Customer Advocacy, Reviews and NPS: How to Build a Feedback Program That Earns Word of Mouth

1. Blog
2. Demand & Channels

By Saroj Jha · October 5, 2026 · 14 min read

Customer advocacy is earned, not requested. Build a feedback loop first, treat NPS as one signal rather than the goal, and ask happy customers for honest reviews, references and stories without ever paying for praise. The five steps below show how, with one example company from start to finish.

Most founders want reviews and case studies before they have a reliable way to hear what customers think. In my view that order is backwards. Advocacy, meaning customers who speak for you in public or to a peer, grows out of a product that works and a team that visibly listens. This post covers feedback, NPS, reviews, references and case studies, and the US rules for each, including patient privacy.

Running case: Example Co. Fictional

Example Co. is a seed-stage MedTech startup with a benchtop blood analyzer for outpatient clinics. From a fingerstick sample, it reports routine tests such as HbA1c (a common blood sugar test) and a lipid panel in about 12 minutes, so the clinician can discuss the results during the visit. In the case it is FDA-cleared and CLIA-waived (a US category for simple lab tests). It sells first to independent primary-care practices with 3 to 10 clinicians. The practice owner or manager signs and buys test cartridges at about $14 each, about $2,100 a month for a typical clinic; medical assistants and nurses run the test every day. After its first quarter it has 12 clinics live, against a target of 25 signed by the end of the second. Its clinic-to-clinic referral program (see my referral post) is kept separate from reviews. Example Co. and all its numbers are illustrative.

Step 1: Build the feedback loop before you ask for advocacy

Ask every customer, on a regular rhythm, what is working and what is not. Asking is not neutral. In a field experiment at a financial services firm, measuring satisfaction changed customers' later behavior, including defection, product use and profitability, with effects that persisted a year later^[4]. The authors say their results “raise questions concerning the design, interpretation, and ethics” of applied marketing research. My reading: a survey is part of the experience, so keep it short and ask only what you will act on.

A loop has four parts: ask, sort the answers into themes, fix the biggest theme, and tell customers what changed. Most teams skip the last part, yet it is what turns a complaint into trust.

Example Co., step 1 Fictional

At quarter end, the team surveys the practice owner or manager and the medical assistants and nurses at its 12 clinics with three questions: the recommend question, “What is the main reason for your score?” and “What one thing should we change?” A founder reads every answer, tags it with a theme and picks one theme to fix next month. The survey asks about the product only and collects no patient information. Fictional.

Do this

Write your three survey questions, and book a date each quarter to read every answer and choose one fix.

Free tool

Build the three-question survey

The free Survey Studio builds, shares and analyzes a short survey like the one in Steps 1 and 2: the recommend question, the main reason and the one thing to change.

Step 2: Use NPS as a signal, not a score to chase

Frederick Reichheld introduced the Net Promoter Score (NPS) in Harvard Business Review in 2003, as the result of “two years of research in which a variety of survey questions were tested by linking the responses with actual customer behavior—purchasing patterns and referrals—and ultimately with company growth”^[5]. Customers answer “How likely are you to recommend us to a friend or colleague?” on a scale of 0 to 10. Bain & Company's Net Promoter System site defines the score as “the percentage of customers who are promoters (those who scored 9 or 10) minus the percentage who are detractors (those who scored 0 to 6)”, so it runs from −100 to +100^[6].

Peer-reviewed tests have not supported NPS's claimed edge in predicting growth. Using data from 21 firms and more than 15,500 interviews, Keiningham and colleagues found that their research “fails to replicate his assertions regarding the ‘clear superiority’ of Net Promoter compared with other measures” in the industries Reichheld cited^[1]. Morgan and Rego found that average satisfaction scores best predicted future business performance, while “metrics based on recommendation intentions (net promoters) and behavior (average number of recommendations) have little or no predictive value”^[7].

In my view, NPS earns its place at seed stage for two jobs: tracking the trend for the same group over time, and finding who to talk to next. The comment matters more than the score. With few customers, one or two answers move the number a lot, so show the response count beside it. And survey the people who pay and the people who use the product separately.

Example Co., step 2 Fictional

Ten of 12 practice leaders answer: 6 promoters, 3 passives and 1 detractor, an NPS of 50. Thirty clinic staff answer: 9 promoters, 12 passives and 9 detractors, an NPS of 0. The gap is the finding. Staff comments point to one theme: typing each result into the patient record takes too long. The team ships a one-click “copy to record” and tells every staff member it was built because they asked. Fictional.

