The Half-Life of a Best Practice

Executive summary

Health systems have long used “best practice” as a finish line: a pilot succeeds, the approach gets the label, and the label stays put. That habit was already risky. Well-designed trials routinely overturn established care, and the commonly cited path from research evidence to routine practice runs to about 17 years.[1, 2, 3] Artificial intelligence now makes the habit untenable. Technology that once changed every 18 months now changes every quarter, and frontier AI models reach the market every few days.[4, 5]

This paper argues for a stricter and more useful definition. An organization may call something a best practice only when two conditions hold at the same time:

• It is proven. The practice has delivered a measured, finance-validated return in this organization, against a baseline agreed before it launched.
• It is current. The evidence has been re-examined since the last material change to the technology, workflow, people, or patients that the practice depends on.

Neither condition is enough alone. A proven practice that has not been reviewed is a legacy practice running on old evidence. A reviewed practice that has never shown a return is an experiment. The practical consequence is the paper’s central operating rule: the review cycle for a best practice must match the clockspeed of the organization that runs it. A system that changes its technology quarterly but reviews results annually spends roughly three quarters of every year operating on unverified claims.

We propose that health systems build a project management office, or PMO, that acts as the organization’s clock keeper. It would own the register of practices, record when each was last proven, track the “evidence debt” that builds with every change, and trigger review before that debt turns into harm or wasted spend.

1. The end of the permanent best practice

Medicine has good evidence that its own standards expire. A review of a decade of original articles in the New England Journal of Medicine found 363 studies that tested an established standard of care. Of those, 146, or 40.2 percent, reversed the practice, while 138 reaffirmed it.[2] A later review of more than 3,000 randomized trials in JAMA, The Lancet, and NEJM identified 396 medical reversals.[3] These were not fringe practices. They were accepted approaches that a stronger test later showed to be ineffective or inferior.

Even the syntheses that underpin best practices age quickly. A survival analysis of high-quality systematic reviews found that the median time before new evidence signaled a need to update was 5.5 years; 15 percent showed such a signal within one year and 23 percent within two.[6]

The system is also slow to act on new evidence. The frequently repeated estimate that research takes 17 years to reach clinical practice is itself contested; its authors note that the figure “hides complexities” because different studies measure different stages.[1] Even so, no credible reading suggests the lag is short. Meanwhile, the volume of knowledge keeps growing. One widely cited estimate projected that the doubling time of medical knowledge would fall from 3.5 years in 2010 to about 73 days by 2020.[7] That is a projection, not a measurement, but its direction is not in dispute.

Put these facts together and a pattern emerges. Best practices get declared quickly and retired slowly, if at all. The label outlives the evidence. AI does not create this problem. It speeds it up to the point where leaders can no longer ignore it.

2. Three clocks, running at different speeds

MIT’s Charles Fine introduced the idea of “clockspeed” to describe how quickly an industry’s products, processes, and organizations change. He observed that different industries, and even different capabilities inside one firm, “change, grow, or obsolesce at different rates.”[8] Health systems now run on at least three clocks, and the gap between them explains why best practices decay.

The technology clock

This is the pace at which the tools change. At its 2026 users group meeting, Epic described how its release cadence has shifted: “Back in the day, we released new versions about every 18 months,” said Seth Howard, Epic’s executive vice president of research and development. “Now we release new versions every three months. … Today’s three-month release contains more code than the typical 18-month release of the past.”[4] Frontier AI moves faster still. Analysis by ARK Investment Management, based on Artificial Analysis data, put the median gap between frontier model releases from leading U.S. labs at 37.5 days in 2023 and 11 days in 2026 so far.[5]

The organization clock

This is the pace at which a particular health system chooses to absorb change, and it varies widely. Health systems upgrade their EHRs anywhere from annually to quarterly. Memorial Hermann upgrades quarterly and stays no more than one release behind. Northwell and Cleveland Clinic upgrade twice a year. Nebraska Methodist Health System updates its Oracle Health EHR annually; its leader explained, “there has to be a reason why we’re going to introduce change into our environment.”[9] Buying is speeding up as well: Menlo Ventures reports that health-system purchasing cycles for AI average 6.6 months, compared with 8.0 months for traditional IT.[10]

The evidence clock

This is the pace at which the organization proves that a practice still works. In most systems it is the slowest clock. Business cases are written once. Benefit reviews, where they exist, are annual or ad hoc. Federal survey data show that in 2024, 79 percent of hospitals using predictive AI reported post-implementation evaluation or monitoring, but 18 percent answered “don’t know” when asked whether they did so at all.[11]

Figure 1. The technology clock now runs far faster than the evidence clock. Sources: Morris et al. (research-to-practice estimate, contested)[1]; Menlo Ventures (buying cycles)[10]; Becker’s Hospital Review (Epic release cadence)[4]; ARK Investment Management via OfficeChai (frontier model intervals)[5]. Cycle lengths are shown on a logarithmic scale.

