The Behavioural Barriers Playbook

Round 2 of 2. Round 1 — Behavioral-Knowing-Doing-Gap — asked does the evidence support MBR’s USP? This round asks the harder question: what actually works to dissolve behavioural barriers in healthy, willing adults, mined from industries far outside behavioural finance. Separate file because Round 1 carries its own source numbering [1]–[79]; citations here are independent.

The framing that drove this round (Talbot, 2026-08-14): the target is healthy and willing — adults who already know what to do and already intend to do it. Not ignorant. Not unmotivated. Not clinical. Many players solve the knowledge/literacy barrier. The win-win-win opportunity is being the best in the world at fast, simple, easy solutions to the behavioural barriers: inertia, procrastination, activation energy, avoidance.

Research Design (as executed)

Stage	What ran
Specialists	6 parallel Opus agents: intention–action science · switching infrastructure · done-for-you agent businesses · conversion & friction engineering · addiction/treatment transfer · cross-industry wildcard
Volume	~290 tool calls, ~670k subagent tokens, ~100 sources assessed
External AI	Gemini 3.5 Flash, independent pass on the same brief
Recovery note	Specialist 6 hit a session limit mid-write; its completed deliverable was recovered intact from scratchpad. No angle lost.
Synthesis	Written in-session against MBR’s strategy SSOTs

Source discipline as Round 1: [Primary] (peer-reviewed/regulator/official full text fetched), [Secondary], [Partial] (abstract/snippet only), [Unverified]. Round 1’s replication-discount rules (Rules 0–6) apply unchanged and are not restated here. Several load-bearing numbers are Partial and are labelled inline rather than laundered.

Executive Summary

Confidence: High (89)

The UK ran MBR’s core assumption as a national experiment for twelve years, and it failed. The Current Account Switch Service (CASS) removed essentially all execution friction from bank switching: guaranteed 7-working-day switch, automatic redirection of every payment for 36 months, automatic closure of the old account, and a financial guarantee against any error. It works — 99.6% complete on time, 90% satisfaction, 11.9 million switches, 166.8 million payments redirected, 53 participating brands [1]. The FCA then fixed what it had identified as the residual barrier — awareness — reaching the 75% target on a campaign reaching 98% of the population [1]. Result: 996,344 switches in the year to 30 June 2025, against roughly 1.2 million in year one [1][2]. The FCA called it in 2015: after the first-year bump, total annual switching sat ~2% above the peak reached under the old, slow, 18-day process [2]. Their conclusion, verbatim: “simplifying and speeding up switching in isolation can only have a limited impact on switching volumes” [2].

The three frictions are not equally binding, and the industry consistently attacks the wrong one:

Friction removed	Natural experiment	Result
Execution (the paperwork)	CASS, national scale, 12 years [1][2]	+22% decaying to ≈0 net
Comparison (which option is best for me)	Kling et al. RCT, Medicare Part D [3]	17% → 28%
Decision (choosing, from a field, under uncertainty)	Ofgem collective switch, 5 replications [4]	2.6% → 26.9% (~10×)

Handel (AER 2013) closes off the information theory too: of employees whose health plan became strictly dominated — worse under every possible state of the world, requiring no search, no comparison, no judgment — 89% stayed in it for a year, 75% for two [5]. Cost of that inertia: $2,032 per employee per year [5]. They were not confused, not uninformed, not blocked. Nobody made the decision for them, so no decision happened.

This does not refute Round 1’s finding — it completes it. Round 1 found pre-filled paperwork moving switching 3%→12% [R1] and FAFSA assistance moving filing +15.7pp [6]. Both were friction removal that arrived unprompted, attached to one named recommendation. CASS is friction removal sitting there waiting for you to go get it. Removing friction from a journey nobody starts changes nothing. The product is not the smooth flow; the product is the arriving, already-decided, already-filled artifact.

The strategic reframe this produces — and it is a stronger USP than the one MBR has:

The loyalty penalty is not why people overpay. People overpay because nobody executes the renewal on their behalf — and the loyalty penalty is what the incumbent charges for that vacancy.

The FCA proved this by accident. It banned price-walking outright in January 2022; the home new-vs-renewal differential nearly halved (£95.38 → £49.17) and the motor penalty reversed entirely. Switching moved by one percentage point — and the share of policies sold with auto-renewal switched on rose from 54% to 70% [7]. A regulator deleted the entire economic reason to shop around, and the market’s answer was that more consumers went on autopilot.

The empirical floor MBR must plan against. Where people are willing, eligible, and face no knowledge barrier whatsoever, self-serve claim rates on free money cluster at 4–10% — median 9% across 149 US class actions, weighted mean 4% [8]; ~8.5% of UK dormant-asset value ever reclaimed despite a permanent free right to it [9]. And it is nearly insensitive to the amount: claims rates for sums under $10 were only 1 percentage point below sums over $200 [8]. It is not the difficulty either — forms demanding the hardest documentation ran 27.3pp lower but were only 5% of the sample, leading the FTC to conclude burden was “unlikely to be the primary factor” [8]. What kills conversion is the existence of a user-owned action step, nearly regardless of its size. When money arrives with no claim required, 55% cash it — 81%/83%/75% in the $50–100 / $100–200 / >$200 bands [8].

Recommendation: MBR should stop selling easier switching and start selling the decision already made and the review that happens by default. Six mechanisms carry the evidence, all shippable without custody: (1) assisted completion ending in submission, never in homework; (2) an opt-out annual re-shop booked to the renewal date; (3) exactly one named recommendation, not a market; (4) plan capture (day, time, what you’re doing beforehand) instead of a “yes” tap; (5) a short, real deadline; (6) a second ask to non-responders. Canada supplies free the one ingredient Ofgem had to manufacture: the mortgage renewal date.

AI Source Analysis

Agreement on the diagnosis. Gemini independently identified CASS as the central paradox and reached the same conclusion — that MBR must be a “friction-slayer,” not an information product, and that “the fallacy of the information-age fintech” is the trap. Convergent from an independent evidence path.
Agreement on calendar-stapled implementation intentions as a top-priority shippable mechanism, matching specialist 1’s and specialist 6’s independent findings.
Divergence — Gemini’s residual-barrier diagnosis is speculative where the specialists’ is measured. Gemini attributes CASS’s failure to “systemic catastrophe bias,” temporal discounting, and satisficing — plausible, but it cites no measurement. The FCA’s own consumer research names the residual directly: consumers “do not even consider switching… because of the lack of a trigger,” with process ranking only second [2]. The specialists’ regulator-sourced finding is favoured, and it points at a different fix (supply a trigger) than Gemini’s (reassure about disruption).
Divergence — Gemini maps F.A.S.T.’s “Tailored” to MINDSPACE Ego/Messenger and treats Fast as covering Defaults. Same error as Round 1: MBR structurally cannot set defaults on an account. Corrected below — MBR’s available substitute is an opt-out default inside its own flow.
Not adopted: Gemini’s “micro-milestone progress tracking” priority. Progress monitoring’s meta-analytic effect (d+ = .40) presupposes a repeated behaviour [10], which MBR does not have, and app-retention data says users will not return to see it [11].

