Hotel groups trust what their AI tells them. They still do not run the business on it. The distance between those two positions is the most useful number in hospitality technology today, and closing it is worth more than any additional tool.
Adoption Ran Ahead Of Reliance
The received wisdom is that hospitality is behind on AI. The evidence says something more specific, and more interesting.
h2c's 2025 global study, supported by fifteen technology vendors including Oracle Hospitality, IDeaS and Cendyn, drew on 189 quantitative responses and 26 executive interviews across 171 hotel chains representing more than 11,000 properties and 1.3 million rooms. Seventy-eight per cent of those chains already use AI. Eighty-nine per cent planned to expand within twelve to twenty-four months.
Adoption is high. Everything that would make adoption dependable is not.
Source: h2c GmbH, AI & Automation in Hospitality, October 2025 (adoption, strategy, business-model centrality); Otelier, The 2026 Hotel Operations Index, January 2026 (readiness, integration). Two studies with different samples, shown together to illustrate sequence rather than a single funnel.
Then the same study asked the two questions that matter more than adoption. How much do you trust AI capability, and how much do you rely on it? Trust averaged 6.6 out of 10. Reliance averaged 4.7.
The gap between believing the output and running the business on it
Source: h2c GmbH, AI & Automation in Hospitality, 2025 global study, 189 responses across 171 hotel chains.
That gap of roughly two points is the whole argument. Leaders broadly believe the output. They still will not act on it unattended. Six per cent operate with a comprehensive company-wide AI strategy. One per cent describe AI as central to their business model.
Adoption is not the constraint. Dependability is. And the reasons hotel chains gave were not about models. They cited a lack of AI expertise at 62 per cent, an unclear strategy at 51 per cent, and integration challenges at 45 per cent.
The barriers are capability and design problems, and chains and independents face them differently
Source: h2c GmbH, AI & Automation in Hospitality, October 2025 (barriers); Cloudbeds, 2026 Independent Hotels Report, April 2026 (chain and independent adoption).
The Otelier respondents pointed the same way. Only 25 per cent said they were ready to adopt AI, and 40 per cent said they were not ready at all. Asked where AI would be most valuable, they named predictive demand modelling and cross-department data collaboration, ahead of guest-facing chatbots.
That inverts the industry's public narrative. The people running hotels are not asking for a better chatbot. They are asking for demand signals they can trust, and for their departments to see the same facts.
One divergence is worth holding onto. Cloudbeds reports nearly 80 per cent of hotel chains using AI in some capacity against 41 per cent of independent hotels. Chains and independents are not in the same conversation, and should not be sold the same answer.
The Evalueserve Edge
Reliance is engineered, and it is engineered in production
A system earns reliance the way a colleague does, by being right repeatedly and being checkable when it is not. That is an engineering discipline, and it is the one most often skipped. We run it as a production control plane: model gateway and orchestration, prompt and context versioning, evaluation sets that are refreshed rather than written once, safety and guardrails, and telemetry covering cost, latency, quality and audit logs.
Two habits matter more than any of it. We hold back an answer key, so quality is measured against known-correct outcomes rather than plausibility. And we keep a named human accountable at every point where a decision carries consequence. Reliance follows evidence, not enthusiasm.
The Cost Of Disconnection Is Measurable, And Two Rivals Measured It Identically
Otelier's 2026 Hotel Operations Index, produced with Agilysys and Sage, surveyed hotel owners and operators. Twenty-seven per cent of respondents reported spending more than eleven hours a week consolidating or reconciling data. Twenty-seven per cent run their hotels on more than seven technology platforms.
11%
have a fully integrated technology stack
91%
still rely on some manual reporting
15%
are very confident in their operational data
Source: h2c GmbH, AI & Automation in Hospitality, October 2025 (barriers); Cloudbeds, 2026 Independent Hotels Report, April 2026 (chain and independent adoption).
Separately, Cloudbeds' 2026 Independent Hotels Report, drawn from 90 million bookings across 180 countries, found that independent properties lose the equivalent of one to two working days a week reconciling data across platforms. Sixty-seven per cent still name managing disparate systems as a top concern.
Both studies come from technology companies with products to sell. Read them with that in mind. Then notice what makes them hard to dismiss. Otelier and Cloudbeds sell different products to different customers and share no sponsorship. They reached the same finding at the same magnitude, by different routes, in the same year. Eleven hours a week and one to two working days a week are the same number.