The key insight is not that technology moves fast. Leaders already know that. It is that a best practice fails when the evidence clock runs slower than the organization clock. Every time the organization changes something a practice depends on, whether a software version, a staffing model, a patient population, or a payer rule, the old proof describes a slightly different organization. The gap between the last proof and the present is what we call evidence debt. Like technical debt, it builds quietly and costs more the longer it is left alone. Cleveland Clinic’s IT leadership described the same dynamic for EHR upgrades: “We have found it’s important not to get behind, which makes for a significant backlog and higher level of effort over time.”[9]

This also frees leaders from a false choice. An organization does not have to match the vendor’s clock. Nebraska Methodist’s deliberate annual cadence and Memorial Hermann’s quarterly cadence can both be sound. What cannot be sound is changing on one clock and proving on a slower one.

3. Two tests that earn the name

If a best practice must be both proven and current, every practice in a health system falls into one of four categories. The grid below is a KinRidge framework, not a published standard, but it gives executives a shared language for an uncomfortable conversation.

The grid gives a practical meaning to the phrase “in today’s age.” In a slow-changing environment, a proven practice could stay in the top-left cell for years. In an AI-enabled environment, it can slide into the legacy cell within a single release cycle without anyone noticing, because the practice itself was never changed. Only its context was.

4. How this builds on prior work

The idea that knowledge decays is not new, and this paper draws on several established bodies of work. What it adds is a way to connect them at the level where health systems actually operate: the individual practice, inside one organization, judged by its return.

Four elements of the paper are, to our knowledge, new in combination: the requirement that a practice pass both an ROI test and a currency test before it earns the name; the rule that review cadence must match the organization’s chosen clockspeed rather than a calendar or a vendor’s release schedule; the concept of evidence debt applied to operating practices; and the proven–current grid as a portfolio tool. These are KinRidge propositions offered for testing, not settled findings.

5. ROI is the entry ticket

Enthusiasm is not evidence, and adoption is not return. A widely cited 2025 study of enterprise generative AI from MIT’s NANDA initiative found that about 95 percent of companies in its dataset were not seeing measurable profit-and-loss impact from generative AI pilots. It attributed the gap less to model quality than to a “learning gap”: tools that “don’t learn from or adapt to workflows.”[12] In healthcare, Bain and KLAS report that executives now prioritize AI with “tangible results and short time-to-value windows,” even as many providers say it is still early to quantify precise financial returns.[13]

Ambient clinical documentation shows why the ROI test matters, and why it has to be local.

• The benefit to clinicians looks real. Peterson Health Technology Institute found that early adopters report reduced burnout and cognitive load, but that “the financial impact is unclear.” Its executive director put the challenge plainly: health systems “need to measure the results.”[14]
• Products in the same category performed differently. In a randomized trial of 238 UCLA Health physicians, one ambient scribe reduced time-in-note by 9.5 percent relative to usual care, while another showed no significant reduction.[15] “Ambient AI” is a category, not a practice. The practice is a specific tool in a specific workflow in a specific organization.
• Scale does not equal return. The Permanente Medical Group reported that ambient AI saved nearly 16,000 hours of documentation across 7,260 physicians and about 2.6 million encounters between October 2023 and December 2024.[16] The total is impressive. Spread across all participating physicians, it averages roughly two hours each over the period. That time becomes ROI only when the organization decides what it is for: more visits, shorter days, better retention, or better notes. Each of those must then be measured.

