Beyond the Headline Numbers: Three Gaps in This Week’s Research

Written by Moe Pouyan

The Research Week · November 2 to 6, 2026

This week's research kept returning to what a single headline metric leaves out. Sales value can rise while the quantity sold, or growth after inflation, barely moves. High satisfaction scores can sit alongside rising cancellations. And plans to use AI for shopping are climbing even as research participants voice discomfort with AI models built from their own data.

The issue draws on new releases from Euromonitor, NIQ, Gallup, Pew Research Center, YouGov, Abacus Data and the Angus Reid Institute, alongside work from Circana, Bain, Prolific and Mintel, and Research Live's reporting on the insights industry.

0.1%forecast 2026 growth in global apparel and footwear sales in constant terms, against 5% in current prices and exchange ratesEuromonitor International
37%of Canadians cancelled a streaming service in the past six months, up from 32% in 2022Angus Reid Institute
34%of likely Black Friday and Cyber Monday shoppers say they are likely to use an AI tool, up from 16% in 2025YouGov
74%of Prolific participants surveyed are uncomfortable with an AI digital twin built from their dataProlific, via Research Live

Gap one: the value holds, real growth stalls

Watercolor illustration of a full shopping basket beside an empty one

Several releases showed sales value moving apart from the quantity or real value behind it, each measured differently. Euromonitor's apparel and footwear outlook forecasts global sales up 5% in 2026 in current prices and year-on-year US dollar exchange rates, but just 0.1% in constant terms. Price changes and currency movements together account for almost all of the headline gain. The most direct evidence on physical quantity came from Circana: US footwear dollar sales were flat in the third quarter while units fell by a mid-single-digit rate, with higher average prices offsetting the drop.

Other markets show a similar pattern in less detail. A Bain report built on NIQ retail data says FMCG value growth across six Southeast Asian markets has been flat from 2023 to mid-2026, with weak volumes even as prices rose; the release gives no volume or price figures. In Europe, NIQ's European retail report shows EU retail turnover growth slowing to 2.1% in 2025, from 5.5% in 2023. The page does not say whether those turnover figures are before or after inflation, so they show slowing sales growth, not necessarily falling volume.

Holiday intentions add context rather than volume data. Gallup's October survey puts planned gift spending at $1,011, on par with $1,007 a year ago, with 24% planning to spend less than last year and 18% more. Gallup forecasts holiday retail sales growth of about 3.1% and notes that inflation is running at about 3.5%. Circana's holiday study asked different questions: 57% rate the economy fair or poor, and nearly three quarters plan money-saving moves, so these percentages are not directly comparable. In the UK, NIQ reports restaurant delivery and takeaway sales up 2.3% like-for-like in September, which NIQ notes is below 3.5% CPI inflation. These are sales values from participating brands, and general CPI does not necessarily reflect restaurant price changes, so this is economic context, not a count of orders.

Why it matters. A flat or rising value line can conceal changes in units, prices or currency. Whether that hides a trade-down is something to check directly, not assume.

Gap two: satisfied subscribers, rising cancellations

Watercolor illustration of a desk where a person weighs which subscriptions to keep

The Angus Reid Institute's streaming study reports two findings that are easy to run together. At least 70% of each platform's current subscribers say they are satisfied with it. Separately, 37% of Canadians cancelled at least one service in the past six months, up from 32% in 2022, and 17% cancelled more than one, up from 8%. Among those who cancelled, 66% cited saving money or affordability, rising to 76% of 18 to 34 year olds. Meanwhile, 26% now hold four or more services, up from 17%.

Satisfaction was measured among each platform's current subscribers, while cancellation covers any service dropped in the past six months, and the published results do not link the two for the same person and service. So they cannot show that people are leaving services they are satisfied with. Among those who did cancel, cost was the most cited reason. The useful question the results raise is whether satisfaction alone is enough to predict retention.

Stated loyalty looks thin elsewhere too. In Bain's Southeast Asia consumer survey (975 respondents), half of shoppers said they would switch brands on the spot if their usual choice were unavailable, a hypothetical question. Shoppers also report trading up and down across categories: seven in ten who traded up in one category traded down in another.

