture on product and category pages, across the composite cohort, typically increases 2 to 4 times relative to the prior passive email pop-up. The "build a regimen" flow described above is the dominant driver of the higher end of that range. Operators who deploy only the default email-capture prompt land closer to a 1.5 to 2 times lift.
Founder time recovered, in composite operator self-reports, typically lands in the 7 to 13 hours per week band. The range is wider than the others because it depends heavily on the prior baseline.
Conversion-rate lift, in the analytics signals we can see, is typically in the 0.2 to 0.5 percentage-point range. As with the cookware composite published alongside this page, conversion lift is the noisiest of these metrics because it is sensitive to seasonality, product mix, paid-acquisition shifts, and TikTok virality cycles, so we calibrate this range cautiously.
Average response time for shade-match questions falls from a typical four to ten hours (depending on when the founder reaches the inbox) to under five seconds for questions the bot can answer from the indexed shade guide.
In every case the right reading is "this is the typical operator outcome in our composite cohort." Not every operator hits every range. A handful outperform the upper end; a handful underperform the lower. The ranges are the honest middle.
What did NOT work
Three failure patterns recur often enough in the composite cohort that they belong in any honest analysis.
The first failure is the image-only shade guide. As described above, a JPEG swatch chart with no alt text and no machine-readable shade map cannot be indexed by a content-grounded bot. Operators who deployed without republishing the shade guide as on-page text saw shade-match deflection stall in week one. The fix is content work, not bot tuning.
The second failure is the over-clinical tone in the first version of the skin-concern flow. Default chatbot voices lean polite and slightly verbose. For a beauty brand whose voice is warm and science-literate, the default reads as flat and a little corporate. Roughly half of the composite cohort needed to rewrite the system prompt in the brand's voice before the skin-concern answers felt right.
The third failure is the over-confident ingredient answer. One illustrative example from the cohort: a buyer asked whether a product was fragrance-free, and the bot answered yes because the product description used the phrase "no synthetic fragrance," but the INCI listing included a botanical essential oil that some sensitised buyers react to. The fix here is two-step: the operator added an explicit "may contain naturally fragrant botanicals" line to the product description, and we tuned retrieval to favour INCI listings over marketing copy when the question is ingredient-specific. Treat these as content gaps, not bot failures.
The honest limits of this case study
The applicability question, told straight. The composite outcomes above generalise reliably to beauty DTC operators whose stores match four traits: a tight to mid-sized catalogue (roughly 20 to 200 SKUs), an existing on-page shade guide and ingredient listing (or the willingness to publish one in the first two weeks), pre-purchase support volume meaningfully driven by shade, ingredient, and skin-concern questions, and a founder or small team whose time is the bottleneck rather than capital.
The outcomes do not generalise reliably to operators in three situations. If your store gets under 1,000 unique visitors per month, there is not enough surface area for the bot to deflect meaningful volume; spend on traffic acquisition first. If your brand sells one or two hero SKUs at high volume rather than a catalogue, the composite ranges overstate the lift you will see, because most of the deflection comes from catalogue navigation. If your inbox is dominated by post-purchase issues (damaged shipments, refund disputes, custom shade requests for prestige tiers), a content-grounded bot trained on public site content cannot help with those tickets because it has no access to order data.
We say all of this to repeat the framing one more time: this is a composite, illustrative analysis. The persona Nia Adeyemi and the brand Velora Botanicals are fictional. The metric ranges are typical bands across the composite cohort, not promises for your store. Use the ranges as calibration, not as a guarantee, and treat the first two weeks of the Unanswered tab as the highest-leverage editorial work you can do.