What Headwater solves, what it rests on, and what could break it
Headwater by TP Infinity. The companies in this demo are invented and every number on screen is synthetic. The problem it points at is real, and so are the assumptions a working version would depend on. Most of them hold today or can be tested in a short pilot. A few may not hold, and one decides whether the idea works at all.
1The problem it solves
A contact-centre outsourcer in Singapore has 60 accounts it would like to win. Its sellers usually learn about a deal when an RFP arrives, and by then the buyer has done most of the work. Green Hat and 6sense surveyed 632 B2B business and procurement professionals across Asia-Pacific in June and July 2025. They found that 60% of the buying journey was over by the time the buyer engaged a vendor. In 76% of cases the winning vendor was the one the buyer engaged first, and 95% of the time the chosen vendor had been on the shortlist from the start [1]. A Gartner survey of 646 B2B buyers, published March 9, 2026, found that 67% prefer a rep-free experience [2]. The decision takes shape upstream, before the seller is invited.
Selling B2B2C widens the blind spot. You sell to a company, but the pain shows up first among that company’s customers: the hotline nobody picks up, the refund still missing after three weeks. Those customers complain in public and in their own language, in app reviews, on TikTok and in Facebook groups. Your CRM records none of it.
Our reading: two things make it worse. If the head office is in Singapore and the complaints are in Indonesian, the account’s own leadership may hear about them late. And when the work will run on a third-party platform, the platform vendor can be in the conversation before you are; what it knows reaches you only if the partnership is close.
Headwater’s job is to put those readings in front of your sellers while the buyer is still deciding, with enough evidence to act on and a way to tell afterwards whether acting made a difference. The demo follows a contact-centre outsourcer, but the method is the same for any company that sells to businesses whose customers talk in public. A payments provider would watch an account’s customers for failed-payment complaints; a logistics software firm, for complaints about late deliveries.
2What it stands on
Buyer surveys. The figures in section 1 [1, 2]: buyers do most of the work before they call anyone, and the winner is usually on their list from the start. CEB’s research for The Challenger Customer found that on average 5.4 people are formally involved in a B2B purchase, and that a purchase becomes less likely as the group grows. Disagreement inside the group peaks about 37% of the way through the purchase, and buyers reach out to suppliers at about 57% [35].
The 95:5 rule. John Dawes of the Ehrenberg-Bass Institute argued in May 2021 that firms change major suppliers about every five years, so only about 20% of business buyers are in the market in a given year, “something like 5% in a quarter” [3]. The B2B Institute at LinkedIn, which worked with Ehrenberg-Bass on how B2B brands grow, presents the same rule [4]. For Headwater it means two things: timing matters, and most of a target list will be quiet in most weeks.
Account-based marketing. ITSMA introduced and named ABM in 2003 [5] and defines it as “treating individual accounts as markets in their own right,” practised one-to-one, one-to-few (usually five to ten accounts at a time) or one-to-many across hundreds or thousands of accounts [6]. Headwater sits between the last two: 60 accounts watched by machine, one play per account.
Intent data. Bombora pools content consumption from thousands of publishers, B2B brands and data providers in a consent-based co-op, and compares each account’s last three weeks of reading on a topic with its own 12-week baseline [7]. 6sense combines visits to the customer’s own site, CRM history and third-party intent (including Bombora, G2 and TrustRadius) into four predicted stages: Target, Awareness, Consideration, Decision [8]. Demandbase lets each customer write its own stage rules, such as “high intent strength in the last 30 days” or “at least one open opportunity” [9]. Both 6sense and Bombora tie anonymous web activity to companies [8, 10]. Bombora does it through IP addresses, publisher cookies and identity graphs, and wrote in October 2021 that the move to working from home had put this to the test [10].
Trigger events. Craig Elias and Tibor Shanto’s Shift! (2010) argued that the best time to reach a buyer is after a trigger event and before they call your competition [11]. Headwater’s signals are trigger events read from data instead of picked up through a seller’s network.
Consumer posts as early evidence. Seshadri Tirunillai and Gerard Tellis collected consumer reviews from several websites over four years, for 15 firms in six markets [34]. The number of reviews led abnormal stock returns by a few days. Negative reviews had a significant negative effect, and positive reviews had none. That supports one part of Headwater’s premise: what consumers post in public carries information about a company before it shows up elsewhere, and volume and complaints count for more than praise. It says nothing about whether a company will buy services, and the lead it found is a few days. Section 5 sets out how we would test the rest.