What Zoom's filings and SEC letters show

Zoom put NPS in its IPO documents. Its March 2019 registration statement says: “In 2018, our average customer Net Promoter Score (NPS) was over 70”^[8]. Its risk factors add: “Any decrease in user satisfaction with our products or support would harm our brand, word-of-mouth referrals and ability to grow.”

The review correspondence is the useful part. Commenting on a confidential draft, SEC staff wrote: “You disclose that your average customer Net Promoter Score (NPS) was over 70 in 2018. Please revise to provide context for this statement, including a brief explanation of this measure and its scoring system”^[9]. Zoom's lawyers replied that the company had revised the disclosure “to provide a brief explanation of this measure”^[10].

The lesson: a bare NPS figure invites the question “measured how?” When you publish one, say who was asked, when, how many answered and how it was calculated. These are what the filings and letters say, not an account of how Zoom ran its surveys or chose its metrics.

Do this

Report NPS with its response count and method, separately for buyers and users, and read every comment before you look at the number.

Step 3: Ask for honest reviews, and never pay for praise

Reviews matter. A meta-analysis of 51 studies found that the sales elasticity of online word of mouth was .236 for volume and .417 for valence, meaning how positive the reviews are^[11]. Put simply, a 1% rise in review count went with about a 0.24% rise in sales, and positivity mattered more.

Replying helps too: hotels that responded to reviews saw “a 0.12-star increase in ratings and a 12% increase in review volume”^[2]. That is hotels, not medical devices, so treat it as a direction.

The rules, as orientation rather than legal advice:

• Rewards tied to sentiment are banned. The FTC's rule on consumer reviews makes it unlawful “to provide compensation or other incentives in exchange for, or conditioned expressly or by implication on, the writing or creation of consumer reviews expressing a particular sentiment, whether positive or negative”^[12]. It also covers fake reviews, undisclosed insider reviews and review suppression, and lets the FTC “seek civil penalties against knowing violators”^[13].
• Hiding bad reviews is a risk. The FTC alleged that the retailer Fashion Nova held lower-starred reviews for approval and that “from late-2015 until November 2019, Fashion Nova never approved or posted the hundreds of thousands of lower-starred, more negative reviews.” The company agreed to pay $4.2 million^[3].
• Google is stricter still. Its policy for Maps reviews prohibits businesses from offering incentives “in exchange for posting any review” and from choosing to “selectively solicit positive reviews from customers”^[14].
• Patient details never belong in a reply. If your customers are clinics, remember the HHS Office for Civil Rights: “OCR continues to receive complaints about health care providers disclosing their patients' protected health information on social media or on the internet in response to negative reviews. Simply put, this is not allowed”^[15]. If a review you receive mentions a patient, ask the reviewer to remove it rather than repost it.

Example Co., step 3 Fictional

The team asks every clinic, whatever its score, for an honest review on the one or two third-party sites that practice managers say they check when they compare products (found by asking in interviews). No reward is attached, and the request never mentions the referral offer. Every review gets a reply within a week, and no reply includes patient details. Four clinic leaders post reviews in the quarter. Fictional.

Do this

Send one review request to every customer, with no reward, and make it a rule that every review gets a reply.

Step 4: Turn advocates into references and case studies

In B2B, the advocacy that closes deals is often private: a reference call or a case study. Research on business reference value found it depends on relationship length, client size, the reference's format and how similar the client is to the prospect, and that the best references are not always the highest-value clients^[16].

Two parts of the FTC's Endorsement Guides apply to case studies. A customer's results “will likely be interpreted as representing that the endorser's experience is representative of what consumers will generally achieve”; if you cannot back that up, the ad “should clearly and conspicuously disclose the generally expected performance in the depicted circumstances”^[17]. And a connection that “might materially affect the weight or credibility of the endorsement”, such as free or discounted products, must be disclosed when the audience would not expect it^[18]. For health products, including diagnostic tests and health equipment, the FTC adds that advertisers “should not make claims through consumer testimonials or expert endorsements that would be deceptive or couldn't be substantiated if the advertiser made them directly”^[19].

