# Every — NPS survey analysis **Deliverables** - `index.html` — static, self-contained NPS dashboard (open directly in a browser; no server or network needed). KPIs, score distribution, segment mix, theme analysis, identity-framing analysis, 11 insight cards with supporting verbatims, and a filterable/searchable explorer of all 195 comments. - `ANSWER.md` — this summary of insights. - `analyze.py` → `nps_analysis.json` — reproducible analysis (NPS, theme tagging, framing stats). `build_dashboard.py` renders the HTML from the JSON. **Data:** `nps_and_descriptions_only.csv`, 195 responses, two columns: NPS score (0–10) and the open-text answer to *"How would you describe Every?"* **Assumptions made:** standard NPS cut-offs (9–10 promoter, 7–8 passive, 0–6 detractor); no date or customer metadata was supplied, so there is no trend or cohort split; themes were assigned by keyword/regex rules and then hand-checked against every comment, so counts are indicative rather than exact. Note the open-text question asks people to *describe* Every, not to justify their score — so sentiment is inferred from *how* people chose to describe it. --- ## Headline numbers | Metric | Value | |---|---| | **NPS** | **45** | | Responses | 195 | | Promoters (9–10) | 107 (54.9%) | | Passives (7–8) | 69 (35.4%) | | Detractors (0–6) | 19 (9.7%) | | Mean / median score | 8.38 / 9 | | Most common scores | 10 (69), 8 (44), 9 (38), 7 (25) | NPS of 45 is solidly good for a paid media/subscription product. The structure of the distribution matters more than the headline: the detractor base is small (10%) and the passive base is large (35%), with **8 the second most common score**. The upside is in converting 8s, not in fixing 0s. --- ## Insights from the comments ### 1. Passives know *what* Every is but not *why it matters* Passives write the shortest, flattest descriptions — median 8 words vs 11 for promoters and 10 for detractors — and they are almost all category labels: *"A modern media company"*, *"AI trends"*, *"Good content"*, *"An AI product lab"*, *"Tech/AI magazine slash software incubating studio"*. Promoters, by contrast, describe an outcome (*"The place to learn how to use AI"*) or a feeling (*"Delights me"*). The passive who comes closest to explaining their 8 says it directly: *"an eclectic forward-looking newsletter that I pay a decent amount for but somehow it's worth it."* Value is felt, but not articulated. ### 2. The writing is the product. Mentioning the apps alongside it *lowers* the score. 57% of respondents describe Every through its writing; 49% mention the apps. But framing matters: | Description mentions… | n | NPS | |---|---|---| | Writing only | 24 | **+75** | | Products only | 48 | +42 | | Writing + products | 43 | +33 | Promoters themselves say the apps are secondary: *"I am there for the writing. It simply comes out of the noise."* (10), *"Subscription would be worth it for the writing alone"* (10), *"a whole bunch of useful apps… (which I still need to fully use!)"* (10). Passives echo it with less warmth: *"Premium newsletter bundle with products I should be using"* (8), *"Newsletter and early AI products"* (8), *"cute software products"* (8), *"scrappy software products"* (7), *"interesting experimental products"* (7). The apps are currently a co-star that dilutes the story for a meaningful slice of subscribers, and ~10 respondents explicitly signal that they are not using them. ### 3. Benefit framings predict promoters; category framings predict passives | How the respondent categorises Every | n | Promoter share | NPS | |---|---|---|---| | Resource / source / **place to learn** | 36 | 72% | **+67** | | Company / team / people | 41 | 61% | +51 | | Studio / lab / incubator | 21 | 52% | +52 | | Newsletter / publication | 67 | 55% | +48 | | Products / apps / tools | 91 | 47% | +37 | | Subscription / bundle | 16 | 50% | +31 | The words *"actually"*, *"practical"*, *"how to use AI"* appear in 15 promoter comments, 5 passive comments and 0 detractor comments (*"The best resource for having any hope of actually using AI in your daily work"*, *"A place to learn about how ai is actually used"*, *"smartest people i've found talking about how to actually leverage ai"*). The positioning that converts is **"the place to learn how to actually use AI"**, not "newsletter + app bundle" — which is how the 4-scorer put it: *"A sub to a bundle of apps."* ### 4. "Signal over noise" is the emotional core — and it has zero detractors 27 respondents praise Every for filtering the AI firehose, and not one of them is a detractor (NPS +59). The language is consistent: *"without getting overwhelmed"* (×2), *"no fluff"*, *"no slop"*, *"no nonsense"*, *"without the usual hype of nonsense"*, *"comes out of the noise"*, *"Rather than trying to catch up on every piece of AI news, I find it far more efficient—and much more insightful—to learn from the things you share."* Frontier language (*"cutting edge"* ×6, *"bleeding edge"* ×3, *"frontier"*, *"front lines"*) shows the same split — 13 promoters, 4 passives, 0 detractors. Curation is the value; the frontier is merely the subject. ### 5. Identity confusion is a detractor signal 12 respondents hedged or literally couldn't answer: *"No idea honestly"* (7), *"I don't know"* (6), *"They build apps and write about it?"* (7), *"An AI content and product studio ?"* (7), *"that's a hard one"* (10), *"Interesting hybrid company"* (8), *"Kinda weird intersection of nerdy AI shit and crafty business thinking"* (5). NPS in this group: **−33**. A further third of all respondents (33%) reach for a hybrid construction — *"media co meets blog meets startup studio"*, *"magazine slash software incubating studio"* — and that group scores below average (NPS +37). The "what is Every?" question is still doing work against you. ### 6. The pivot to AI is costing pre-AI-era subscribers Six respondents describe Every as something it *used to be*, and three of the lowest scores in the survey are in that group: - *"It was... a collection of the best writers in tech & business as one bundle. Now.. I think it's 'a collection of the best on AI?'"