Statistical process control. Walter Shewhart proposed the control chart in a memo dated May 16, 1924 [12]. A chart has a centre line and limits, and a point outside the limits is a reason to look for a cause [13]. Headwater, and the demo’s charts, use Donald Wheeler’s version for one count per week [33]. The limits sit at the mean plus or minus 2.66 times the average change from one week to the next. A standard deviation taken over all the weeks would be inflated by the very shift being looked for; the average week-to-week change barely moves. The limits stay fixed until a change is confirmed, so a lasting rise cannot quietly become the new normal. Three rules raise a signal: one week outside the limits (a Threshold signal), three of four weeks nearer a limit than the mean, or eight weeks in a row on one side of the mean (both Pattern signals). Five or six weeks of counts are enough to start. Wheeler also treats comparisons with last week, last year or a target as weak evidence, because one noisy number can swing them [33]. Before raising a signal, Headwater also reads two similar accounts measured the same way. If they rose in the same week, the cause is probably shared, such as a holiday, a sale or a platform outage, and the signal is held. That check is our own addition.
Controlled comparison. Google’s ad platform measures lift by comparing people who saw an ad with a control group who did not [14]. Headwater uses time as the control: each qualifying moment is drawn at random to act now or to wait four weeks, and the waiting moments are the comparison [38]. No account is left alone for good; some wait a month.
The signal library. The six mechanisms (Threshold, Pattern, Absence, Presence, Meta, Outcome), the confidence rule and the checks before raising come from the TP Infinity Lab signal library, first built for consumer brands. Its central rule carries over unchanged: the model decides whether to raise a signal, and the arithmetic (baseline, limits, spread, record count) decides how much to trust it. Headwater changes one piece: its limits follow Wheeler, as described above. It also adds a rule for when to act. A moment opens when one signal reaches confidence 0.80, or when weaker signals on one account reach that together, which is the library’s Meta mechanism. Later signals on the account join the open moment. A signal that the window has closed, such as a vendor already chosen, never opens one. The 0.80 bar is a starting setting; the pilot’s back-test should set it.
Our reading: Headwater helps with the 5%. It does nothing for the 95%, and the shortlist figure says the winner is almost always a vendor the buyer had in mind from the start. Being known before the moment is still marketing’s job. For accounts that are in the market, a signal is worth most between the 37% and 57% marks, while the buying group still disagrees and before it calls anyone.
3What already exists, and where Headwater sits
Three kinds of product already watch accounts for signs of buying. Each was built for a different job and reads different evidence. Alongside them, ABM practice decides how much attention each account gets: one-to-one, one-to-few or one-to-many [6].
- Intent and stage platforms, such as 6sense and Demandbase, with Bombora as a data source. They are built to show which accounts are researching a category and how far along they are. They combine anonymous web research matched to companies, visits to their customer’s own website and CRM history into predicted stages [8] or stage rules each customer writes [9]. The campaign reports we read show pipeline from the accounts an ad reached [31], or compare targeted accounts’ page views with their own 30 days before launch [30].
- Signal-based selling tools, such as Common Room, UserGems and Clay. They are built to tell sales whom to contact and when, from job changes, hiring, web visits, product use and funding news. Several work down to the person: Common Room resolves buying committees “down to individuals, not accounts” [15], and UserGems tracks more than 30 signals to the contact [16]. Clay draws on more than 200 data providers [17], and its pre-built signals include brand mentions and customer sentiment, the latter read from software review sites such as G2, Capterra and TrustRadius [32].
- Social listening suites, such as Brandwatch, whose archive goes back to 2008. They are built for a brand to monitor itself, its products and its competitors [18]. They read consumer posts at volume, for the brand being discussed.
Where Headwater sits. It reads a different kind of evidence: what an account’s own customers say in public, in their own language, counted against that account’s normal week. It runs on a list of target accounts and public data, so it needs no intent platform. If you already license 6sense or Demandbase, keep it: Headwater takes the account-level exports the contract allows as one more input, and writes each moment back to the CRM, which both platforms already read [8, 9]. The demo’s tenant licenses both, so their exports appear on screen as one source among five; the numbers are synthetic, and no partnership with either company is implied. It only helps with accounts whose customers talk in public; section 5 has the test.