That is not a reporting problem. It is a structural one. No model improves that. It is fixed upstream, in the data.
The Evalueserve Edge
Trusted data is the first deliverable, not the first assumption
This is where our data and AI work begins, because nothing downstream survives a weak foundation. We build the trusted layer that fragmented estates lack: a lakehouse or warehouse with curated data products, ingestion from existing systems of record, master data and a semantic layer so that a rate, a room night and a guest mean one thing across the business, and data quality rules, catalog, lineage and policy so that a number can be traced to its origin.
It is built around the property, reservation, revenue, customer and finance platforms already in place, not as a replacement for them. Most estates do not need fewer systems. They need one version of the truth flowing between the ones they have.
Why Hospitality Specifically
McKinsey and Skift published a joint analysis of agentic AI in travel in September 2025, based on surveys of 1,002 travelers and 86 mostly US-based travel executives. This is travel-wide evidence rather than hotel-specific, and it should be read that way. Its diagnosis is still the clearest available.
They identify two causes of the sector's lag. The first is siloed data and incompatible systems, with hospitality named as especially fragmented, and an absence of centralized data ownership that limits the network effects and feedback loops which normally make AI improve with use. The second is that the industry tends to see itself as specializing in service rather than technology, so technology talent and investment lag.
The Finding That Matters Most
Horizontal deployments, meaning enterprise copilots and chatbots, scaled quickly and delivered diffuse, hard-to-measure gains. Vertical use cases specific to the sector could be more transformative, and most of them remain stuck in pilot.
McKinsey and Skift, Remapping travel with agentic AI, September 2025
Their survey measured the two technologies separately, and at two depths each. For generative AI, 90 per cent of executives said their organization used it in some capacity, and within that group 22 per cent described the use as widespread. For agentic AI, 38 per cent said they were not using it at all, and only 2 per cent described their use as widespread. The widespread figures are subsets of use, not separate populations, which is why each technology accounts for 100 per cent of respondents on its own.
Generative AI is nearly universal but shallow. Agentic AI is neither.
Source: McKinsey and Skift, Remapping travel with agentic AI, September 2025, survey of 86 mostly US-based travel executives. The middle bands, and the generative AI not-using band, are arithmetic complements of those figures.
The most-cited barrier was a lack of technical expertise and talent. The second was the absence of a clear road map for transforming business domains. Neither is a technology gap. Both are capability and design gaps, and that distinction determines what actually fixes them.
The Evalueserve Edge
The pilot trap is a design problem, so we start with the decision
Horizontal tools stall because they are layered onto processes that were designed for people working around missing information. Automating that process preserves the workaround. The harder and more valuable work is re-engineering how the work gets done.
So we do not start with a tool. We start with a decision that needs to improve, and we put a domain engineer on it whose job is problem framing and workflow reality, not code. What is the decision, who owns it, which signals inform it, where does it get stuck, what must a human still approve, and how will we know it improved. Only then does the engineering begin. Vertical use cases escape pilot when they are built as workflows with owners and controls, rather than as features hoping to find a process.
Disconnection Has A Price In The P&L, And On The Balance Sheet
The operational argument is easy to nod along to and easy to defer. The financial one is not.
Hotel economics tightened through 2025. Cloudbeds reports that labour now represents between 47 and 60 per cent of operating expenses depending on region, that acquisition costs through online travel agencies have grown faster than RevPAR since 2019, and that global RevPAR for independent hotels declined 5.4 per cent in 2025 while the online travel agency share of independent bookings rose to 63.4 per cent. The market also split. Ultra-luxury RevPAR grew 10.6 per cent while US economy hotels recorded eighteen consecutive months of RevPAR decline. Operators are increasingly measuring GOPPAR alongside RevPAR, because what a hotel keeps has become a more urgent question than what it earns.
Against that backdrop, consider what sits inside loyalty data.
Marriott's Form 10-K for the 2025 financial year discloses that, based on conditions at 31 December 2025 and holding other factors constant, a one percentage point decrease in its estimate of loyalty point breakage could increase the guest loyalty programme liability by approximately $50 million.
