The Digital Transition got the data ready; the decision it was meant to improve would start with its formulation

Evidence markers: [V] verified · [D] derived · [E] estimate with an explicit assumption · [H] unproven hypothesis

Evidence markers: [V] verified against the primary source · [D] derived, with the calculation shown or in the supporting evidence book · [E] estimate with stated assumptions · [H] unproven hypothesis. Citations of SF-3, DT-2026-001, DS-2026-004, PE-1, and the storage Perspective inherit the marker those pieces assigned; PE-1 and the storage Perspective are still in production and not certified. Every figure has its entry, with URL, date of access, and method, in the supporting evidence book.

A company that invested in controlling, making available, standardizing, and governing its data built the physical infrastructure without which a Decision System would have nothing to run on [H].  Having the data would not guarantee that its business decisions are formulated, solved, governed, and ready to be reopened when the state changes [H]. The Analytical Transition would be the second stage. In it, the company would institutionalize that capability on the infrastructure it already paid for and then choose, starting from the problem, the analytical methods that solve it [H].

Six of the 12 headings in this piece, those of sections 2, 5, 6, 7, 9, and 11, are InsightIA hypotheses, and that is why they are written in the conditional.  The other 6 rest on definitions, on open public sources, or on material InsightIA has already published. Every claim carries its marker in the text, and every figure has its entry in the supporting evidence book.

0 · Structure of this note

Sections 1 and 2 define the Digital Transition and the formulation of decisions. Sections 3 and 4 examine declared investment and analytical occupations. Sections 5 to 7 propose responsibility, use cases, and intellectual infrastructure. Sections 8 and 9 relate analytical families to problems. Sections 10 and 11 present examples and the order in which technology is chosen. Section 12 gathers questions for checking the argument. Facts, derivations, estimates, and hypotheses are kept separate.

1 · The Digital Transition left data controlled, available, standardized, and governed, and that physical infrastructure is the condition for any better decision that relies on data

The Digital Transition delivers 4 things, and each is recognized by what it makes possible.  Control establishes who can see and change each piece of data. Availability puts data within reach of whoever needs it, when they need it. Standardization makes the same quantity mean the same thing in every area. Governance establishes who decides about the data and under what rules. The second edition of the international standard for data governance, ISO/IEC 38505-1, was published in August 2026 [V]. It applies the governance model of ISO/IEC 38500 to data [V]. It guides the organization's governing bodies toward the "effective, efficient, and acceptable use" of data [V].

This piece does not estimate how much was invested in the Digital Transition.  The only official series with a published method that this version located measures software investment in the United States. That category is broader than data and belongs to another country. Using it as the size of the investment discussed here would go beyond what it measures. For that reason it stays in the supporting evidence book and outside the argument.

This piece takes that investment as given and does not dispute it.  Without controlled and governed data, a Decision System has nothing to run on [H]. The next question is which business decision improved with that investment and who can demonstrate it.

2 · Having controlled and governed data would not guarantee that business decisions are formulated, because the objective function, the constraints, and the shared representation would not be found in any data

Data records what happened. The formulation states what the next decision is meant to achieve [H].  A bank with perfectly governed deposits knows how much it took in, at what rate, and in which segment. Before looking at the data, someone sets the margin the bank seeks, the liquidity buffer it holds, and the target it measures against. In the universal banking architecture published by InsightIA, that criterion is produced in a second-order system. It is the system of targets, budget, and allocation of ambition. Its output is the objective function of the other systems [H, SF-3 §8].

The boundary describes a missing guarantee [H].  Some Digital Transition programs already incorporate business rules, optimization, or automated decisions. This piece acknowledges that possibility. It holds that having the data and governing it, by themselves, leave the formulation, solution, and institutionalization of a decision unsecured [H].

The formulation has 3 parts, and none of them would be data [H]:

• The objective function: which quantity is maximized or minimized, and who set it

• The constraints: which limits cannot be violated (regulatory, physical, contractual, capacity) and which can be bought at a price

• The shared representation: the live model of the whole system, with the competing decisions, their constraints, and their interdependencies. They are held together as a single object on which decisions are made. If 2 areas consume the same quantity and each has its own version, a shared representation is missing [H, SF-3 §9]. What remains is a reconciliation that someone would have to do by hand, every time [H, SF-3 §9]

The table describes 2 complementary capabilities that can be developed in overlapping stages:

	Digital Transition · physical infrastructure	Analytical Transition · intellectual infrastructure
What it produces	Controlled, available, standardized, and governed data	Formulated, solved, and governed decisions
Its unit	Data and its lineage	The decision and its owner
Which question it answers	What happened, where is it, and who can use it?	What is decided, on what criterion, under what constraints, and when is it reopened?
When it is complete	When the data is reliable and accessible	When a decision changes its criterion and its result is recorded

Source: InsightIA formulation [H].