Our reading: the pieces exist. Clay can watch brand mentions, and the suites read consumer posts at volume. In the vendor material listed below, we did not find a product that turns an account’s consumer complaints into a dated, account-level signal with its own baseline, holds that signal back when collection is incomplete or when similar accounts moved the same week, and judges each play against moments drawn at random to wait four weeks. That combination is what Headwater adds. It stays at account level on purpose, because person-level data brings the consent questions in section 6.
4Why this has not been common
The premise is partly wrong: similar tools exist, as section 3 shows.
Our reading: what was missing was the join.
- The people who read customer complaints work for the brand being complained about. The B2B seller who could treat those complaints as a buying signal never sees the feed.
- Sorting a week of posts in Indonesian, Thai, Vietnamese and Tagalog by what went wrong took native readers, or a separate model for each language.
- Proving that a play worked needs a comparison group. B2B target lists are small, and sellers will not leave a live account alone, so few teams try.
The second obstacle has shrunk. Stanford’s 2025 AI Index reported that the inference cost of a system performing at the level of GPT-3.5 dropped more than 280-fold between November 2022 and October 2024 [19]. Collecting posts went the other way. Meta shut down CrowdTangle on August 14, 2024, and only researchers could apply for its replacement [20]. TikTok’s research API excludes commercial use [21], and X now sells API access pay-per-use [22].
Our reading: the cost has moved from reading posts to collecting them lawfully. That is why the data table below matters more than the choice of model.
5Assumptions, in three tiers
The biggest assumption is that signals appear early enough before an RFP to act on. We have not tested it, and we found no published test for this market. It needs a back-test on two to three years of won and lost deals: for each deal, did the signals fire, how many weeks before the RFP, and how many false alarms came with each true one?
| Assumption | Demo step | Why we think so, or how we would find out |
|---|---|---|
| Holds todayCan be built now with known methods | ||
| An account’s weekly baseline and limits can be computed from counts. | 4The signal card | Control-chart arithmetic [13, 33]. Bombora runs a similar comparison, three weeks against twelve [7]. Five or six weeks of history give first limits [33]. |
| Hiring can be read from careers pages and job boards: role, seniority, country, first appearance. | 3Corvia Air | Postings are public and dated. Calling one the first in a year needs a year of posting history, bought or collected. Check each board’s terms. Never LinkedIn [23]. |
| Checks can hold a signal back: incomplete collection, duplicates, too few records. | 3Corvia Air | Rules on our own collection logs. The held TikTok signal (58% collected against a 90% floor) is one. |
| A rise can be checked against similar accounts before it is raised. | 4The signal card | Two similar accounts, measured the same way, sit on the same chart. If they rose the same week, the cause is shared: a holiday, a sale, a platform outage. The held Sundry Lane signal on the Signals page is one. The check is our own addition. |
| A play can be written to the CRM with an owner and a review date. | 6Start the play | Ordinary CRM integration work. |
| Needs proof firstPlausible; must be tested with real data or contracts | ||
| Accounts can be matched across the CRM, the intent platforms and public sources by company domain. | 190 days ago | Regional arms can trade under their own names and domains. Match the target list by hand in week 1 and count the misses. |
| The intent contracts allow account-level scores to be exported and combined with other data. | 3Corvia Air | Depends on each contract. Read them all before week 1. |
| The intent platforms cover Southeast Asian and Chinese accounts well enough to use. | 3Corvia Air | Count how many target accounts have twelve straight weeks of scores. |
| A general model sorts Indonesian, Thai, Vietnamese and Tagalog complaints correctly, slang included. | 4The signal card | Write a codebook. Two native readers label the same 200 posts per language without seeing each other’s work; report how often they agree. Then report precision and recall for each category, on posts not used to tune the rules [36]. When the rules change, re-label the history and rebuild the baselines, or the change itself will look like a signal. |
| Enough customers post about each account to form a baseline. | 4The signal card | Corvia’s normal is 8.4 posts a week. A freight forwarder may see one a week, and limits built on one post a week mean little. Count twelve weeks per account. |
| Enough accounts move each quarter to fill a weekly brief. | 2Today | Dawes’s arithmetic gives about 5% a quarter, three of 60 [3]. The demo shows seven to make the point. Count from the opportunity history. |
| The platform partner both sides use will make the introduction. | 5The moment | Partners are wary of sharing account lists. Bob Moore, whose company Crossbeam sells account mapping, describes a neutral service that shows each side only the accounts they have in common [39]. Ask two partners what they would pass on, and when. Try it on two accounts. |