Health-system leaders acknowledge the difficulty. Mayo Clinic’s chief AI implementation officer noted that on the care delivery side “it’s really hard to apply line of sight” to ROI, and Ardent Health’s chief digital officer cautioned that success metrics must be “clearly defined and measurable alongside a baseline.”[17] The answer is not to drop ROI as the standard. It is to widen what counts while keeping it disciplined. We recommend a value ledger with six lines, agreed with finance before launch:

6. A best practice in four acts: the sepsis model

No example shows the half-life of a practice more clearly than the Epic Sepsis Model, one of the most widely deployed predictive tools in U.S. hospitals.

• Act one: adoption. Hospitals across the country implemented the model on the strength of developer-reported performance.
• Act two: local evidence. A 2021 external validation at Michigan Medicine covering 38,455 hospitalizations found an area under the curve of 0.63, “substantially worse than the performance reported by its developer.” The model missed 67 percent of patients with sepsis while alerting on 18 percent of all hospitalized patients.[19]
• Act three: the context moved. In April 2020, the University of Michigan deactivated the model because of spurious alerting tied to pandemic-era changes in patient characteristics, which “fundamentally altered the relationship between fevers and bacterial sepsis.”[20] The model had not changed. The organization had.
• Act four: the practice was rebuilt locally. Epic released a revised model in 2022 that could be fine-tuned on each site’s historical data. A 2026 prospective study across four health systems found better discrimination, with AUROC between 0.82 and 0.92, but high variability across institutions: the alert threshold needed to reach 60 percent sensitivity ranged from 14 to 37 depending on the site. The authors concluded that health systems “should conduct internal evaluations to validate local performance.”[21]

Across all four acts, the product carried the same name. Whether it was a best practice depended entirely on when and where someone looked. That is the thesis of this paper in a single case.

7. Review at the speed of your organization

If evidence debt builds with every material change, then the right review interval for a practice is not a fixed calendar date. It is set by the fastest-changing element the practice depends on, and it is shortened by risk. A practice that relies on an AI model updated monthly cannot be governed by an annual review, however disciplined that review is. A stable process with stable inputs does not need monthly scrutiny, and imposing it would burn out the teams doing the work.

The table below translates that principle into an operating design. The intervals are illustrative starting points for calibration, not published standards.

Three design rules keep this approach workable:

• Choose your clock deliberately. An organization’s clockspeed should be a leadership decision, not a side effect of vendor release schedules. Then align the review cadence to the clock you chose.
• Review the dependency, not just the practice. Tag each practice with the systems, models, roles, and populations it relies on. When any of those changes, the practice becomes due for review automatically.
• Make “last proven” visible. Every practice in the register should display the date it last passed both tests. A best practice without a date is a belief.

8. Why AI shortens the half-life

Traditional practices decay when people drift from the design. AI-enabled practices can decay even when everyone follows the design exactly, because the model, the data, or the way people use its output has changed. Researchers describe this as dataset shift, which occurs “when a machine-learning system underperforms because of a mismatch between the data set with which it was developed and the data on which it is deployed.” Common causes include changes in technology, in population and setting, and in behavior, such as new reimbursement incentives.[20] The FDA has similarly described how changes in clinical practice, patient populations, data inputs, and infrastructure can degrade the real-world performance of AI-enabled medical devices; that discussion concerns regulated devices, not every workforce tool.[22]

Accreditors are now moving toward the position this paper argues for. In September 2025, the Joint Commission and the Coalition for Health AI released guidance calling for policies, “appropriate local validation,” and monitoring, with Joint Commission’s CEO noting “how quickly AI is changing healthcare – and at a scale I’ve never seen.”[23] In June 2026, the Joint Commission issued a review guide for its Responsible Use of AI in Health Care certification. The guide asks organizations to show evidence such as “performance/bias dashboards, drift alerts, review cadence, revalidation triggers,” and to explain how they “validate vendor-reported performance.”[24] NIST’s AI Risk Management Framework likewise treats risk management as continuous across the AI life cycle.[25]

AI can also speed up the evidence clock itself. Researchers writing in the Journal of Medical Systems describe language-model systems that reproduced and updated an entire issue of Cochrane reviews, 12 reviews covering 146,276 citations, in under two days of work, compared with an estimated 12 person-years by traditional methods. They caution that performance is uneven and requires human oversight.[26] The implication for health systems is significant. The same technology that shortens the half-life of a practice can also shorten the time needed to re-prove it, if the organization builds the capability to use it.