When the Digital Transition is limited to leaving data controlled and available, the capability to decide better is not guaranteed [H]. The Analytical Transition would institutionalize that capability, even if some of its components were already built during the first transition [H].

3 · In Gartner's 2026 surveys, 3 of every 4 functional leaders cite productivity as a key objective of artificial intelligence (AI), and 1 of every 5 AI projects in finance leans toward decision quality

The 3 Gartner publications from 2026 that this piece uses measure declared investment objectives, and none measures improvements achieved [D].

• Survey of finance leaders, July 20, 2026.  Gartner surveyed 204 finance leaders in March [V]. According to the body of the press release, 45% of AI investments in finance lean toward productivity [V]. Another line reports that 20% of AI projects in finance lean toward decision quality [V]. The 2 figures use different nouns: investments and projects [V]. They may not share a base, and they are not parts of the same total. The headline of the press release attributes them to chief financial officers [V]. The body attributes them to 204 finance leaders [V], and this piece cites the body

• Survey of functional leaders, September 1, 2026.  Gartner surveyed 1,303 respondents from January to April [V]. Productivity is a key objective for 75% of functional leaders and takes about 30% of functional AI spending on average [V]. The press release titles its figure 1 "Productivity Is the Top Outcome Functions Target with AI" [V]

• Generative AI adoption matrix, June 22, 2026.  Gartner classifies adoption in 7 use cases against 11 industries [V]. The use cases are to search, converse, analyze, draft, simulate, produce images, and program [V]. None of the 7 labels names a decision [D]. In the same figure, its architecture for agents closes the context layer with "context graphs for decision and workflow tracing" [V]. The labels, the key, and that phrase are printed in the image of the figure and not in the text of the article. Neither the article nor the figure states what the correlation measures, the sample, the period, or the method [D]. For that reason the matrix is cited as what Gartner publishes and never as a measurement of adoption

What these figures support is limited, and each claim states its base.  According to Gartner, among the 1,303 respondents surveyed from January to April, productivity is the outcome functions pursue most with AI [V]. Among the 204 finance leaders surveyed in March, 20% of AI projects lean toward decision quality [D]. That is a minority of projects. The 2 surveys have different populations, dates, and units [D]. Together they do not form a ranking of objectives [D]. The press releases leave the geography of the 2 surveys unspecified. For that reason the evidence is not credited to Latin America. That this way of investing leaves the decision the budget was meant to improve without a line item is an InsightIA hypothesis [H, DT-2026-001]. That outcome lies outside Gartner's findings.

4 · The data engineer and the data scientist contribute necessary occupational skills; who answers for a business decision being formulated, solved, and reopened is an institutional responsibility, distinct from the occupation

The 2 occupations are described here in the words of the official occupational classification of the United States, O*NET. Its resource center is sponsored by the U.S. Department of Labor.

• The data engineer.  O*NET lists "Data Engineer" among the titles reported by incumbents of the database architects occupation [V]. It describes that occupation as designing "strategies for enterprise databases, data warehouse systems, and multidimensional networks" [V]. It is the occupation that builds the physical infrastructure. It decides where data lives, how it moves, how it scales, and how it is protected

• The data scientist.  O*NET describes this role as one who develops analytical techniques or applications "to transform raw data into meaningful information" [V]. It applies data mining, modeling, natural language processing, and machine learning [V]. It visualizes, interprets, and reports findings [V]. It is the occupation that turns data into models and findings

Neither of the 2 descriptions, nor the first 5 tasks O*NET publishes for each, includes formulating a business decision [D]. A decision is understood as formulated when it has an objective function and constraints [D].  That reading describes the classification and not companies. The omission of a task from an occupational description leaves unproven that organizations lack someone who performs it [D]. Both occupations do well what the classification assigns them, and the Analytical Transition needs both. Without the data engineer, data in production is missing. Without the data scientist, the model is missing.