A single estimation assumption inside operational data carries a disclosed fifty-million-dollar consequence
Source: Marriott International, Inc., Form 10-K for the fiscal year ended 31 December 2025, and Marriott 2025 Annual Report; Hilton Worldwide Holdings Inc., Form 10-K for the fiscal year ended 31 December 2025. Both filed with the US Securities and Exchange Commission.
Read that as a data statement rather than an accounting one. A single estimation assumption, derived from member behaviour data, moves a disclosed liability by roughly $50 million per percentage point. The programme it describes reached about 271 million members at the end of 2025 after adding 43 million during the year and accounted for approximately 75 per cent of Marriott's US room nights and 68 per cent of its global room nights. Hilton, for scale, closed 2025 with 243 million Hilton Honors members across 9,158 properties and 1,351,351 rooms in 143 countries.
Loyalty data is not a marketing asset that happens to be large. At this scale it is a financially material accounting input, subject to audit, with a disclosed sensitivity. Data quality in hospitality has a number attached to it, and the number is in the annual report.
The majors have already moved from interest to expenditure
The commercial ground is shifting at the same time. Cloudbeds reports that the share of US travelers using traditional search engines for trip planning fell from 51 per cent to 36 per cent in a single year while use of generative AI platforms more than doubled, and that online travel agencies account for more than half of citations in AI-generated hotel recommendations
What is not in doubt is how the largest operators are responding. Marriott initially guided $1.0 to $1.1 billion of investment spending for 2026, with roughly 35 to 40 per cent directed at digital technology transformation and corporate systems. By its second quarter August 2026 earning call, the total had risen to $1.25 to $1.35 billion, with about 25 per cent going to that category. The bulk of the technology spend is replatforming its property management, central reservations and loyalty systems, which are being rolled out to a meaningful number of hotels during 2026. In June it began a phased rollout of Ask Bonvoy, a natural language search experience on Marriott.com and the Bonvoy app, and it is working with Google on its AI Mode travel product and with OpenAI on its advertising pilot, both at an early stage.
What This Means For Everyone Else
A billion dollars of transformation spending, with core systems being replaced mid-flight, sets the pace for the sector. Groups without that budget cannot answer it with a matching platform programme. They can answer it by making the estate they already own behave as one, which is a far cheaper problem and a faster one.
The Evalueserve Edge
We work in industries where a data assumption has an audited consequence
Loyalty breakage is an estimation assumption inside operational data with a disclosed effect on a published balance sheet. That is not a marketing analytics problem. It is the kind of problem our data and AI practice was built on, across twenty-five years in banking, capital markets, insurance and investment management, where covenant monitoring, credit spreading, model validation and index calculation all carry the same requirement: a number a regulator or an auditor can follow back to its source.
The technical answer is deterministic retrieval rather than approximate recall. We ground AI answers in graph-structured, verified data, so every entity mapping can be traced and every figure is explainable on demand. The same discipline transfers directly to loyalty liabilities, revenue recognition, distribution cost and GOPPAR reporting, because the question is identical. Can you prove it.
Four loops where connection pays
The valuable opportunities are not isolated use cases. They are loops, in which a signal produces a decision, the decision produces an action, and the outcome improves the next signal. Four of them cover most of where hotel decisions live.
A loop closes only when the signal, the action, the accountable human and the measure are designed together
Source: Evalueserve analysis.
1. Commercial. Sense Demand, Then Shape It.
Booking pace, competitive rates, event compression, channel economics and cancellation patterns already exist as signals. They typically reach decision-makers at different times, through different tools, in different formats. The loop closes when a material change is detected, its context is assembled automatically, an explained recommendation reaches the revenue process, and the revenue manager retains the decision. The measure is not forecast accuracy. It is the elapsed time between a signal becoming available and a rate decision being made, and the share of that time spent reconciling rather than judging. This is the use case hotel operators themselves ranked highest.
2. Guest. Understand, Then Coordinate The Response.
A guest experiences one brand, not a CRM plus a survey platform plus a property system. Synthesis across booking, loyalty, preference, service, review and survey data is the easy part, and it is not personalization. The loop closes when insight changes a response. A recurring complaint becomes an operational intervention with an owner. A service-recovery signal reaches a named employee with context. A cross-property pattern informs a decision about service design, training or investment. Customer data management was the single largest area of planned AI expansion in the h2c study, at 50 per cent, which suggests the industry already knows this.