9. The PMO as the organization’s clock keeper

A conventional project office coordinates schedules, budgets, and deliverables. The PMO proposed here would also act as an internal team of experts in process design, implementation, change management, data, finance, clinical operations, and technology. It would not replace service-line or frontline ownership. It would give those leaders a consistent method for moving from idea to evidence to adoption, and then back to evidence. AHRQ describes systematic learning from internal data and external evidence in learning health systems, and published healthcare case studies describe PMO and transition-support teams contributing project, change, and evaluation expertise. Those studies support feasibility, not a universal ROI claim for the model proposed here.[27, 28, 29]

The living practice register

The PMO’s central tool is a register in which each practice records the problem it addresses, the essential steps that produce the benefit, allowable local adaptations, the sites where it has been tested, its implementation cost, its value ledger, its dependencies, its clockspeed tier, its last-proven date, its review owner, and its next review trigger. A practice advances from candidate to locally proven to transferable only as evidence builds up. It moves to under review when a trigger fires, and to retired when the evidence no longer supports it. Retirement is a success of the system, not a failure of the team.

IHI’s Model for Improvement uses small tests of change to learn how an intervention works locally before broader implementation, and the StaRI reporting standard emphasizes describing both fidelity and adaptation.[30, 31] Together they support a practical rule: preserve the elements that cause the benefit, and let the implementation approach respond to local conditions. Variation between sites, like the sepsis thresholds that ranged from 14 to 37, should be explained rather than averaged away.[21]

Three rhythms, one system

The PMO sets three connected rhythms rather than one fixed calendar. Frontline teams monitor critical exceptions as work happens. Initiative owners review adoption, benefit, and unintended effects at the interval set by the practice’s clockspeed. Executives periodically decide which practices to expand, refine, replace, or stop. IHI’s work on sustaining improvement places responsibility in the daily management system, supported by leaders at several levels, and the dynamic sustainability framework argues for continued adaptation in changing settings rather than fixed fidelity to an original design.[33, 34]

10. Independent implementation, internal ownership

An independent implementation partner can bring a useful outside view at the start. It can map the real process across departments, surface inconvenient baseline data, test assumptions made by software vendors, and align the business case with clinical and financial leaders. That independence matters more as vendors move faster: the Joint Commission’s certification guide explicitly asks how organizations validate vendor-reported performance.[24] The MIT NANDA research also found that line managers, not only central AI teams, need to drive adoption.[12]

Evidence on facilitation is nuanced. Trials and comparative studies have examined external and blended internal–external facilitation, with effects that vary by intervention and setting. They do not show that an external team is always better.[32] The claim here is narrower: an independent partner can help establish credible methods and challenge conflicts of interest, while internal ownership is necessary to embed and adapt the work. The goal is local capability, not permanent dependence on a consultant.

11. Five questions for the executive team

Leaders can test their own position against this thesis with five questions:

• What is our clockspeed? How often do we change the technology, workflows, and staffing models our most important practices depend on, and did we choose that pace on purpose?
• Which of our “best practices” have a last-proven date? For those that do not, what evidence would we show a board member who asked today?
• Where is our evidence clock slower than our organization clock? Start with AI-enabled and EHR-dependent practices.
• What have we retired in the past year? A portfolio that never retires anything is not learning.
• Could our own team defend the ROI method without the vendor in the room? If not, the organization does not yet own its evidence.

12. How KinRidge can help

A practical engagement begins with a clockspeed diagnostic: an inventory of active practices and AI-enabled workflows, their dependencies, their last-proven dates, and the gap between the organization clock and the evidence clock. KinRidge and the health system then design the PMO charter, decision rights, staffing, value ledger, practice register, and review cadence together.

The teams then run a small number of initiatives through the full cycle, including at least one AI-enabled implementation and one process redesign, while training internal leaders to do the analysis and facilitation themselves. KinRidge helps stand up the practice-tracking project office — the register, the review triggers, the value ledger — and then hands ownership to the client’s own team. The transition is complete when the health-system team can independently evaluate a new idea, lead adoption, defend its ROI method, keep its evidence current, and recommend when a practice should change or retire.

References

Sources are linked for verification. This paper distinguishes published evidence from KinRidge recommendations and illustrative operating models. Frameworks labeled as KinRidge designs are proposals, not published standards.

