What outside research says for and against PSI.

(core thesis). The founding TAM paper. Perceived usefulness correlated with current usage (r = .63) and predicted future usage. Ease of use looked more like an antecedent of usefulness than a direct driver of usage. Supports the idea that adoption depends on how users see the change, not just objective value. Caveat: PSI's Experience factors are operational metrics, not TAM's perception constructs.

(Experience factors), with caveats. UTAUT unified eight earlier models around four determinants: performance expectancy, effort expectancy, social influence, and facilitating conditions. PSI's training and support factors map loosely to facilitating conditions, and champion strength to social influence. PSI has no moderators and no performance or effort expectancy terms.

(core thesis). Adoption depends on relative advantage, compatibility, complexity, trialability, and observability, not on relative advantage alone. Complexity and compatibility correspond roughly to PSI's Pain dimensions.

Net benefits are an outcome of system, information, and service quality, flowing through use and user satisfaction. This matches PSI's split between value created and value captured through use.

CFIR's five domains (intervention, outer setting, inner setting, individuals, process) support PSI's view that organizational context drives implementation. The authors say CFIR does not specify relationships between constructs, so PSI's weights and formula cannot be derived from it.

The update puts more weight on the people who receive the innovation. That is relevant to PSI's adoption inputs (end users, champions). Its constructs are still descriptive, not a scoring formula.

Adoption in health service organizations depends on organizational and contextual factors, not only on the merits of the innovation.

(core thesis), with a caveat on the formula. Programs with complexity across several domains "rarely, if ever" became mainstream, which directly backs the split between creating and capturing value. NASSS stresses interacting complexity that changes over time, which sits uneasily with averaging four independent Pain ratings.

Defines implementation outcomes (acceptability, adoption, feasibility, penetration, implementation cost, sustainability) as separate from service and clinical outcomes. This gives formal backing for scoring "will they capture it" apart from "is the value real", plus vocabulary for the Experience factors (for example, penetration in place of adoption %).

(weak evidence base). Eight common errors drawn from observing 100+ companies. Supports PSI's weight on change management, but it is a practitioner essay, not an empirical study.

(popular statistics). Found no valid, reliable evidence for the "70% fail" figure. PSI material should not cite it.

Defines readiness as shared commitment to the change plus shared belief in the ability to carry it out. Supports including readiness in PSI. But a seller's single 1 to 10 rating does not measure that shared state, which is an argument for collecting Pain and Experience from the buyer side as well.

Across 1,471 IT projects the average cost overrun was 27%, but about one in six was a "black swan" averaging a 200% cost overrun. Supports wide uncertainty bands. Challenges PSI's triangular distributions, which are bounded and cannot produce fat tails.

(popular failure statistics). Argues the Standish CHAOS figures are meaningless because they average estimates with unknown biases. Also a caution for PSI, which averages ratings from scorers who may have different biases.

Attributes forecast error to optimism bias and strategic misrepresentation, and recommends reference class forecasting. Supports grounding Gain and timelines in past deals. Challenges taking seller-estimated Gain at face value.

92% of recent articles reached positive conclusions, but provider dissatisfaction remained a barrier. Value creation is real; capture depends on people.

Most of 236 studies reported positive effects on quality, but 61% reported little about implementation context. That limits how far published evidence can calibrate PSI's organizational and adoption weights.

(core thesis). RAND's 2005 projection of more than $81 billion a year in savings had not materialized. The authors tie the gap to usability, interoperability, and care-process redesign. A direct example of business-case size not predicting realized value.

Across 26 practices, productivity and net income dipped after EHR go-live, then recovered to or near earlier levels by 12 months. Supports a time-to-value factor and a post-go-live dip. The effects were smaller and shorter than expected.

(input quality), SUPPORTS (calibration loop). Executives routinely overstate benefits and discount costs. Applies directly to seller-entered Gain (likely inflated) and Pain (likely understated). Companion: Flyvbjerg, B. (2003). Delusions of success: Comment on Dan Lovallo and Daniel Kahneman. HBR, December 2003, 121-122. https://arxiv.org/abs/1305.0741. Argues that deliberate misrepresentation also drives forecast error, which matters because sellers have an incentive to inflate PSI inputs.

(indirect). B2B buying has become complex enough to stall customers, so sellers should track buying progress. This matches PSI's attention to buyer-side friction and staleness. Practitioner research, not peer-reviewed. The widely quoted "6.8 stakeholders" figure could not be confirmed from the primary text, so do not cite it from here.

(leans CHALLENGES). Normalization, weighting, and aggregation choices change results and should be stress-tested. PSI's Monte Carlo covers input uncertainty, not uncertainty about how the index itself is built.

(the "NPS for healthcare tech" positioning). Net Promoter did not clearly beat other metrics at predicting revenue growth. PSI should prove predictive validity before claiming single-number status.

(the arithmetic on Pain). 1 to 10 ratings are ordinal at best. Using their average as a ratio denominator assumes equal intervals and a true zero the ratings do not have.

(partial defense). Parametric statistics hold up well on Likert-type data, which partly defends averaging the ratings. It does not justify dividing by an average rating as if it were a ratio-scale quantity.

Supports diminishing sensitivity to larger gains. But value is measured from a reference point, losses weigh more than gains, and the curve is not specified as a logarithm. Under prospect theory, Pain arguably deserves more weight than Gain.

The origin of logarithmic utility, which supports the log transform in principle. But in Bernoulli the value of a gain depends on existing wealth, which implies Gain should be scaled to the buyer's size, not taken in absolute dollars.

(the calibration loop), with a precondition. The standard accuracy score for probability forecasts. PSI outputs an index, so it must first be mapped to a predicted probability (of close, or of value realized) before Brier scoring applies.

Training, teaming, and tracking improved forecast calibration. Supports training the people who enter PSI inputs and tracking their accuracy over time.