This is also where primary research earns its place. Transactional data records what happened and rarely explains why. Surveys and interviews test hypotheses and surface motivation. AI accelerates coding and synthesis at a scale manual analysis cannot reach. Humans still set the questions and decide what the finding means.
The Evalueserve Edge
Research is not an adjacent service for us. It is the origin of the firm.
Most technology partners can process the guest data you already hold. Fewer can go and get the evidence you do not hold. Evalueserve was built as a research business before it was an engineering one, and that combination is unusual: primary and voice-of-customer research, competitive and market intelligence, and survey design and analysis, run by domain specialists and industrialised with AI rather than replaced by it.
In practice that means the guest loop can be closed with both halves present. Behavioural data tells you what changed. Designed research tells you why, and whether the intervention worked. Recognition in Forrester Wave evaluations for market and competitive intelligence and for customer analytics reflects where that capability sits.
3. Operating. Detect Friction, Then Resolve It.
Hotel operations accumulate small exceptions with large aggregate effects: inventory imbalance, maintenance delay, staffing gaps, invoice mismatch, standards drift. Reporting explains these after the fact. The loop closes when a signal is caught earlier, joined to context, and routed into a controlled workflow with clear ownership. The measure is time to resolution and recurrence rate, not volume of alerts.
Procurement follows the same logic. Hospitality businesses generate substantial supplier, contract, invoice, category and spend data, and the opportunity is rarely another platform. It is better intelligence flowing into and around the systems already in use: classifying and enriching supplier data, reading contracts, linking spend to risk, surfacing exceptions and giving sourcing teams external market context. The transactional platform stays the system of record. The intelligence layer improves the quality and usability of the decision.
McKinsey and Skift identify automated room allocation, predictive maintenance, housekeeping task management and menu engineering as candidates here. Their report frames these as potential, using conditional language throughout, and claims no deployed outcome. Treat any demonstrated agentic result in hotel operations as a claim to be evidenced. Ask for the primary source and the measurement method. Including for the figures in this article.
4. Market. Read The Outside, Then Adapt.
Openings, competitor moves, destination dynamics, distribution shifts, regulation and investment activity are high-volume and uneven. More monitoring does not produce more intelligence. The capability required is judgement about which developments are material, how reliable the evidence is, and which part of the business must respond. AI extends coverage and speed. Domain experts decide what matters. The measure is decisions changed, not reports produced.
The Evalueserve Edge
A squad, not a hand-off
Loops fail at the seams, so we staff to remove them. A single squad carries a use case from framing to production and then keeps it running: a domain engineer for process and scope, a data engineer for pipelines and foundations, a data modeller for ontology and semantics, an AI engineer for prototyping and agent build, and a change consultant accountable for adoption and measured impact.
The same model then shifts into operate, with distinct run, improve and adopt disciplines on a daily, weekly, monthly and quarterly cadence. That matters because the loop is the asset. Value compounds only if someone owns quality, cost, adoption and reuse after go-live, which is precisely where most AI programmes quietly stop.
Governed action is a dated obligation, not a maturity goal
When AI moves from advising to acting, governance stops being a principle and becomes a compliance calendar. For any hotel group operating in Europe, some of those dates have already passed.
Transparency duties are live now. The high-risk deferral is time to prepare, not a pause.
Source: Regulation (EU) 2026/1744 of 8 July 2026 (Digital Omnibus on AI), Official Journal L, 24 July 2026, in force 27 July 2026, amending Regulation (EU) 2024/1689 (Artificial Intelligence Act).
The EU AI Act's Article 50 transparency obligations became applicable on 2 August 2026 and were not deferred. Where a guest interacts with an AI system, or where synthetic content is generated, disclosure duties attach. For systems already on the market before that date, the Article 50(2) marking and detection requirements apply from 2 December 2026.
The high-risk obligations did move. Under the Digital Omnibus on AI, adopted as Regulation (EU) 2026/1744 on 8 July 2026 and in force from 27 July 2026, obligations for standalone Annex III high-risk systems shifted from 2 August 2026 to 2 December 2027, and for AI embedded in regulated products under Annex I to 2 August 2028. This matters for hospitality more than it first appears, because AI used in employment contexts can fall within Annex III, and hotels are labour-intensive businesses experimenting with AI in recruitment, scheduling and workforce management. The deferral buys time for conformity assessment, technical documentation and human-oversight design. It is not a pause. Penalty exposure runs to €35 million or 7 per cent of turnover for prohibited practices, and €15 million or 3 per cent for transparency and high-risk breaches.