The classification does contain an occupation that formulates problems, and it deserves an exact name.  The U.S. Bureau of Labor Statistics (BLS) describes the operations research analyst [V]. It presents this analyst as one who uses "mathematics and logic to help solve complex issues" [V]. Among the analyst's duties are identifying business problems and developing and testing quantitative models [V]. The duties also include writing memos with findings and recommendations "for managers, executives, and other officials" [V]. The list of duties ends with the recommendation to management [D]. Answering for the decision being solved in production and reopened when the state changes remains outside it [D].

The piece separates 2 things.  An occupation comprises the skills to formulate and solve problems. These 3 occupations have them in different measure. Institutional responsibility is who guarantees that the decision is formulated, solved, put into operation, leaves evidence, and is reopened when appropriate. The next section addresses that responsibility.

5 · The missing responsibility would guarantee that the decision is formulated as a problem with an objective function and constraints and is solved in production; the company would keep it as a capability, regardless of who exercises it

The company would need to institutionalize the capability to formulate, solve, govern, and reopen its business decisions on the data infrastructure it has already built [H].  This is the central thesis of the piece. The responsibility has 3 parts. Hiring more data engineers or data scientists does not, by itself, assign them [H]:

1. Formulate.  Write down who decides, what is maximized, which constraints apply, what can be bought at a price, and what representation the decision-makers share. This is work with the chief financial officer, the chief operating officer, and the area that executes, before opening the data

2. Select the analytical architecture.  Identify the families and methods the problem requires (section 8), taking into account its constraints, uncertainty, the available data, and the conditions of implementation

3. Answer for the solution in production.  Ensure that the decision runs at the cadence the business requires. Ensure that a record is kept of what was decided, under what state, and what happened next. Ensure that the decision is reopened when a variable crosses a declared threshold

This piece describes that responsibility by what it does, without proposing a position, because what matters is that the company retains it [H].  InsightIA's published architecture for universal banking poses a question. "If the people who do the analysis today disappeared, what would have to remain for the bank to keep making that decision?" [SF-3 §10]. That this capability remains when the people change is the design hypothesis of this piece [H]. Developing it within the organization is what InsightIA calls Organizational Upskilling.

6 · A use case that does not name the decision it must improve could produce valuable information without changing the way decisions are made

The 2 paths start with different questions [H].  The first asks what can be done with the available data. Its natural product would be an indicator, a dashboard, or a forecast that someone consults. The second asks which decision, with which owner, and on what criterion, and what data that decision lacks. Its intended product is a decision that changes its criterion and leaves a trace. Neither path determines the result. A use case that originates in the data can end up changing a decision. One that originates in the formulation can stall at a dashboard if its implementation fails [H]. The second path adds that someone will have to answer for which decision improved. Decision Thesis DT-2026-001 translates that proposal into the budget [H]. Each AI investment line item would have to name the decision it was meant to improve and who answers for it [H]. It would also have to name the expected economic result and the baseline against which the improvement will be measured [H].

The public evidence is consistent with this hypothesis and with others.  Gartner reported a survey of 644 respondents from organizations in the United States, Germany, and the United Kingdom in the fourth quarter of 2023 [V]. It reported that 48% of AI projects reach production and that moving from prototype to production takes 8 months [V]. The main obstacle to adopting AI is the difficulty of estimating and demonstrating its value [V]. Another survey had 432 respondents from the United States, the United Kingdom, France, Germany, India, and Japan, none from Latin America [V]. It was conducted in the fourth quarter of 2024 [V]. Among leaders of organizations with low AI maturity, 37% named finding the right use case among their main barriers [V]. The same survey records a rival explanation [V]. Data availability and quality was among the main challenges for 34% of those leaders [V].

What these figures do not say matters as much as what they say.  They measure AI projects, rather than analytical projects in general. They leave unspecified why the rest did not reach production. Nor do they distinguish a use case that originates in the data from one that originates in the decision [D]. The data is consistent with the reading that the difficulty of demonstrating value comes from use cases without a formulated decision. The data leaves that reading undemonstrated [H]. It would be refuted by a portfolio of use cases that did not name their decision and still changed criteria with a recorded result. That portfolio would have to do so in the same proportion as the use cases that did name it.