| Sellers act on the hand-off. | 5The moment | Each moment opens with something new about the account’s own customers, the kind of content that changes a buying group’s direction [35]. Count how many hand-offs get a follow-up within five working days, and how often sellers mark a signal wrong. If sellers ignore them, stop. |
| Public posts can be collected lawfully, at a price that works for 60 accounts. | 4The signal card | Legal review in Singapore and Indonesia; quotes from licensed vendors. Indonesia’s law has no explicit exemption for publicly available data [24]. |
| May not holdCould break the idea; how we would find out | ||
| Signals appear early enough before an RFP to act on. | 5The moment | Back-test on two to three years of won and lost deals: weeks of lead time, false alarms per true signal. If the lead is under a month, or false alarms swamp the true ones, stop. |
| A complaint spike comes before the account decides to buy. | 4The signal card | The same back-test: how often a complaint spike came before an opportunity, and how often it faded or was fixed in-house. |
| The complaints are news to the account. | 5The moment | What moves a buying group is learning something new about its own business [35]. If the account’s service team already tracks these complaints, the signal is only a reason to call. Ask in the first meetings whether the pattern was known, and keep count. |
| Arriving early wins. | 5The moment | The 76% in [1] counts vendors the buyer chose to engage first. Whether a vendor that approaches early gets the same edge is untested. Compare win rates for deals we entered before and after the RFP. |
| An account under pressure will reach a decision. | 6Start the play | Two things stand in the way. The buying group may not agree (section 2). And in Matthew Dixon and Ted McKenna’s analysis of about 2.5 million sales calls, 40% to 60% of deals ended in no decision; in 44% of those the buyer preferred the status quo, and in 56% the buyer wanted to act but could not commit [37]. A complaint spike argues against the status quo. It does nothing for the fear of choosing wrong, and re-arguing the case for change with a buyer who already intends to act hurt the outcome 84% of the time [37]. So each play in the demo starts small and easy to undo, such as a four-week overflow pilot. Count how many moments end in no decision. |
| Drawing moments at random to act now or wait shows whether plays work. | 6Start the play | Only with enough of them. Telling a 10% meeting rate from 30% takes about 60 moments in each group (5% significance, 80% power); the demo’s Results page shows 13 and 12. Pool plays over several quarters, and expect sellers to resist waiting on an account that looks ready. |
6Data and sources
The minimum is a list of target accounts; everything else adds evidence. Below are four kinds of data, each with an owner and a catch. “You” is the company using Headwater. Every account, post and score in the demo is invented, and the screen says so. 6sense and Demandbase are the only real names on screen, shown as data sources.
| Data | Where it comes from | Owner or licensor | Access route | The catch |
|---|---|---|---|---|
| Your own data | ||||
| Opportunities, won and lost, 24 months or more | CRM | You | Export or API | Loss reasons may be blank. The back-test is only as good as this history. |
| Company-level visits to your website | Web analytics, matched to companies by an intent platform | You; the matching is licensed | Platform export | Matches are probabilistic, and visits from home networks are harder to place [10]. |
| Events, meetings, email engagement | Event and marketing systems, sellers’ notes | You | Export | Contact records are personal data. Singapore’s PDPA generally does not protect business contact information [25]; check each market’s law before reuse. |
| SDR call transcripts | The call platform | You | Inside your own systems only | Transcripts stay where they are. Only counts come out, such as how often a competitor is named. |
| Licensed data | ||||
| Intent scores by account and topic | Intent platforms you license, such as 6sense or Demandbase; 6sense draws on Bombora, G2 and TrustRadius [8] | The vendors, licensed to you | Platform export or API, where the contract allows | Optional: Headwater runs on a list of accounts and public data. Export and combination rights depend on your contract. Scores are modelled and noisy, so the limits sit far out. History may be short. |
| Reviews of other companies’ apps | App-data vendors | The vendor | Licence | Google Play’s own review API serves only the app’s developer, and only the past week [26]. |
| Public posts at volume | Social listening vendors | The vendor | Licence | Price per account. Coverage of TikTok and Facebook groups in local languages has to be tested. |
| Public data | ||||
| Job postings | Accounts’ careers pages; job boards | The account; the board | Collection within each site’s terms | No LinkedIn: in November 2022 a US court found that hiQ breached LinkedIn’s user agreement by scraping [23]. |
| Posts and reviews in local languages | X, TikTok, Facebook, Google Play, App Store | The people posting; the platforms | X’s pay-per-use API [22]; licensed vendors. CrowdTangle is gone [20] and TikTok’s research API excludes commercial use [21] | Posts speak only for the people who post [36]. They count as personal data when they can be linked to a person [24]. Indonesia’s law (No. 27 of 2022) reaches processing abroad that affects Indonesians [27]. Keep counts and short excerpts; drop handles after classification. Thailand, Vietnam, Malaysia and the Philippines need the same review. |