Part 1 — Who MBR Is Actually Serving

Confidence: High (90)

Talbot’s framing has a precise name in the literature, and that matters: it converts a marketing claim into a measurable segment.

Intentions translate into action approximately half the time [12]. The predictive picture flatters: intention at T1 correlates r+ = 0.53 with behaviour at T2 across 10 meta-analyses / 422 studies (Sheeran 2002, as reported in [12]). But correlation is not leverage — experiments that actually manipulated intention found a medium-to-large intention change produced only a small-to-medium behaviour change, d+ = .36 (Webb & Sheeran 2006, as reported in [12]).

The gap is asymmetric, and one segment owns it. Decomposing the intention × behaviour matrix: “it is people who intend to change their behavior but do not (‘inclined abstainers’) who are mainly responsible for the intention–behavior gap” [12]. That sentence is the empirical charter for MBR’s target customer.

What inclined abstention looks like in administrative, appointment-like, one-shot tasks — structurally the closest published analogues to moving your savings:

31% of women invited for cervical screening failed to make the appointment despite strong measured intentions (M = 4.60 on a 1–5 scale) [13]. Motivation was measured, and it was high.
70% of those who intended but did not perform breast self-examination explained it as “forgetting” [13].
8% of young healthy adults forgot to enact an intention after a delay of five seconds — rising to 24% under divided attention [12].

Forgetting is not a character flaw. It is the modal failure mode.

The three getting-started failures — and which one MBR faces

Gollwitzer & Sheeran decompose failure-to-initiate into three problems [13]:

Remembering to act. (See above.)
Seizing an opportune moment — “especially likely [to fail] when such opportunities are brief or infrequent… involve deadlines, or when multiple ways to achieve the intention are available and the person is undecided about how best to attain their goal” [12]. Every clause of that sentence describes rate-shopping.
Second thoughts at the critical moment — the problem of “overcoming initial reluctance,” arising when benefits are long-term and costs immediate [13].

⚠ The premise to stop assuming: “willing” is not observable

Round 1 established that targeting on low baseline outcomes is actively detrimental, and intermediate-baseline targeting works best [R1]. Stated intention predicts action weakly — which is why this literature exists. So “our users are willing” cannot be an unexamined premise; MBR must detect willingness from behaviour, not assume it from a signup.

This is not academic. Gollwitzer & Sheeran’s own moderator analysis says implementation intentions work “predominantly when the underlying goal intention was strong and activated,” and are “superfluous” where there are few barriers [13]. MBR’s flagship mechanism has a documented precondition it must screen for.

The counterweight MBR must confront — and it is the best-matched study in existence. Gravert (2025) ran a randomised field experiment on consumer switching in a liberalised, low-friction market (Danish retail electricity): 200,000 randomly-sampled households on administrative smart-meter data, plus a nationally-representative survey of 9,047 [14]:

	Control	Targeted savings info	Broker (switching cost removed)
Stated intention to switch	8.5%	29.5% (+21pp)	36% (+27.5pp)
Actual switching, 3 months	2.54%	+0.8pp	+1.3pp

Powered to detect 1pp, so the small effects are not an artefact. Most recent switchers said the process took under ten minutes. Denmark has 40+ suppliers, a government comparison site with one-click switching, and monthly switching still sits near 1.5% — leaving ~1.65bn DKK unclaimed annually [14].

Massively moving intention moved behaviour by roughly one twenty-fifth as much. Removing the friction entirely added half a point over merely informing. Gravert’s own recommendation is smart defaults and automatic enrolment — the two things MBR cannot do.

But the lever is buried in the same paper, and MBR can pull it:

Consumers who said they would switch today were 22.5 percentage points more likely to actually switch than those planning to switch in 3–6 months [14].
Treatments raised “switch today” intention from a ~0.1% baseline, and this acceleration explains most of the treatment effect [14].
Author’s interpretation: “the survey acts as a prompt or deadline for those who had already planned to switch” [14].
Those who reported previously having delayed were +2.4pp more likely to switch (p<0.001) [14].

The lever is not information, and not even friction removal. It is collapsing the planned execution date to now — for people who had already decided. That is the sharpest single sentence in this entire research project, and it should govern the product.

Part 2 — The Claim-Rate Evidence: the empirical floor

Confidence: High (94)

This is the purest measurement of inertia available anywhere: free money, zero knowledge barrier, willing and eligible people, one small action step.

US class actions — FTC staff report, 149 consumer class actions, data subpoenaed under FTC Act §6(b) from seven of the largest claims administrators [8]:

Condition	Outcome
Settlements requiring a claim	Median 9% file (weighted mean 4%); 77% of resulting cheques cashed
Settlements paying automatically	55% cash — 81% ($50–100), 83% ($100–200), 75% (>$200)
Effect of money at stake	<$10 vs >$200: only 1 percentage point apart. FTC calls it “surprising”
Effect of paperwork difficulty	Hardest documentation −27.3pp, but only 5% of sample → “unlikely to be the primary factor”
Effect of the artifact	Notice packet with claim form ~10% · postcard ~6% · email ~3% · postcard with a detachable form back to ~10%
Objection / opt-out	0.01% opted out; 0.0003% objected
Eligibility	86% of submitted claims approved

Read those last three rows together. People are not sceptical, not ineligible, and not objecting. They simply do not act. And a physically detachable form recovered a 3× channel penalty.

The 14-year independent replication. UK dormant assets: £2.16bn transferred into the scheme since inception; £183.3m ever reclaimed by account holders across 210,000 accounts [9] — roughly 8.5% of value, despite a permanent, free, publicised right to your own money. Corroborates the FTC’s 9% from a completely different institutional setting.

And when someone does it for you, it multiplies. Colorectal-screening RCT (n=945): six-month adherence 12% (usual care) → 33% (test materials mailed to the patient) → 38% (materials plus a navigator call); at twelve months 18% → 36% → 43% [15].

The planning statement for MBR: in willing, eligible, knowledge-unconstrained populations, self-serve conversion on free money clusters at 4–10%, is nearly insensitive to the amount, and roughly triples when a completed artifact is placed in the person’s hands instead of an instruction. Any forecast built on “we will inform them well” should assume single digits.