Then there is a live precedent specific to this industry.
In December 2024 the US Federal Trade Commission finalized an order against Marriott International and Starwood following three data breaches between 2014 and 2020 affecting more than 344 million customers worldwide. Marriott made no admission of liability and stated that many of the relevant privacy and security enhancements were already in place or in progress. The obligations are the instructive part.
|
Obligation under the FTC order
|
Design consequence for a data or AI team
|
|---|---|
|
Comprehensive information security programme, certified to the FTC annually for twenty years
|
Control evidence must be producible on demand, not reconstructed at audit
|
|
Personal information retained only as long as reasonably necessary
|
Retention limits become a constraint on context stores, feature stores and evaluation datasets
|
|
Deletion on request, tied to an email address or loyalty rewards account number
|
Every downstream copy of a loyalty identifier needs traceable lineage
|
|
Review of loyalty accounts on request and restoration of stolen points
|
Loyalty ledger integrity is a customer-facing legal commitment
|
Source: US Federal Trade Commission, order finalised 20 December 2024. A separate settlement with 49 states and the District of Columbia carried a $52 million civil penalty. Design consequences are Evalueserve analysis.
For a Chief Data Officer the implication is direct. Data minimization and deletion obligations attach to precisely the identifiers that personalization and agentic workflows depend on. Retention limits are therefore a design constraint on context stores, feature stores and evaluation datasets, not a legal footnote to be handled later. A loyalty identifier that must be deletable on request cannot be quietly duplicated into six downstream systems and an embedding index with no lineage.
None of this requires inventing a governance model. The NIST AI Risk Management Framework, published in January 2023 and organized around Govern, Map, Measure and Manage, provides a common vocabulary, and its Generative AI Profile of July 2024 extends it to twelve risk categories including confabulation, information integrity, data privacy and information security. It is voluntary. Its value is that it produces a documented, auditable position rather than a defensible-sounding one.
The Evalueserve Edge
Governance designed in, and evidenced continuously
Retrofitting controls onto a working agent is expensive and rarely convincing. We design the control envelope at the same time as the workflow: enterprise safety, data protection and risk controls; regulatory adherence, model auditing and error mitigation; approval gates, exception handling and escalation paths; and cross-functional oversight bringing the CFO, CIO and business owners into the same forum.
Evaluation, observability and audit evidence then run continuously rather than at review points, so the question of whether a system is behaving is answered by logs and evals rather than by assurance. In an industry now holding dated statutory obligations and twenty-year consent decrees, that is not overhead. It is the condition for letting AI act at all.
Who does what, and where the thin layer is
No single type of provider covers this ground, and the useful question is not who is best but what each layer is for.
Every layer is necessary. None is designed to own the middle work.
Source: US Federal Trade Commission, order finalised 20 December 2024. A separate settlement with 49 states and the District of Columbia carried a $52 million civil penalty. Design consequences are Evalueserve analysis.
Hyperscalers and model providers supply the substrate: compute, foundation models, managed AI services, identity, regional data controls and security posture. Capability at this layer is improving quickly and is broadly available to everyone.
Hospitality platforms, meaning property management, central reservation, revenue management, CRM and point of sale systems, hold the systems of record and the transactional integrity of rate, inventory, folio and profile. They own the surfaces where action lands, and they are adding AI features directly.
Niche hospitality technology vendors bring deep capability in specific jobs: revenue management, reputation, upselling, housekeeping, energy, labour scheduling. For a defined problem they are frequently the fastest route to value and often the best in category.
Strategy firms contribute framing, portfolio choices, operating-model and organizational design, and the board-level conviction required to fund multi-year change.
None of them, by design, is responsible for the sustained middle work: making the data trustworthy enough to act on, encoding the decision logic that currently lives in an experienced manager's head, integrating into the workflow where the action occurs, proving output quality repeatedly, and then operating the result long enough for it to improve.
That is not a claim that other providers cannot do this work. Many do. It is an observation drawn from what hotel leaders report as their own constraints: expertise at 62 per cent, unclear strategy at 51 per cent, integration at 45 per cent, and a fully integrated technology stack at 11 per cent. Those are not gaps in available technology. They are gaps in sustained, domain-specific execution, and they persist because that work is unglamorous, continuous and hard to package.