Why it matters. Satisfaction can reflect price, value and service, but on its own it does not capture every reason people stay or leave. Testing whether it predicts retention would need data not provided in these published results: satisfaction measured first, then later cancellation tracked for the same customers and service.

Gap three: planning to use AI, wary of being modelled by it

Watercolor illustration of a person at a drawing desk beside a mesh-line portrait suggesting an AI digital twin

Plans to use AI for shopping are rising. YouGov found 34% of likely Black Friday and Cyber Monday shoppers say they are likely to use an AI tool to navigate the sales, up from 16% a year earlier, while 40% say they will use none. That is stated intention, collected in July and August, not behaviour. In Southeast Asia, Bain reports that about 80% of consumers say they already use AI somewhere in their shopping journey. Concern appears to be rising as well: in Abacus Data's November survey, 26% of Canadians named AI as one of the two greatest threats to Canada. Abacus reports 10% in April but cautions that the question changed between waves (rank three threats, then select two), so the comparison is directional only.

The research industry sits on both sides of this. MFour will now capture opted-in panel members' Gemini conversations and link them to the same people's browsing, store visits, purchases and survey answers. A Prolific survey of 3,046 of its participants found a conditional picture rather than a simple rejection.

These are different people in different countries answering different questions, so they do not show the same person embracing one use of AI and rejecting another. My reading, which these studies did not test, is that acceptance depends on the application and on the control people keep. A shopping assistant the user directs is a different proposition from a model of that user built by an organization.

Why it matters. Prolific's participants say review and correction rights matter. Whether such terms actually change who takes part is untested, and worth measuring before twin or conversation data becomes routine.

A note on method: definitions move the numbers

A Pew Research Center Decoded post shows how much a definition matters. Its estimate of the US Middle Eastern and North African population ranges from 3.1 million using ancestry alone to 5.2 million when birthplace, language, race and family ties are added. For anyone setting quotas or weights for a group, the definition chosen can change its size by about two thirds.

The benchmarks behind those weights are also in question. Fourteen business groups led by the Insights Association wrote to oppose a proposed rule that would drop race and ethnicity questions from the next US census. Census counts supply weighting targets and sampling frames for many US surveys. The rule is proposed, not final.

Finally, Pew's first look at 2026 polling performance found that, on preliminary vote counts, polls captured the direction of results but understated the Democratic margin by about 4 percentage points on average across 27 governor and Senate races. Pew says it is too early to identify causes. One concern it raises, greater survey participation among more engaged Democratic supporters, is a familiar methodological risk well beyond elections: weighting may not fully correct for differences in who chooses to respond.

Also worth knowing

  • Pew Research Center: 62% of US adults have at least some trust in national news organizations, up from 56% in 2025; local news rose to 75% (November 2).
  • Resolution Foundation, via Research Live: UK job satisfaction fell from 81% in 2015 to 74% in 2025, a British Social Attitudes low, and willingness to work harder than required fell from 69% to 55% (November 5).
  • NIQ: 11.8% of GB households, about 3.3 million, had a GLP-1 user by June 2026, and they spend 1.9% less on groceries than other households (November 5).
  • Abacus Data: 64% of Canadians say the country should plan for a relationship with the US that does not return to what it was, against 26% who would wait (November 2).
  • Mintel: half of US consumers already use, or are open to using, generative AI to manage their health and wellness, part of its 2027 prediction that brands will compete on outcomes rather than products; no sample size is given (November 6).

Reading these numbers together

These studies are not directly comparable. They mix retail tracking and forecasts, brand-contributed sales data, probability panels, telephone polls, online opt-in panels and a sample provider's own participants, across several regions and several definitions of growth. What they share is not one direction but a reminder that any single metric is a partial view.

For researchers and CX teams, each gap suggests a next step. On spending, report units and constant-price growth beside value where the data allow, and ask where cuts land, not just whether people plan to cut. On retention, test whether questions on affordability, value and changing needs add to satisfaction in explaining churn, measuring satisfaction first and tracking later cancellation for the same customers and service. On AI, track planned and actual use of AI assistants as separate measures, and set out consent, review and opt-out terms before using twin or conversation data, then measure whether those terms affect participation. A single number is a question, not an answer.

Sources

Moe Pouyan
Next

Three Checks: Spending Plans, AI Sign-off and Who Is Really Answering Your Survey