| News, annual reports, filings | Company sites, stock exchanges, news outlets | Public; full news text is licensed | Collection; a news licence | Slow: an annual report comes once a year. |
| Government tenders | Public tender portals | Public | Collection | Public-sector buyers only. |
| Partner directories; job ads that name tools | Platform vendors’ partner pages; job ads | Public | Collection | Directories show who partners with whom, rarely who uses what. Our reading: a job ad asking for experience on a named platform is the better clue. |
| Signals on mainland China accounts | Chinese job sites, tender notices, your own China site and WeChat official account (article-level totals only) | Public; yours | Process personal data inside mainland China; send out only account-level counts | PIPL: personal information leaving mainland China needs a security assessment, certification or a standard contract, plus separate consent [28]; rules issued on March 22, 2024 relaxed this for smaller volumes [29]. Legal review comes before any transfer. |
| What the demo uses | ||||
| Everything on screen | Invented for the demo | Nobody; it is synthetic | None | Sixty invented companies with made-up posts and scores, including the rows shown as 6sense and Demandbase data. The top bar and the rail say so. |
7What a real pilot would look like
Eight weeks, at account level, with a written scope and a budget.
- Weeks 1 and 2. Next year’s target list, 24 months of opportunities from the CRM, and exports from any intent platform you license; Headwater runs without one. Match accounts by company domain. Describe the opportunity history before predicting anything, as a customer-base audit does for a customer base [40]: deals a year, weeks from first contact to decision, the share that ended in no decision. Write the first 10 to 15 signal definitions, and the codebook for sorting posts.
- Weeks 3 to 6. A weekly account brief for SDRs and sellers, four issues in all. Each qualifying moment is assigned at random: its play starts now, or four weeks later [38]. Either way it reaches sales as a hand-off: an opening fact about the account’s own customers, a small first step, and two names the seller fills in.
- Weeks 7 and 8. The back-test. For each won and lost deal in the history: did the signals fire, how many weeks before the opportunity opened, and how many false alarms came with them? Then decide whether to build a tool, against a pass mark written down before week 1: for example, signals at least four weeks ahead of the opportunity, and a follow-up on more than half the hand-offs within five working days.
Judging the plays. Count at most one moment per account per month, so that no account counts twice, and analyse the results by account. Judge near-term outcomes first: a reply, a meeting within 30 days, an opportunity opened. Win rates come later. Telling 10% from 20% takes about 200 accounts on each side (5% significance, 80% power), so pool the plays across quarters. If volume allows, contact a third group without mentioning the signal; that separates the effect of the signal from the effect of calling at all. Record harm as well, such as a contact who asks not to be approached again, because outreach can backfire [38].
If eight weeks is too long, start with four: 50 to 100 accounts, their customers’ public posts and their job ads, platform exports merged by domain where they exist, the ten accounts with the most sustained movement, and one question for sales. Do you recognise these?
Our recommendation: if you license an intent platform, settle two things before week 1: whether its contract allows export and combination with other data, and how many months of account-level history each vendor can export, since the back-test depends on it. Personal data stays where it was collected; only account-level counts move.
8What comes next, and what it leaves to others
In the demo. Each moment ends in a hand-off for sales: an opening fact about the account’s own customers, a small first step that is easy to undo, a reading of whether the account is happy as things are or wants to act but cannot commit [37], and two names the seller fills in after the first call, the people likely to push for the purchase and against it [35]. The Checks page shows how well posts are sorted in each language, how complete this week’s collection is, and how signals did against events already known [36]. The Results page shows plays drawn at random to act now or wait four weeks, and how much the numbers can show so far [38]. Neither page is part of the six steps, and both use synthetic numbers.
Next, in order. First, a pilot on your own target list, as in section 7. Second, connections to what you already license: account-level exports from an intent platform, and moments written to the CRM, which those platforms already read [8, 9]. Third, a partner route: a neutral account-mapping service that shows only the accounts you and a partner have in common [39]. Likelihood estimates come last, after several hundred moments have outcomes. Until then the Results page reports measured rates.