Part 3 — Where MBR’s Funnel Will Actually Bleed

Confidence: High (87) — the anchor dataset is a government agency publishing raw counts.

IRS Direct File 2024 pilot, published by the IRS with absolute counts at every stage [16]:

Stage	Users	Loss
Started Eligibility Checker	3,340,500	—
Completed Eligibility Checker	677,663	−79.7% (mostly screening — only 12% were eligible)
Created / signed in to an IRS account	423,450	−37.5%
Started a tax return	395,483	−6.6%
Submitted a tax return	161,042	−59.3%
Submitted an accepted return	140,803	−12.6%

The two bolded steps are not screening. Identity cost 37.5% of users who had already been told they qualified. Form completion cost 59.3% of users who were already authenticated and had already started. Combined, those two steps retain roughly 25% of users reaching them — on a free government service with maximal institutional trust.

MBR faces the same two steps with less trust and a commercial ask.

#	MBR step	Best evidence on drop-off	What to do
1	Alert received	MBR’s internal claim of “60%+ push action rate vs 10% email” has no external corroboration and is currently an unsourced assumption under a top-30 strategic rock	Measure it before building on it
2	Open & comprehend	Nearest analogue: 12% abandon when they can’t see total cost upfront [17]	Net dollar benefit in the alert itself
3	Decide to act	This is the step to stop investing in	See Part 1 — supply the decision, don’t argue for it
4	Identity / KYC	−37.5% [16]; 19% distrust sites with card data [17]; passkeys 93% vs 63% sign-in success [18]	Worst step in the funnel. Defer it past value delivery — Stage-1 no-KYC MVP is the critical path, not a simplification
5	Fill the transfer form	−59.3% [16]; US checkout averages 23.48 form elements vs Baymard’s ideal 12–14 [17]; autofill −75% abandonment [19]; >90% accept prefill when offered [20]	Second worst, most fixable. Prefill everything MBR already knows; hard cap at 14 visible elements
6	Sign	76% of sent DocuSign agreements complete <24h, 41% <15min — no control group, no denominator, no comparison to paper exists [21]	Do not assume this step is free. Instrument it
7	Submit / route	−12.6% at IRS accept/reject [16]	Design for graceful partial success — failures here are the most damaging
8	Confirmation	Direct File set a 10-minute status SLA to prevent abandonment [16]	Adopt an explicit confirmation-latency SLA

Design rule from Amazon’s own patent. US 5,960,411 (filed 12 Sep 1997, granted 28 Sep 1999, expired 12 Sep 2017 — public domain) claims: display item information; in response to a single action, send an order request plus a purchaser identifier; the server retrieves previously stored customer information and generates the order — explicitly without a shopping-cart model [22]. That is MBR’s product thesis in one sentence: pre-stored identity + single action, with data collection moved out of the moment of intent.

Part 4 — The Master Playbook

Tiers: [FR] field-replicated at scale · [MA] meta-analysis · [1RCT] single RCT · [LAB] lab/correlational · [NULL] measured null · [UNEV] no causal estimate retrieved. “Alone?” assumes no custody, no employer relationship, Phase 1 read-only open banking.

A. Decision-supply mechanisms — the highest-value group

#	Mechanism	Measured effect	Tier	Alone?
1	Exactly one named recommendation (not a market)	Ofgem “Open Market” arm without the single named tariff underperformed by 5–6pp while still beating control hugely [4]	[FR]	Yes
2	Personalised £/$ saving + one option + deadline + minimal user input (the full Ofgem recipe)	26.9% / 24% / 29.5% vs controls of 2.6% / 3.5% / 4.5% — ~10×, five replications [4]	[FR]	Yes
3	*Do the comparison for* them, not make it available to them**	Kling RCT: 28% vs 17%, n=406; identical info was already free and advertised [3]	[1RCT]	Yes
4	Collapse the planned date to “today"	"Switch today” intenders followed through +22.5pp vs 3–6 month planners; acceleration explains most of the treatment effect [14]	[FR]	Yes
5	Short, real deadline	1-week deadline beat 3-week (p=0.01) and was statistically indistinguishable from paying people [R1]	[1RCT]	Yes
6	Second ask to non-responders	Ofgem re-engagement: 14% vs 2%; few recalled the first contact, none objected [4]	[FR]	Yes
7	Trigger on a real dated event	FCA: the main barrier is “lack of a trigger to consider switching”; process ranks second [2]	[FR]	Yes — Canada supplies the renewal date free

B. Execution mechanisms — necessary, not sufficient

#	Mechanism	Measured effect	Tier	Alone?
8	Assisted completion, finishing in-session	FAFSA: dependents 39.9% → 55.6% (+15.7pp); independents 16.1% → 42.8%; enrolment 34.2%→42.3% (p=.019); Pell 29.6%→40.2% (p=.002); 8 minutes, $88/participant [6]	[1RCT, admin outcomes]	Yes
9	Pre-filled “just sign here” return form	Switching 3% → 12% (4×), n=124,000 [R1]	[FR]	Yes
10	Ship the artifact, not a link to it	Notice packet with form ~10% vs email ~3%; detachable form recovered a postcard to ~10% [8]; mailed test materials 12%→33% [15]	[FR]	Yes
11	Never end a session with homework	FAFSA dependents’ near-complete forms “were not actually filed unless applicants followed up by mailing these forms to the DOE” [6]	[1RCT]	Yes
12	Prefill from data already held	42–48% of US returns fully prefillable; $8.2bn owed to 11M (20%) of non-filers purely to filing friction [23]; >90% accept prefill when offered [20]	[Primary/FR]	Yes
13	Autofill / stored identity	−75% form abandonment, −35% fill time (correlational, vendor-disclosed) [19]	[LAB/corr]	Yes
14	Passkeys for returning users	93% vs 63% sign-in success (30pp, ≈48% relative); 8.5s vs 31.2s; −81% login support [18]	[Vendor-consortium]	Yes
15	Form-element budget ≤14	US average 23.48 vs Baymard ideal 12–14 [17]	[Vendor-research, documented method]	Yes
16	Defer account creation past value	18% of US adults abandon over forced account creation; 62% of sites bury guest checkout [17]; IRS identity step −37.5% [16]	[FR]	Yes
17	Fee reimbursement (remove a real cost, not a framed one)	Round 1: pre-approved zero-up-front-cost offers achieved 13.0–24.3% take-up [R1]	[1RCT]	Yes