1. Morris ZS, Wooding S, Grant J. The answer is 17 years, what is the question: understanding time lags in translational research. Journal of the Royal Society of Medicine. 2011. https://journals.sagepub.com/doi/10.1258/jrsm.2011.110180
2. Prasad V et al. A decade of reversal: an analysis of 146 contradicted medical practices. Mayo Clinic Proceedings. 2013. https://pubmed.ncbi.nlm.nih.gov/23871230/
3. Herrera-Perez D et al. A comprehensive review of randomized clinical trials in three medical journals reveals 396 medical reversals. eLife. 2019. https://pmc.ncbi.nlm.nih.gov/articles/PMC6559784/
4. Becker’s Hospital Review. 34 new Epic features coming — and when. August 24, 2026. https://www.beckershospitalreview.com/healthcare-information-technology/innovation/34-new-epic-features-coming-and-when/
5. OfficeChai. Frontier labs are releasing new models faster than ever, shows data (ARK Investment Management analysis of Artificial Analysis data). April 29, 2026. https://officechai.com/ai/frontier-labs-are-releasing-new-models-faster-than-ever-shows-data/
6. Shojania KG et al. How quickly do systematic reviews go out of date? A survival analysis. Annals of Internal Medicine. 2007. https://pubmed.ncbi.nlm.nih.gov/17638714/
7. Densen P. Challenges and opportunities facing medical education. Transactions of the American Clinical and Climatological Association. 2011. https://pmc.ncbi.nlm.nih.gov/articles/PMC3116346/
8. Fine CH. Industry Clockspeed and Competency Chain Design: An Introductory Essay. MIT Sloan working paper. 1996. https://dspace.mit.edu/bitstream/handle/1721.1/2614/SWP-3893-36119547.pdf?sequence=1
9. Becker’s Hospital Review. Health systems rethink EHR upgrades amid AI shift. April 28, 2026. https://www.beckershospitalreview.com/healthcare-information-technology/innovation/health-systems-rethink-ehr-upgrades-amid-ai-shift/
10. Menlo Ventures. 2025: The State of AI in Healthcare. October 2025. https://menlovc.com/perspective/2025-the-state-of-ai-in-healthcare/
11. Office of the National Coordinator for Health Information Technology. Hospital Trends in the Use, Evaluation, and Governance of Predictive AI, 2023–2024. https://healthit.gov/data/data-briefs/hospital-trends-use-evaluation-and-governance-predictive-ai-2023-2024/
12. Fortune. MIT report: 95% of generative AI pilots at companies are failing (on MIT NANDA, The GenAI Divide: State of AI in Business 2025). August 18, 2025. https://fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo/
13. Bain & Company and KLAS Research. Healthcare AI investment focusing on hard-dollar returns and clinical workflows amid shift from pilots to widespread production. 2025. https://www.bain.com/about/media-center/press-releases/20252/healthcare-ai-investment-focusing-on-hard-dollar-returns-and-clinical-workflows-amid-shift-from-pilots-to-widespread-productionbain--company-and-klas-research/
14. Peterson Health Technology Institute. Leading Health Systems: AI-Powered Scribes Alleviate Clinician Burnout; Financial Impact Unclear. March 25, 2025. https://www.phti.org/announcement/ai-scribes-reduce-clinician-burnout/
15. Lukac PJ et al. Ambient AI Scribes in Clinical Practice: A Randomized Trial. NEJM AI. 2025. https://pmc.ncbi.nlm.nih.gov/articles/PMC12768499/
16. Kaiser Permanente Division of Research. AI-assisted notetaking gains steady support from Kaiser Permanente physicians. 2025. https://divisionofresearch.kaiserpermanente.org/ai-assisted-notetaking-gains-steady-support-from-kaiser-permanente-physicians/
17. Healthcare Dive. Health systems are racing to adopt AI. But can they prove ROI? October 27, 2025. https://www.healthcaredive.com/news/healthcare-ai-adoption-return-investment-roi-hlth/803653/
18. Identifying value in healthcare transformation initiatives: an evaluation of an approach to benefits realisation. BMJ Open Quality. 2023. https://doi.org/10.1136/bmjoq-2023-002349
19. Wong A et al. External Validation of a Widely Implemented Proprietary Sepsis Prediction Model in Hospitalized Patients. JAMA Internal Medicine. 2021. https://pmc.ncbi.nlm.nih.gov/articles/PMC8218233/