What it leaves to others. Headwater does not build an intent network, a contact database or an advertising platform. Those rest on publisher co-ops, identity graphs and data partnerships that specialists already run [7, 8, 10]. Your sellers find people in the contact tools they already use, and an account list can go to the ad platforms you already run. Nor is there a single buying score: the map asks three questions, and the Results page counts what happened.
Our recommendation: before a pilot, list what you already license and what each contract allows. Headwater should fill the gaps and add nothing you already pay for.
·Sources
- Green Hat with 6sense. “The APAC B2B Buyer Journey Research Report 2025.” 2025; survey of 632 B2B business and procurement professionals, June and July 2025. https://19579357.fs1.hubspotusercontent-na1.net/hubfs/19579357/B2B%20Buyer%20Journey%20Research%20Hub/Green%20Hat%20APAC%20B2B%20Buyer%20Journey%20Research%20Report%202025.pdf
- Gartner. “Gartner Sales Survey Finds 67% of B2B Buyers Prefer a Rep-Free Experience.” Press release, March 9, 2026. https://www.gartner.com/en/newsroom/press-releases/2026-03-09-gartner-sales-survey-finds-67-percent-of-b2b-buyers-prefer-a-rep-free-experience
- John Dawes, Ehrenberg-Bass Institute. “Advertising effectiveness and the 95-5 rule: most B2B buyers are not in the market right now.” May 2021. https://marketingscience.info/news-and-insights/advertising-effectiveness-and-the-95-5-rule-most-b2b-buyers-are-not-in-the-market-right-now
- The B2B Institute at LinkedIn. “95-5 Rule.” LinkedIn Marketing Solutions, undated web page. https://business.linkedin.com/advertise/resources/b2b-institute/b2b-research/trends/95-5-rule
- ITSMA. “The Rise of Account-Based Marketing (timeline).” May 5, 2016. https://www.itsma.com/account-based-marketing-timeline/
- ITSMA. “Defining the Three Types of Account-Based Marketing.” February 15, 2017. https://www.itsma.com/defining-the-three-types-of-account-based-marketing/
- Bombora. “Bombora Company Surge®.” Product page, read October 6, 2026. https://bombora.com/intent/
- 6sense. “How 6sense Turns Buying Signals into Account Priorities.” June 8, 2026. https://6sense.com/guides/account-prioritization/
- Demandbase. “Getting Started with Journey Stages.” Support article, updated June 2, 2026. https://support.demandbase.com/hc/en-us/articles/360053277611-Getting-Started-with-Journey-Stages
- Bombora. “Bombora’s Company Surge® Intent Data continues to deliver despite challenges like remote work and the cookieless future.” October 28, 2021. https://bombora.com/blog/bomboras-company-surge-intent-data-still-works-despite-challenges-like-remote-work-and-the-cookieless-future/
- Craig Elias and Tibor Shanto. “Shift! Harness the Trigger Events That Turn Prospects into Customers.” iUniverse, 2010 (bookseller listing). https://www.abebooks.com/9781450240079/Shift-Harness-Trigger-Events-Turn-1450240070/plp
- ASQ. “Walter A. Shewhart.” Honorary members page. https://asq.org/about-asq/honorary-members/shewhart
- NIST/SEMATECH. “e-Handbook of Statistical Methods, 6.3.1: What are Control Charts?” Web handbook. https://www.itl.nist.gov/div898/handbook/pmc/section3/pmc31.htm
- Google. “About Conversion Lift.” Google Ads Help. https://support.google.com/google-ads/answer/12003020
- Common Room. Home page, read October 6, 2026. https://www.commonroom.io/
- UserGems. Home page, read October 6, 2026. https://www.usergems.com/
- Clay. Home page, read October 6, 2026. https://www.clay.com/
- Brandwatch. “Consumer Research.” Product page, read October 6, 2026. https://www.brandwatch.com/products/consumer-research/
- Stanford HAI. “The 2025 AI Index Report.” Released April 7, 2025; the figure quoted is from its press release. https://www.businesswire.com/news/home/20250407539812/en/Stanford-HAIs-2025-AI-Index-Reveals-Record-Growth-in-AI-Capabilities-Investment-and-Regulation