C. Default and calendar mechanisms — MBR’s substitute for a real default

#	Mechanism	Measured effect	Tier	Alone?
18	Whoever owns the default owns the outcome	Same products, split only by who must act to continue: motor attrition 33–35% with auto-renewal vs 49–54% without; home 19.6–21.5% vs 32.2–38.5% [7]	[FR]	n/a — the principle
19	Opt-out enrolment inside MBR’s own flow	Opt-out vs opt-in tobacco treatment: 22% vs 16% verified quit at 1 month, and the advantage was flat across baseline desire to change (posterior 0.976) [24]	[1RCT]	Yes
20	Opt-out recurring escalation	Workplace giving: automatic annual increase sign-up 6% → 49% opt-out vs opt-in [25]	[1RCT, secondary-reported]	Yes
21	Opt-out annual re-shop booked to the renewal date	Synthesis of 18–20 — MBR’s strongest available default substitute	[FR-derived]	Yes
22	Scheduled proactive outbound re-contact	Recovery Management Checkups: treatment receipt 61% vs 33% at 12 months, AOR 3.85; d=+0.38 treatment days [26]	[1RCT]	Yes
23	Future-dated commitment (“give more tomorrow”)	+32% donations vs same-day ask [25]	[1RCT, secondary]	Yes
24	Opt-out appointment / pre-scheduled slot	Italian RCT: uptake +3.2pp (32% relative) [R1]	[1RCT]	Partly — own flow only

D. Plan-capture mechanisms

#	Mechanism	Measured effect	Tier	Alone?
25	Implementation intentions (if-then)	d = .65 goal attainment; d = .61 specifically on failures to get started; d = .77 anti-derailment; 94 studies [12][13]	[MA]	Yes
26	Plan capture: when / where / what beforehand	N=287,228: plan formation ATT +4.1pp (SE 1.7); +9.1pp (SE 2.3) single-voter households [27]	[1RCT, very large]	Yes
27	Date AND time, not date alone	Flu-shot prompt: date+time +4.2pp; date only +1.5pp, n.s. [R1]	[1RCT]	Yes
28	Literal if-then format	Looser inductions produce weaker effects [13]	[MA moderator]	Yes
29	Screen for strong, activated goal intention	Effects obtained “predominantly when the underlying goal intention was strong and activated”; “superfluous” where few barriers exist [13]	[MA moderator]	Yes
30	Failure does not backfire	Blocked if-then planners showed more frequent and “higher quality and more strenuous” subsequent attempts; fears of handicap “would seem unfounded” [13]	[MA]	n/a — reassurance

E. Messenger, timing and placement

#	Mechanism	Measured effect	Tier	Alone?
31	Messenger identity is worth ~2×	Identical content and offer: supplier-branded 26.9% vs regulator-branded 15.0%; the envelope alone moved it to 18.5% [4]	[FR]	No — this is the partnership case
32	Embed the ask in an adjacent transaction the person is already sitting through	Will-writing legacy ask: 4.9% → 10.8% (plain) → 15.4% (social-norm), >1,000/arm, against a 35%-want/7%-do gap [28][25]	[1RCT]	No — needs partners
33	Hand a pre-simplified artifact into an existing queue	ServiceOntario, N>10,000: form at reception 2.3× odds; up to +143% registrations — and plain placement beat persuasive copy [29]	[1RCT, govt eval]	No — needs a counter
34	Front-page vs back-page placement	Front page 3%→6%; identical content on the back page: no effect [R1]	[FR]	Yes
35	Fresh-start / temporal landmark timing	Gym visits +33.4% new week, +14.4% new month, +7.5% post-birthday [R1]	[LAB/archival]	Yes
36	Multi-touch, multi-channel follow-through	Vehicle recalls: mail alone tops out; dealers who call and text achieve materially higher rates (direction only, vendor source) [30]	[Trade/vendor]	Yes

F. Content and framing

#	Mechanism	Measured effect	Tier	Alone?
37	Goal-specific reminder content	Reminders naming the specific goal were ~2× more effective than those that didn’t [R1]	[FR]	Yes
38	Plain personalised feedback + written next steps — no counselling layer	Cochrane brief alcohol: −20 g/wk at 12 months (moderate GRADE); “longer counselling probably provided little additional benefit” [31]. SIPS (n=756): 20-min counselling vs a leaflet, OR 0.78 — null [32]	[MA + pragmatic RCT]	Yes — already the plan
39	Simplify the notice	EITC response 0.14 → 0.23 from simplification alone [R1]	[1RCT at scale]	Yes
40	Benefit salience inside a simplified notice	0.23 → 0.28 [R1]	[1RCT at scale]	Yes
41	Situational, not dispositional, attribution in stall messages	Weakest item here — first-lapse self-blame did not predict relapse [33]. Cost is a copy decision	[Partial]	Yes
42	Immediacy and certainty of confirmation	The transferable half of contingency management once payment is stripped out (CM d=0.42–0.46; immediacy a positive moderator) [34]	[MA]	Yes
43	In-cohort descriptive norms only	A truthful national norm says most Canadians don’t switch (12–18%) and would backfire [R1]	[FR w/ caveat]	Yes, carefully
44	Hawthorne condition (“we’re tracking follow-through”)	+2.5pp — the benign salvage from social pressure [35]	[1RCT]	Yes
45	Behavioural activation: schedule the action, don’t raise the desire — and deliver it cheaply	COBRA (n=440): BA non-inferior to CBT (PHQ-9 diff 0.1, CI −1.3 to 1.5), £262.29 cheaper per participant, delivered by graduates with 5 days of training vs accredited psychotherapists [36]	[Non-inferiority RCT]	Yes

G. Method

#	Mechanism	Measured effect	Tier	Alone?
46	Formal sludge audit (NSW/OECD 8-step method)	Codified, tool-supported, 16 orgs across 14 countries; microbehaviour journey mapping including non-interaction steps, plus per-step completion rates to locate attrition; “hidden sludge” checklist = decision points, wait times, document preparation, search time. NSW death registration ease 69% → 74% [37]	[FR — method]	Yes
47	Run a megastudy, not a campaign	Expert forecasters — professors, practitioners, laypeople — cannot predict which intervention wins [R1]	[FR]	Yes
48	Instrument completion, not consent	Every domain shows the registered-intention illusion (see Part 5)	[FR-derived]	Yes

Part 5 — The Registered-Intention Illusion

Confidence: High (88)

Every domain measured here mistakes “registered an intention” for “the outcome occurred,” and MBR is about to inherit the same error.