* (3) - *"I would about 2 years ago, but not anymore. I find Every less interesting after it switched over to AI and building products"* (3) - *"It WAS a great resource for how people were using the AI for life and biz"* (7) Two promoters describe the same pivot approvingly (*"started out as a newsletter collective but has pivoted to being an AI product development house that is at the frontier"*), so this is not a verdict on the strategy — it's a churn-risk cohort that the writer-collective era acquired and the AI era isn't retaining. With no tenure data in the file, this can't be sized, but it's worth a cohort cut of churn by signup date. ### 7. The people are a differentiator — promoters describe humans, passives describe categories 38 respondents describe Every through its people (NPS +66; 28 promoters vs 7 passives). The adjectives are unusual for a media brand: *"humble (willing to be honest about failures or blindspots)"*, *"humility"*, *"non-ego oriented"*, *"abundance mindset"*, *"educators at heart"*, *"a mentor, a buddy in a multiplayer game who opened a hidden part of the map and tells you to follow them."* Dan Shipper is the only individual named. Superlatives (*"best"*, *"must-read"*, *"go-to"*, *"top 3"*, *"in the world"*) appear in 18 promoter comments vs 3 passive comments. ### 8. Mac-only is small but sharp: 3 mentions, 0 promoters, 1 cancellation Only three respondents raise platform, but every one is a non-promoter and one 0-scorer says it is why they are leaving: *"While the marketing talks about the tools I can use, I did not find out until after I subscribed that the majority are for Mac users."* A 7 wishes the suite *"ran on additional platforms besides Mac — Linux, especially"*; an 8 notes Lex was valued precisely because it was the non-Mac product, and its spin-out was felt (*"I upgraded after it branched off"*). Stating platform requirements on the pricing page is a cheap fix for an expensive surprise. ### 9. Price sentiment is polarised — and two of the three zeros are about money and trust Promoters: *"I tell them it's a steal!"*, *"Good value"*, *"worth it for the writing alone."* Detractors: *"Good content, but they will abuse the trust and relationship and fleece you for more money"* (0); *"I think you were out of touch offering a High three figure Claude class"* (0). Both zeros are about *additional* asks of existing subscribers — upsells appear to be where trust is being spent. The median view is the passive's: *"a newsletter that I pay a decent amount for but somehow it's worth it."* ### 10. Word-of-mouth is already happening, unprompted Asked only to *describe* Every, six respondents volunteered that they already recommend it — *"I have told approximately a dozen friends"*, *"Have recommended it to two other people"*, *"I've spent a lot of time talking about Every with founders"*, *"It's where I send people who want to understand…"* All six are promoters (NPS +100 in that group). A referral mechanism would harvest behaviour that already exists. Named product love is concentrated: **Monologue** (3 mentions, *"which I absolutely adore"*), Cora (2, *"just a short trial, but it was effective"*), Sparkle (1), Lex (1). **Workshops / Claude sessions / community** get 8 mentions at NPS +63 (*"The value in the workshops/learning session is almost worth it alone"*, *"really beneficial workshops"*, *"learn alongside others while you teach"*) — and even the 0-scorer rated the Claude sessions 8/10. If the apps need a front door, Monologue is it; if the bundle needs a second pillar, it's the live learning. ### 11. Detractors are engaged, specific and fixable Detractors write the longest comments (mean 19 words vs 15 promoters, 11 passives) — the unhappy minority cares enough to explain. Their complaints are concrete: - confusing product line-up and *"underwhelmed with customer service"* (4) - apps unusable *"for security reason"* (2) — likely a corporate/IT constraint - Mac-only (0, 7, 8) - too technical / too dense: *"Nerdy"* (6), *"nerdy AI shit"* (5), *"Cutting edge and insightful but not easy reading"* (7), *"non-stop claude if you're into that sort of thing"* (7) - the survey itself: *"did anyone at Every try to use this survey tool? The text box is so tiny… It seems odd that a company preaches design and aesthetics has such a poor experience when asking about user experience."* (0) There is also an audience tension worth watching: one 9 calls Every *"designed and communicated for non-engineers"* while another calls it *"a must have if you're a developer in 2025"*. Some passives and detractors are finding the current mix too technical. --- ## What I'd do with this 1. **Lead with the writing and the outcome ("learn how to actually use AI"); make the apps an earned discovery**, not the headline of the bundle. The data says the writing-only framing produces the happiest subscribers and that "bundle" framing produces the least happy. 2. **Fix the app onboarding**: ~10 respondents (including 10-scorers) admit they don't use the apps. Start with Monologue, the only app people spontaneously say they love. 3. **Put platform requirements (macOS) on the pricing page.** It's one line and it cost at least one subscriber. 4. **Audit upsells to existing subscribers** (high-ticket courses, "Claude class"). Both money-related zeros are about being asked for more *after* subscribing. 5. **Run a tenure cohort cut of churn** to size the pre-AI-era attrition the comments hint at. 6. **Add a referral path** — promoters are already recommending by the dozen. 7. **Fix the survey text box.** A detractor told you so, at length. --- ## Method notes - NPS = % promoters − % detractors = 54.9 − 9.7 = **45.1**. - Themes: regex rules in `analyze.py` (`THEMES` dict), reviewed against every comment and adjusted (e.g. case-sensitive *"WAS"* for the pivot theme; "platform" excluded from the Mac-only theme). A comment can carry several themes. - Per-theme NPS on small groups (n < 15: Mac-only, pivot, word of mouth, price, workshops, apps-underused) is directional only. - "Identity framing" and "writing vs products framing" are keyword-based: a description counts as mentioning writing if it uses newsletter/publication/magazine/blog/media/journalism etc., and products if it uses apps/tools/software/products.