Patient stories are a separate question. HIPAA treats as marketing a communication that “encourages recipients of the communication to purchase or use the product or service”^[20], and a covered entity “must obtain an authorization for any use or disclosure of protected health information for marketing”, except a face-to-face communication or a promotional gift of nominal value^[21]. HHS settled with a physical therapy practice that “posted patient testimonials, including full names and full face photographic images, to its website without obtaining valid, HIPAA-compliant authorizations”^[22]. The simplest rule: tell stories about the clinic's workflow and staff, and leave patients out.

Ask in order of effort: any customer can leave a review; a promoter who looks like your best prospects can take a reference call; a promoter with a result they can show can be a case study.

Example Co., step 4 Fictional

Of the six promoter clinics, four agree to take up to one reference call a month (the reference list from the referral post), matched to prospects by clinic size. Two agree to a case study. The Case Study Kit structures each one: the problem, what changed in the clinic's workflow, and the result in the clinic's own words, with numbers only where the clinic can show them. The case studies contain no patient names, photos, quotes or visit stories, so there is no patient authorization to obtain. Example Co. gives no reward for referrals (see the referral post), so neither case study mentions one. Fictional.

If you sell an AI product

Buyers of AI products read reviews and case studies with one question: will it work on data like mine? In my view, an AI case study should say what the customer's data and setup were, how the results were checked, and how much human review stayed in the loop. One customer's accuracy or time saved may not be what others get, so the FTC's point above about “generally expected performance” matters even more when results depend on the customer's data.

Never let AI write the praise. When the FTC announced its reviews rule in 2024, it said the rule “addresses reviews and testimonials that misrepresent that they are by someone who does not exist, such as AI-generated fake reviews, or who did not have actual experience with the business or its products or services”^[13]. In my view, if a tool helps a customer tidy their own words, they should approve the final text, and it must still describe what they actually experienced.

Feedback on an AI product needs one more step. Let users flag wrong outputs inside the product, and tell them what happens to that feedback, including whether you use it to improve your models. Closing the loop (Step 1) matters more when users are deciding whether to trust what the product tells them.

UiPath, which sells AI-powered automation, describes advocacy built on a user community. Its annual report for the fiscal year ended January 2025 lists “community evangelism” first among its “approaches to reach prospective customers”, and says: “Our annual UiPath Community Most Valuable Professional award program recognizes global UiPath Community members for their outstanding contribution, innovation, and evangelism.” It adds: “We believe that our highly engaged user community sets us apart”^[23]. This is what the filing says, not how the company decided internally or how it runs the program. The lesson I take: recognizing people who help other users can grow advocacy without paying for praise. If recognition comes with perks, the disclosure point above still applies.

Do this

List your promoters, and next to each name write the one ask that fits them best: review, reference call or case study.

Step 5: Measure the program, not just the score

Track five numbers every quarter:

1. Response rate, for buyers and users separately.
2. NPS trend with response counts, plus the top three comment themes.
3. Detractors contacted, and what you changed as a result.
4. Review volume and average rating on the platforms your buyers actually use.
5. References and case studies used in deals, and the win rate when one was used.

Example Co., step 5 Fictional

The quarterly dashboard shows practice-leader and staff NPS with response counts, the theme fixed, the review count and how many demos used a reference call. The number the team watches most is next quarter's staff score, because practice managers listen to the people who run the test every day. Fictional.

Do this

Add a “proof used” field to each deal in your CRM (review, reference call or case study) so you can see which kind of advocacy helps you win.

What changes by stage?

• Pre-seed — Talk to every customer yourself. Skip formal NPS; write down what each person says and fix things fast.
• Seed — A quarterly three-question survey for buyers and users, a review request to every customer, three to five references and one or two case studies.
• Series A–B — Survey at key moments, such as after onboarding and before renewal. Route detractors to customer success within a set time, and track how references affect win rates.

What mistakes should I avoid?

• Chasing the NPS number. It is no better a predictor than other measures; the value is in the comments.
• Publishing NPS without a method. Say who was asked, when, how many answered and how it was calculated.
• Rewarding positive reviews, or asking only happy customers. The first breaks the FTC rule; both break Google's policy.
• Hiding negative reviews. Suppressing them while implying you show them all can be deceptive.
• Putting patient information in a story or a reply. HIPAA can require a signed authorization first, and answering a review with patient details is not allowed.
• Never closing the loop. Customers who never hear back stop answering.