Domain	Intention registered	Outcome
Wills	35% want to leave a legacy	7% of wills contain one [28]
Organ donation (Wales)	38% actively registered a decision; only 6% opted out vs 10% anticipated [38]	Consent conversations remained the binding step; Bangor researchers report consent levels “no better now than when the legislation was introduced” a decade on, citing complex paperwork and misunderstanding of deemed consent [39]
FAFSA	Near-complete forms mailed to households	Not filed unless the household mailed them onward [6]
Voting	Self-prediction — literally asking someone to state their intention: ATT +2.0pp, not significant	Plan formation: ATT +4.1pp [27]
Danish electricity	Intention +21pp / +27.5pp	Behaviour +0.8pp / +1.3pp [14]

Direct implication: a user tapping “yes, find me a better rate” is the self-prediction condition — the arm that produced nothing. It is a leading indicator, not an outcome. MBR’s north-star metric must be dollars of rate differential actually realised on a live account, verified downstream, with explicit tracking of decay between said yes → started → submitted → took effect.

Wales is the cautionary case: an entire national policy optimised on the registration metric while the outcome metric refused to move [38][39].

Part 6 — The Done-For-You Business Model

Confidence: Medium-High (79)

How the survivors get paid

Model	Example	Structure	Fit for MBR
Contingency on recovered money	AirHelp: 35% of compensation, +15% only in the ~3% of cases needing litigation; no fee for eligibility checks, no fee if unsuccessful [40]	User pays from money they’d otherwise never see	Works because the counterfactual is €0
Contingency on created savings	Rocket Money 35–60% of first-year savings, success-only [41]; Billshark 40%, “no savings, no fee” [42]	Converts a recurring benefit to a one-off fee	Poor fit — MBR’s user’s counterfactual isn’t zero (they’d renew), so % -of-savings reads expensive
Paid by the receiving institution	Capitalize: free to the individual, paid by the destination firm [43]	No price decision for the user at all	Best structural fit — identical to MBR’s position
Freemium subscription	Nous: free tier + £6.99/mo, 100,000+ UK households, £534 average saving among members who save [44]	Recurring	Viable; Nous is the closest live analogue

The pattern that should worry MBR

Consumer-direct acquisition is unaffordable in this category, so successful agents become infrastructure or get absorbed. Trim → OneMain [45]. Paribus → Capital One → decommissioned [46]. Honey → PayPal [47]. Truebill → Rocket Companies. Capitalize → pivoted to an enterprise API [43]. Not one scaled profitably as a standalone consumer subscription. The two that stayed consumer-facing (AirHelp, the bill negotiators) both have contingency economics where CAC is repaid from a single successful transaction.

Combined with Round 1’s Flipper finding — the UK’s no-custody, user-paid, savings-gated switching agent that closed in September 2021 when spread dispersion collapsed [R1] — the message is: build the execution engine so it can run inside someone else’s flow from day one, even if MBR launches as a destination.

The moat is not the AI

Capitalize’s own differentiator is that it has “mapped the rollover process for all recordkeepers, covering virtually every participant and exception” [43]. Bill negotiators’ value is knowing each provider’s retention playbook. The scarce, compounding asset is procedural coverage of every counterparty’s idiosyncratic process — for MBR: every lender’s discharge, assignment and transfer-in process, and the documents each actually requires. Start recording it from transaction one.

Two disclosure disciplines to adopt before they’re forced

Honey lost ~3 million of 20 million users within two weeks of allegations that it overrode publishers’ affiliate links, and >4 million by May 2025, plus class actions and platform-policy changes [47]. The lesson isn’t “don’t take affiliate revenue” — it’s that the revenue mechanism must survive being explained in one sentence, on the same screen as the recommendation, because someone hostile will eventually explain it. Note Nous’s own disclosure that savings “may come from rewards only” [44] — exactly the trap a client-first positioning cannot afford.
DoNotPay: FTC order finalised 11 February 2025 — $193,000 in monetary relief, notice to 2021–23 subscribers, and a prohibition on advertising professional-equivalent performance “unless it has sufficient evidence to back it up” [48]. “We do it for you” is a performance claim. Hold documented success rates, defined scope and honest failure modes before the copy goes live.

The precedent that says the market is there

Mortgage broking is the done-for-you agent model at national scale. Canada: 38% of recent buyers, 48% of first-time buyers [49]. UK ~80%; Australia 77.6% of new home loans [50]. Canada sits at roughly half the penetration of comparable markets. And the demand is already partly behavioural rather than informational: while 54% cite access to the best rate, roughly a third cite help understanding options or the process, and 25% cite “assistance with paperwork” — rising to 40% for first-time buyers, up 14pp in a year [49].

Part 7 — The Kill List

Things with real evidence saying do not build:

Don’t build	Evidence
A savings calculator as the product	FAFSA information-only arm: no significant effect; equality with the assistance arm rejected at 5% [6]. Danish info arm: intention +21pp, behaviour +0.8pp [14]
A “yes, I’m interested” tap as the alert’s terminus	That is structurally the self-prediction condition: ATT +2.0pp, not significant [27]
A conversational persuasion / counselling layer	Cochrane: longer counselling adds little [31]. SIPS: 20-min counselling vs a leaflet, OR 0.78, null [32]
Motivational interviewing as core interaction design	Effect in 40% of one-session trials vs 87% of >5-encounter trials [51]; d≈0.04–0.09 vs active comparators [52]; process meta-analysis: the relational/empathy hypothesis was not supported [53]
Stages-of-change segmentation	Zero demonstrations that stage-tailoring improves outcomes anywhere in the retrieved evidence [54]
Anything requiring users to return to an app	Median 4.0% daily open, 3.3% 30-day retention across 93 real-world apps — healthy consumers, not patients [11]
Progress-tracking as a motivational metaphor	Progress monitoring’s d+ = .40 presupposes a repeated behaviour [10]; MBR’s recurs every few years
Habit / streak mechanics	No substrate — the behaviour is too infrequent for automatisation
Public pledge / social-sharing features	Announcing an identity-relevant intention can reduce action, because the person “feels they possess the identity and no longer needs to act” [12]
Confidence-building copy	Self-efficacy and perceived behavioural control do not consistently moderate intention–behaviour consistency [12]
Social-pressure disclosure	Highest ROI in the whole report — ~8pp at $1.93/vote vs ~$20 canvassing [35] — and it generated enough anger that “many recipients called the number listed on the mailer to complain” [35]. MBR has no civic mandate and operates in a trust-sensitive category. Reject; salvage only the Hawthorne variant (+2.5pp)
Contingency payments for completing a switch	CM’s entire effect rests on objective verification (urinalysis-equivalent) MBR cannot obtain without custody [34]
A human concierge sold as a completion lever	Patient navigation doubled screening initiation (OR 2.0) but was null on completing diagnostic follow-up (OR 2.1, CI 0.99–4.4); it buys speed (−9.9 days), not finishing rate [55]. In the CRC trial, mailing the instrument moved 12%→33% while adding a navigator call added only 5 more points [15]
Features contingent on a cancellation-friction mandate	FTC click-to-cancel vacated in full by the 8th Circuit, July 2025 [56]