- Rappler. “Meta shuts down analytics tool CrowdTangle despite pleas from researchers.” August 14, 2024. https://www.rappler.com/technology/social-media/meta-shutdown-crowdtangle-august-14-2024/
- TikTok for Developers. “Research API.” Read October 6, 2026. https://developers.tiktok.com/products/research-api/
- X. “X API (developer documentation).” Read October 6, 2026. https://docs.x.com/x-api/introduction
- Jeffrey Neuburger, Proskauer. “Court Finds hiQ Breached LinkedIn’s Terms Prohibiting Scraping, but in Mixed Ruling, Declines to Grant Summary Judgment to Either Party as to Certain Key Issues.” New Media and Technology Law Blog, November 11, 2022. https://newmedialaw.proskauer.com/2022/11/11/court-finds-hiq-breached-linkedins-terms-prohibiting-scraping-but-in-mixed-ruling-declines-to-grant-summary-judgment-to-either-party-as-to-certain-key-issues/
- Nikita Gundala, Isentia. “Indonesia’s PDP law and social listening: what buyers need to know in 2026.” January 6, 2026. https://www.isentia.com/latest-reads/indonesias-pdp-law-and-social-listening-what-buyers-need-to-know-in-2026/
- Quahe Woo & Palmer LLC for Multilaw. “Data Protection Guide: Singapore.” 2025. https://www.multilaw.com/Multilaw/Multilaw/Data_Protection_Laws_Guide/DataProtection_Guide_Singapore.aspx
- Google for Developers. “Reply to Reviews (Google Play Developer API).” Read October 6, 2026. https://developers.google.com/android-publisher/reply-to-reviews
- Future of Privacy Forum. “Indonesia’s Personal Data Protection Bill: Overview, Key Takeaways, and Context.” October 19, 2022. https://fpf.org/blog/indonesias-personal-data-protection-bill-overview-key-takeaways-and-context/
- Office of the Privacy Commissioner for Personal Data, Hong Kong. “Mainland’s Personal Information Protection Law.” Read October 6, 2026. https://www.pcpd.org.hk/english/data_privacy_law/mainland_law/mainland_law.html
- Claire Zhao and Jason Chen, Faegre Drinker. “China Releases New Regulation on Cross-Border Data Transfers.” April 11, 2024. https://www.faegredrinker.com/en/insights/publications/2024/4/china-releases-new-regulation-on-cross-border-data-transfers
- Demandbase. “Measuring the Results of an Advertising Campaign.” Support article, updated September 18, 2026; read October 6, 2026. https://support.demandbase.com/hc/en-us/articles/4402348717069-Measuring-the-Results-of-an-Advertising-Campaign
- 6sense. “Performance Trend Report.” Support documentation, read October 6, 2026. https://support.6sense.com/docs/performance-trend-report
- Clay University. “Types of Signals in Clay.” Lesson, read October 6, 2026. https://university.clay.com/lessons/types-of-signals-in-clay-signals-abm
- Donald J. Wheeler. “Understanding Variation: The Key to Managing Chaos.” Second edition. SPC Press, 2000.
- Seshadri Tirunillai and Gerard J. Tellis. “Does Chatter Really Matter? Dynamics of User-Generated Content and Stock Performance.” Marketing Science 31(2), 2012, pp. 198-215. https://doi.org/10.1287/mksc.1110.0682
- Brent Adamson, Matthew Dixon, Pat Spenner and Nick Toman. “The Challenger Customer: Selling to the Hidden Influencer Who Can Multiply Your Results.” Portfolio, 2015.
- Justin Grimmer, Margaret E. Roberts and Brandon M. Stewart. “Text as Data: A New Framework for Machine Learning and the Social Sciences.” Princeton University Press, 2022.
- Matthew Dixon and Ted McKenna. “The JOLT Effect: How High Performers Overcome Customer Indecision.” Portfolio, 2022.
- Robert C. Blattberg, Byung-Do Kim and Scott A. Neslin. “Database Marketing: Analyzing and Managing Customers.” Springer, 2008.
- Bob Moore. “Ecosystem-Led Growth: A Blueprint for Sales and Marketing Success Using the Power of Partnerships.” Wiley, 2024.
- Peter S. Fader, Bruce G. S. Hardie and Michael Ross. “The Customer-Base Audit: The First Step on the Journey to Customer Centricity.” Wharton School Press, 2022.
Sources were opened and checked on October 6, 2026. Figures not found in a source we could open were left out.