How Evalueserve Closes The Gap
Our position is that this middle layer is the work, and that it is done around a client's existing technology rather than in place of it. Start with domain. Scale with AI.
That is delivered as six connected service layers, and a client can enter at any one of them and scale through the same operating model.
|
Layer
|
What it delivers
|
|---|---|
|
Domain-led AI advisory
|
Opportunity discovery and prioritization, readiness assessment, target operating model, phased roadmap, and financial-grade ROI baselines with stage-gate reviews
|
|
Data platforms
|
Lakehouse and warehouse build, migration, ingestion from existing systems, data quality, catalog, lineage and policy, master data, semantic layer, knowledge graph and retrieval
|
|
AI platform layer
|
Model gateway and orchestration, LLMOps and MLOps, prompt and context operations, evaluation, safety and guardrails, telemetry, cost and latency operations
|
|
Agentic applications
|
Role-specific copilots for commercial, service and operations teams, tool and API actions inside business systems, reasoning, memory and release evaluations
|
|
Agentic workflows
|
Process re-engineering rather than automation of the status quo, multi-step execution, human approvals and exception handling, outcome telemetry against business KPIs
|
|
Build and operate
|
Managed run, incident and SLA management, continuous quality and model improvement, adoption programmes, and value-realization reporting
|
Three things distinguish how that is staffed and run.
Domain first, and domain in the room. More than a thousand data and AI specialists sit within a global expert base of over five thousand people across forty-five countries, and the people framing the problem are the people who understand how the decision is made.
Research and engineering in one team. The ability to generate primary evidence, not merely process existing data, is what turns a guest-experience dashboard into an explanation of why behaviour changed.
Built to be operated. Reusable platforms, accelerators and solution patterns shorten time to value, and strategic partnerships with Google Cloud, OpenAI and Databricks mean the substrate is enterprise-grade from the outset. But the differentiator is the operate discipline that follows go-live, because a decision loop only compounds if someone owns it.
How To Start, In One Quarter
- Pick one recurring decision with a named owner and a number attached to it. Rate release for a compression date. Service recovery for a repeated complaint type. Not a use case. A decision.
- Measure the assembly cost. From the moment the relevant signal existed somewhere in your estate to the moment the decision was made, how long elapsed, and what share of that time was spent gathering and reconciling rather than judging? This single measurement usually settles the business case.
- Write down the decision rule. Interview the person who makes it well. Most of the value in an intelligent workflow sits in tacit logic you are about to discover is undocumented.
- Define the control envelope before building. What the system may do unattended, what requires approval, what must escalate, and what is logged for audit. Name the accountable human.
- Check the compliance surface now. Does the workflow touch guest personal data, an employment decision, or a guest-facing AI interaction? If so, identify the specific obligation and its date before code is written, not after.
- Set the measure before you start, and hold back an answer key, so you can tell whether the output was right rather than merely plausible.
- Then scale through the same model. A proven loop, with its data foundation, controls and adoption pattern already established, is the cheapest possible starting point for the next one.
A two to four week opportunity sprint produces the use-case portfolio, value case and roadmap. A data foundation launch or an agentic pilot follows in six to twelve weeks. An operate squad then carries the run discipline. Small entry, production controls from day one, operated long enough to stick.
The Advantage Is In The Connections
Hospitality's next advantage will not come from owning more AI tools. Access to models and platform features is converging, and quickly.
It will come from connecting four things that currently sit apart: fragmented data, domain context, human judgement and governed action, across commercial, guest and operating workflows. That layer is harder to buy and harder to copy, and it is where the eleven hours a week are currently going.
The industry does not need persuading that AI matters. It needs to get from trusting AI at 6.6 to relying on it at something better than 4.7. That is a data, context and control problem, and it is solvable this year.
Hospitality has always been a human business. Closing this gap does not change that. It decides whether the people delivering it spend their week assembling the picture or acting on it.
Where is your intelligence getting stuck?
Pick one decision-heavy workflow, measure how much of its cycle time is spent reconciling data rather than making the call and bring us the number. That single measurement is a better starting point than any technology assessment, and it is the first thing we would ask for. Let us take one loop from decision to production, with the controls and the measure-built in.