Numbers that would not survive an investor deck

“Baymard: forced account creation costs 20–30% conversion.” Not on any Baymard page — misattributed. Use the 18% stated-reason figure [17].
“Passkeys give a 30% conversion lift.” It is 30 percentage points (93% vs 63%), ≈48% relative [18].
“Each form field costs 3–5% of conversion.” No traceable primary — agency blogs citing agency blogs.
Any DocuSign figure other than 76%/41% — and even those only as time among agreements already sent, with no denominator and no control [21].
Estonia’s “820 working years saved.” The same search returned both 820 and 1,345 — mutually contradictory, no primary.
“Direct File users were 90% satisfied.” True but survey-of-completers only, 13% response rate — the IRS says so itself [16].
Amazon 1-Click’s “$2.4bn value” / Apple licence terms. Press-repeated, absent from the patent record [22].
All SaaS activation benchmarks. Uniformly vendor marketing.
MBR’s own “60%+ push action rate vs 10% email.” Internal, unsourced, and currently load-bearing on the roadmap.

Methodological proof from this session: a search snippet reported Baymard’s abandonment reasons as “39% / 19%”; the fetched page says 40% / 18%. Small drift, same session, on the best-documented source in the field.

RANKED LIST A — MBR can ship this alone

Three-axis scoring per Round 1 (effect after replication discount × build cost × deployability with no custody).

Gate 0 remains Round 1’s spread-durability test. Flipper and Look After My Bills both died when dispersion collapsed. Model the subscription against the narrowest historical Canadian dispersion. Nothing below rescues a vanished spread.

Rank	Build	Why here	Evidence
1	Opt-out annual re-shop, booked to the renewal date at signup	The single highest-leverage thing MBR can do without custody. Whoever owns the default owns the outcome: 15pp swing on identical products [7]; opt-out beat opt-in 22% vs 16% and held across all motivation levels [24]; opt-out escalation 6% → 49% [25]; scheduled outbound re-contact AOR 3.85 [26]. MBR’s product becomes “the review happens unless you stop it”	[FR]
2	Assisted completion that ends in a submission, never in homework	FAFSA +15.7pp / 16.1%→42.8% at 8 minutes and $88 [6]; pre-filled return form 3%→12% [R1]. Exit criterion: a submitted application. Anything handed back to finish later is a failed session [6]	[1RCT/FR]
3	One named recommendation + personalised dollar figure + short real deadline + minimal input (the Ofgem recipe)	~10× control, five replications [4]. MBR currently ships a market; narrowing to one is worth 5–6pp on its own [4]	[FR]
4	Fire on the mortgage renewal date	The FCA names “lack of a trigger” as the primary barrier [2]. Ofgem had to manufacture a deadline; Canada supplies one that is dated, known in advance and individually timed — and OSFI dropped the stress test for uninsured straight switches on 21 Nov 2024 [57]	[FR]
5	Plan capture instead of a confirmation tap	Self-prediction ATT +2.0pp ns; plan formation ATT +4.1pp, +9.1pp for solo decision-makers [27]; date+time +4.2pp vs date-only n.s. [R1]. Three questions and a calendar invite — a small build	[1RCT]
6	Second ask to non-responders, on a schedule	14% vs 2% [4]; RMC AOR 3.85 [26]. A single-shot funnel discards most available conversion	[FR]
7	Attack the identity and form steps in that order	−37.5% and −59.3% on a free government service with maximal trust [16]. Stage-1 no-KYC MVP is the critical path; hard-cap forms at 14 elements [17]; prefill everything already known [23]	[Primary]
8	Instrument completion, not consent	The registered-intention illusion appears in every domain measured [6][14][27][28][38][39]. North star = realised dollars on a live account	[FR-derived]
9	Run the NSW/OECD sludge audit — on the banks, not just on MBR	Public, codified, 8 steps, 14 countries [37]. Running it on each lender’s onboarding produces a brandable public asset: MBR as the institution that measures and publishes where each provider’s sludge sits	[FR method]
10	Passkeys + autofill on the returning-user path	93% vs 63%, 8.5s vs 31.2s [18]; −75% form abandonment [19]. Authentication friction compounds across every alert	[Vendor]
11	Behavioural-activation staffing model for any human tier	BA non-inferior to CBT, £262 cheaper, delivered by 5-day-trained graduates [36]. Script-driven, low-training role — not an advisor role	[RCT]
12	Loss framing, in-cohort norms, Hawthorne tracking, situational stall copy	All near-zero cost; all currently absent from F.A.S.T. [12][35][43]	[Mixed]
13	Megastudy discipline	Forecasters cannot pick winners [R1]. Test many variants against one objective outcome with a real control arm	[FR]

RANKED LIST B — requires a partner or institution

(Chilton / Jeff pitch material.)

Rank	Unlock	Partner	Evidence
1	The messenger	A trusted distributor (Chilton)	Identical letter, identical offer: 26.9% supplier-branded vs 15.0% regulator-branded [4]. MBR is neither incumbent nor regulator — the weakest position in that experiment. This is the hard quantitative case for the Chilton relationship, and it is worth roughly 2×.
2	The adjacent-transaction moment	Mortgage brokers, tax preparers, realtors/closers, will-writers	The two largest conversions retrieved were both embedded asks by the professional already doing adjacent paperwork: legacy ask 4.9% → 15.4% [28]; FAFSA prefill from data already on screen [6]
3	Embedded distribution (the Capitalize pivot)	Lenders/institutions	Free to the user, paid by the receiving institution; Capitalize moved its centre of gravity to an API because that’s where the volume was [43]
4	Pre-simplified artifact in an existing queue	In-branch, brokerage, closing package	ServiceOntario 2.3× odds, +143%, N>10,000 — Canadian, government-run, recent, directly citable to partners [29]
5	Write access / execution	Open banking Phase 2	Phase 1 is read-only; nobody can execute a switch today [R1]
6	True defaults	Bank or employer	The field’s highest-effect intervention, permanently unavailable to MBR alone

The USP, restated

Round 1 recommended reordering the three differentiators to put execution first. Round 2 says that’s necessary but not sufficient — CASS proves execution alone moves nothing. The complete statement:

MBR’s USP
Section titled “MBR’s USP”

Every other player tells you what to do. MBR decides, schedules, and fills it in — and does it again next year without being asked.