20. Finlayson SG et al. The Clinician and Dataset Shift in Artificial Intelligence. New England Journal of Medicine. 2021. https://pmc.ncbi.nlm.nih.gov/articles/PMC8665481/
21. Wong A et al. Multicenter Prospective Validation of an Updated Proprietary Sepsis Prediction Model. JAMA Network Open. 2026. https://pmc.ncbi.nlm.nih.gov/articles/PMC12949446/
22. U.S. Food and Drug Administration. Measuring and Evaluating AI-Enabled Medical Device Performance in the Real World: Request for Public Comment. 2025. https://www.fda.gov/medical-devices/digital-health-center-excellence/request-public-comment-measuring-and-evaluating-artificial-intelligence-enabled-medical-device
23. Joint Commission and Coalition for Health AI. Joint Commission and CHAI release initial guidance to support responsible AI adoption across U.S. health systems. September 17, 2025. https://www.chai.org/news/joint-commission-and-coalition-for-health-ai-chai-release-initial-guidance
24. The Joint Commission. Responsible Use of Artificial Intelligence in Health Care Certification: Review Process Guide. June 1, 2026. https://digitalassets.jointcommission.org/api/public/content/7bd26620a3e54c0ead2138afa37fc167?v=5f349b78
25. NIST. AI Risk Management Framework Core. https://airc.nist.gov/airmf-resources/airmf/5-sec-core/
26. Peltonen LM, Topaz M, Zhang Z. From Research to Practice in Days, not Decades: Why Leaders Must Act Now. Journal of Medical Systems. 2025. https://pmc.ncbi.nlm.nih.gov/articles/PMC12672604/
27. Agency for Healthcare Research and Quality. How Learning Health Systems Learn: Lessons from the Field. https://www.ahrq.gov/learning-health-systems/how-lhs-learn.html
28. Implementation of evidence-based practices in the context of a redevelopment project in a Canadian healthcare organization. 2012. https://pubmed.ncbi.nlm.nih.gov/23121763/
29. A Health Care Project Management Office’s Strategies for Continual Change and Continuous Improvement. 2017. https://pubmed.ncbi.nlm.nih.gov/28953067/
30. Pinnock H et al. Standards for Reporting Implementation Studies: StaRI statement. BMJ. 2017. https://www.bmj.com/content/356/bmj.i6795
31. Institute for Healthcare Improvement. Model for Improvement: Testing Changes. https://www.ihi.org/library/model-for-improvement/testing-changes
32. Smith SN et al. Comparative effectiveness of external versus blended facilitation on collaborative care model implementation. Health Services Research. 2020. https://pmc.ncbi.nlm.nih.gov/articles/PMC7704469/
33. Chambers DA et al. The dynamic sustainability framework: addressing the paradox of sustainment amid ongoing change. Implementation Science. 2013. https://link.springer.com/article/10.1186/1748-5908-8-117
34. Institute for Healthcare Improvement. Sustaining Improvement. White paper. https://www.ihi.org/library/white-papers/sustaining-improvement
35. Australian Living Evidence Collaboration. Living Guidelines Handbook, version 1.1. August 2025. https://livingevidence.org.au/wp-content/uploads/Living-Guidelines-Handbook-v1.1-Aug2025.pdf
36. Healthcare Innovation. VAMOS Collaborative: An Open Source Platform for Trustworthy AI. May 2026. https://www.hcinnovationgroup.com/analytics-ai/ai-governance/article/55379896/vamos-collaborative-an-open-source-platform-for-trustworthy-ai
37. De-Implementation of Low-Value Care. 2023. https://pmc.ncbi.nlm.nih.gov/articles/PMC10842822/
38. Arbesman S. The Half-Life of Facts: Why Everything We Know Has an Expiration Date. Current/Penguin. 2012. https://archive.org/details/halflifeoffactsw0000arbe
39. McKinsey & Company. The health system CEO imperative: Turning AI’s promise into performance. June 15, 2026. https://www.mckinsey.com/industries/healthcare/our-insights/the-health-system-ceo-imperative-turning-ais-promise-into-performance
40. Arthur D. Little. Moving from “best practice” to “next practice” to drive effective digital transformation. https://www.adlittle.com/sites/default/files/viewpoints/adl_moving_from_best_practice_to_next_practice_to_drive_effective_digital_transformation-min.pdf