Three claims, in evidence order:

1. The decision arrives already made. Not a market — one recommendation, with your dollar figure, and a date it expires. (Ofgem: ~10× control, five replications.)

2. The paperwork arrives already filled. Your job is one signature, in one sitting. Nothing is ever handed back to you to finish later. (FAFSA: +15.7pp; pre-filled switching form: 3% → 12%.)

3. The review happens by default. Booked to your renewal date, opt-out not opt-in, with a second ask if you don’t respond. (Auto-renewal defaults: 15pp swing. Second ask: 14% vs 2%.)

What this replaces: “Higher savings rates. Lower debt rates. Only notified when Better is worth it.” That leads with rate outcomes and an alert-suppression promise, of which only the suppression logic survived Round 1 — and the “<3 alerts/year” dose remains untested and should be dropped.

What stays out of the USP: the Anti-Pitch and the money-back guarantee. Both are genuine commitments and belong in brand/values, but neither has measured evidence as a conversion lever, and the FCA found consumers already assume comparison sites act in their interest [R1].

Next Steps

Decide the USP restatement above, then update MBR/Strategy/Unique Selling Proposition.md (deliberately untouched across both rounds) and propagate to the pitch deck and Chilton proposal.
Take the messenger finding into the Chilton conversation as a number, not a hope — 26.9% vs 15.0% on identical content [4] is the strongest quantitative argument for the partnership in either round.
Reframe the roadmap around the opt-out annual re-shop. It is rank 1 of List A, ships without custody, and converts MBR from an alert product into a default-owning one.
Build the counterparty process map from transaction one — every lender’s transfer-in/discharge process and required documents. That is the compounding moat, not the intelligence layer [43].
Kill the items on the Kill List before they consume roadmap — especially any calculator-as-product, any conversational persuasion layer, and any feature requiring return visits to an app.
Replace the internal “60%+ push vs 10% email” assumption with a measurement, and stand up the instrumented funnel (steps 1–8 in Part 3) on the first hundred users. Those users are worth more as a funnel than as revenue — no published source measures drop-off on a consumer financial switching flow, so this is a genuinely novel data asset.
Verify before external use: OSFI’s straight-switch stress-test change against OSFI primary (currently industry commentary only) [57]; the open-banking retail-volume threshold; and whether MaxMyInterest takes bank compensation (Round 1 open item).
Consider the publishable RCT. No source tests if-then planning or plan capture on a financial account switch. Running it would be novel, cheap, and a durable credibility asset.

Sources

[1] CASS Dashboard Issue 47, Q2 2025 — Pay.UK [Primary]
[2] Making current account switching easier — FCA post-implementation review, March 2015 [Primary]
[3] Comparison Friction: Experimental Evidence from Medicare Drug Plans — Kling, Mullainathan, Shafir, Vermeulen & Wrobel, QJE 127(1), 2012 [Primary]
[4] Collective Switch Trials Final Report — Ofgem, September 2019 [Primary]
[5] Adverse Selection and Inertia in Health Insurance Markets — Handel, AER 103(7), 2013 [Primary]
[6] The Role of Application Assistance and Information in College Decisions (H&R Block FAFSA Experiment) — Bettinger, Long, Oreopoulos & Sanbonmatsu, QJE 127(3), 2012 [Primary]
[7] Evaluation Paper 25/2: An evaluation of our general insurance pricing practices remedies — FCA, July 2025 [Primary]
[8] Consumers and Class Actions: A Retrospective and Analysis of Settlement Campaigns — FTC Staff Report, September 2019 [Primary]
[9] Annual Report and Accounts 2024/2025 — Reclaim Fund Ltd [Primary]
[10] Harkin, Webb, Chang, Prestwich, Conner, Kellar & Sheeran (2016), Psychological Bulletin 142(2):198–229 — 138 interventions; d+ = .40 [Secondary, as reported in [12]]
[11] Objective User Engagement With Mental Health Apps — Baumel et al., JMIR 21(9):e14567, 2019 [Primary]
[12] The Intention–Behavior Gap — Sheeran & Webb, Soc Pers Psychol Compass 10(9), 2016 [Primary]
[13] Implementation Intentions — Gollwitzer & Sheeran, NCI-hosted chapter summarising their 94-study meta-analysis [Primary — not the 2006 Adv Exp Soc Psychol article, which was not retrievable]
[14] Can Attention Overcome Consumer Inertia? Experimental Evidence from a Liberalized Market — Gravert, July 2025 [Primary — working paper, not yet peer-reviewed]
[15] A Randomized Controlled Trial of Tailored Navigation and Standard Intervention in Colorectal Cancer Screening — Myers et al., n=945 [Primary]
[16] Publication 5969: Direct File Pilot Program Filing Season 2024 After Action Report — IRS [Primary]
[17] 50 Cart Abandonment Rate Statistics and Ecommerce Checkout UX — Baymard Institute [Primary — vendor-research, documented method]
[18] Passkey Index — FIDO Alliance, October 2025 [Partial — confidential self-reported consortium survey, no control]
[19] Autofill in action: real-world insights — Chrome for Developers [Primary — vendor engineering; explicitly correlational]
[20] GAO-25-106933: Direct File — US Government Accountability Office [Primary]
[21] Docusign eSignature Datasheet [Vendor-marketing — no method, no denominator, no control]
[22] US Patent 5,960,411 — Amazon.com Inc. [Primary — patent record]
[23] Automatic Tax Filing: Simulating a Pre-Populated Form 1040 — Goodman, Lim, Sacerdote & Whitten, NBER WP 30008 [Primary]
[24] Opt-out vs opt-in tobacco treatment — JAMA Netw Open, n=739 [Primary]
[25] From Intentions to Action: The Science Behind Giving Behaviours — Rideau Hall Foundation × Behavioural Insights Team, Canada, 2018 [Primary for BIT’s own trials; Secondary for the Rasul & Huck +26% and “Give More Tomorrow” +32% figures]
[26] Recovery Management Checkups in primary care — Scott et al., 2023, n=266 [Primary]
[27] Do You Have a Voting Plan? Implementation Intentions, Voter Turnout, and Organic Plan Making — Nickerson & Rogers, Psychological Science 21(2), 2010, N=287,228 [Primary]
[28] Applying Behavioural Insights to Charitable Giving — BIT, summarised [Secondary — BIT primary returned no content; corroborates [25]]
[29] Behavioural Insights Pilot Project — Organ Donor Registration — Government of Ontario, N>10,000 [Primary — single centre, 8 weeks, small absolute control base]
[30] NHTSA Recall Compliance 2015–2025 — Recall Masters [Secondary, trade — vendor content marketing to dealerships; every NHTSA primary PDF returned 403]
[31] Effectiveness of brief alcohol interventions in primary care populations — Cochrane CD004148 [Primary]
[32] SIPS trial — Kaner et al., BMJ 2013, n=756 [Primary]
[33] Smoking-cessation lapse dynamics — PubMed 21787035 [Partial — cookie wall; search summary only]
[34] Prize-based contingency management meta-analysis — Benishek et al., 19 studies [Primary for effect sizes]; Lussier et al. 2006 moderators [Partial]
[35] Gerber, Green & Larimer (2008), APSR — Yale ISPS record [Partial/Secondary — the APSR PDF is a scanned image with no text layer; per-condition effects and the $1.93/vote figure come from secondary summaries]
[36] COBRA trial — Lancet 2016, n=440 non-inferiority RCT [Primary]
[37] Fixing Frictions: ‘Sludge Audits’ Around the World — OECD / NSW Government, 2024 [Primary]
[38] Attitudes to organ donation in Wales — NHSBT ODT [Partial]; NHSBT news release on Welsh Government figures [Secondary, official — promotional framing, no counterfactual]
[39] Wales’ landmark opt-out organ donation law has had little impact — reporting McLaughlin & Noyes, Bangor University [Secondary — journalism reporting named academics; no peer-reviewed venue named, no figures; the peer-reviewed sceptical source returned HTTP 402]
[40] Service Fees and Compensation Payment Information — AirHelp [Primary]
[41] Bill Negotiation charge explained — Rocket Money [Primary]
[42] Billshark [Primary]
[43] Capitalize [Primary — outcome figures are self-reported and unaudited]
[44] Nous [Primary]
[45] OneMain MyMoney (formerly Trim) [Primary for the rebrand; Partial on bill-negotiation discontinuation]
[46] Paribus decommissioning / Capital One Shopping Price Protection ending [Secondary — search results only; re-verify dates before external use]
[47] PayPal Honey [Secondary — the affiliate-override claims are allegations, not adjudicated findings]; PayPal completes acquisition [Secondary]
[48] FTC Finalizes Order with DoNotPay — FTC, 11 February 2025 [Primary]
[49] Mortgage broker share rises to 38% — Canadian Mortgage Trends, reporting Mortgage Professionals Canada [Primary trade press reporting an industry-body survey — MPC is an interested party]
[50] UK ~80% / Australia 77.6% broker share (MFAA, June-2025 quarter) [Secondary — industry bodies, not fetched]
[51] Motivational interviewing: a systematic review and meta-analysis — Rubak, Sandbæk, Lauritzen & Christensen, Br J Gen Pract 55(513), 2005 [Primary — note the “74% of trials” figure is vote-counting, not an effect size]
[52] Lundahl & Burke (2009), J Clin Psychol In Session 65(11):1232–1245 — practice-friendly review of four MI meta-analyses [Primary]
[53] Magill et al. (2018), J Consult Clin Psychol — MI process meta-analysis, 58 reports / 36 studies / N=3,025 [Primary]
[54] West (2005), Addiction 100:1036–1039, “Time for a change: putting the Transtheoretical Model to rest” [Partial — HTTP 402; the recommendation rests primarily on the absence of positive stage-tailoring evidence across all retrieved sources, not on West’s text]
[55] Patient navigation for breast screening: meta-analysis of RCTs — 15 RCTs [Primary]
[56] FTC click-to-cancel / Negative Option Rule — 8th Circuit vacatur July 2025, ANPRM January 2026 [Secondary — law-firm alerts; ftc.gov returned 403 twice and federalregister.gov redirected to a blocking interstitial]
[57] OSFI stress-test exemption for uninsured “straight switches” at renewal, 21 November 2024 [Secondary — mortgage-industry commentary only; re-verify against OSFI primary before external use]
[R1] Round 1 report — Behavioral-Knowing-Doing-Gap, sources [1]–[79]

Explicitly not retrieved this session (stated rather than inferred): Johnson & Goldstein (2003, Science) on opt-in/opt-out defaults; Green & Gerber’s cost-per-vote-by-channel meta-analysis; automatic voter registration and vote-by-mail effect sizes; EU Payment Accounts Directive switching evaluations (EUR-Lex returned empty); mobile number portability churn effects; GPO/reverse-auction savings (only vendor marketing surfaced); change-of-address bundling; real-estate closing coordinators; donor-advised funds; loyalty status matching and credit-card price protection; professional-licensing renewal; energy-retrofit one-stop-shop conversion data; Fogg (2009) full text — cite the B=MAP model, do not quote it; a publication-bias-corrected re-estimate of the d = .65 implementation-intentions figure; and Lally et al. (2010) on habit-formation timing. Morningstar’s behavioural team, Ontario BIU, US OES and Impact Canada’s behavioural science pages remained inaccessible across both rounds.

A shared 200-call WebSearch budget was exhausted during this round; the gaps above are consequences of that ceiling and are stated as absences of evidence, not as findings.

Notes

This report is frozen prior art (see Behavioural-Solutions), unfrozen only for this Notes glossary plus acronym links below — every occurrence, not just first use (Talbot, 2026-08-15) — no other body edits. Evidence-tier tags ([Primary]/[Secondary]/[Partial], [FR]/[MA]/[1RCT]/[LAB]/[NULL]/[UNEV]) are defined inline at their own first use (see §“Tiers” near the strategy tables) and are not repeated here.

FCA — UK Financial Conduct Authority (financial-services regulator)
CASS — Current Account Switch Service (the UK bank-switching guarantee scheme)
BIT — UK Behavioural Insights Team (“Nudge Unit”) — originated MINDSPACE, then EAST
MINDSPACE — BIT’s earlier, retired 9-lever framework: Messenger, Incentives, Norms, Defaults, Salience, Priming, Affect, Commitments, Ego
USP — Unique Selling Proposition
KYC — Know Your Customer (mandatory identity-verification step for financial accounts)
RCT — Randomized Controlled Trial
AOR — Adjusted Odds Ratio (a statistic measuring effect size, adjusted for confounders)
ATT — Average Treatment effect on the Treated (a causal-inference statistic)
AER — American Economic Review; QJE — Quarterly Journal of Economics; JAMA — Journal of the American Medical Association; NBER — National Bureau of Economic Research — journal/publisher names, citation context only
FAFSA — Free Application for Federal Student Aid (US)
EITC — Earned Income Tax Credit (US)