7 · The intellectual infrastructure of the Analytical Transition would be the 6 components that institutionalize a Decision System, and all of them would be built on the physical infrastructure the company already paid for

The 6 components come from the universal banking architecture InsightIA published [SF-3 §10]. Here they are generalized as hypotheses to any industry [H].  The table shows which part of the physical infrastructure each would be built on:

Component of the intellectual infrastructure	What it is	Part of the physical infrastructure it would be built on
Formulation of the decision	Explicit objective function and constraints, written down and kept live	The catalog and the standardized definitions of the data
Shared representations	With an owner, a version, and periodic calibration	Standardized, governed data: a single version of each quantity
Joint resolution mechanism	The procedure that produces the answer, with its assumptions stated	Computing and access to data at the decision's cadence
Decision and outcome log	What was decided, under what state of the world, and what happened next	Data storage and lineage
Governance of the decision	Who can change what, with what evidence, and what is reserved for judgment	Access control and the data governance rules
Reopening triggers	Declared thresholds on monitored variables, so that the decision is reopened when the state changes	The continuous availability of those variables

Source: the 6 components, SF-3 §10; the correspondence with the physical infrastructure, InsightIA formulation [H].

The table shows why the Analytical Transition would arrive as a second stage and not as a replacement [H].  Each intellectual component rests on a physical one. The decision and outcome log needs storage and lineage. The triggers need variables available in time. No physical component produces an intellectual component by itself. Lineage records where a piece of data came from and leaves unrecorded what was decided with it [H].

8 · Each analytical family solves a class of problem and asks something different of the data: optimization, causal inference and econometrics, simulation, time-series forecasting, pattern learning, rules and heuristics, and language models and agents

Family	What it is for	What problem it solves	What it is not for	What it asks of the data
Optimization	Choosing the best alternative within a feasible set	Allocation of resources, prices, production schedules, portfolios	Without a written objective function and constraints, it cannot be set up	Cost, capacity, and constraint parameters, at the decision's cadence
Causal inference and econometrics	Estimating what changes if one intervenes	Effect of a price, a policy, or a campaign; elasticities	Answering without assumptions about what was not observed	Variation in the intervention and covariates that make it possible to sustain the identification assumptions
Simulation	Reproducing the behavior of a system under scenarios	Systems with interactions, queues, rare events, or physical dynamics	Choosing the best alternative on its own: it compares the alternatives it is given	Parameters and distributions of each component, and validation against what is observed
Time-series forecasting	Anticipating a quantity over time	Demand, flows, prices, occupancy	Deciding: it delivers an expected value and its uncertainty, and the decision is made separately	Sufficient history and knowledge of the factors that move the series
Pattern learning	Classifying, detecting, or predicting from examples	Credit risk, fraud, maintenance, segmentation	Distinguishing, on its own, correlation from the effect of an intervention	Labeled examples in volume, representative of the conditions in which it will be used
Rules and heuristics	Applying a known criterion in an auditable way	Policies, thresholds, compliance, low-frequency cases	Adapting to a change of state unless someone rewrites the rule	The variables the rule uses, reliable and on time
Language models and agents	Reading, summarizing, drafting, and chaining steps across text and tools	Document extraction, user interface, task orchestration	Guaranteeing accuracy without verification: they can produce false content that appears certain	Reliable documents and context, and a mechanism to verify their outputs

Source: definition and mechanism of each family; the capabilities and limits with a primary source carry their marker in the text that follows.

Four of the limits in the table have a primary source.

• Optimization.  The standard formulation of an optimization problem minimizes an objective function subject to constraint functions. Its typical applications include resource allocation, scheduling, and portfolio operations [V, Boyd and Vandenberghe]. It is the family that depends most on the formulation. Without a written objective function, the problem to be solved is missing [D]

• Causal inference.  Hernán and Robins discuss the example study of their chapter 2 in their reference text [V]. They write: "we are generally unable to determine whether exchangeability holds in our study" [V]. Exchangeability is the assumption that allows a difference to be read as an effect. That data alone does not demonstrate it is the general reading this piece takes from that passage [H]. For that reason this family would ask the business for a question and a design before more data [H]

• Forecasting.  Hyndman is a coauthor of a reference text on forecasting [V]. He writes that the predictability of a quantity depends on how well the factors that move it are understood [V]. It also depends on how much data is available and on whether the forecast can affect what is being forecast [V]. The decision that uses the forecast can change the result, as when a price moves demand. When that happens, the forecast alone is no longer enough and the decision moves to another family [H]

• Language models.  The U.S. National Institute of Standards and Technology (NIST) published its generative AI profile in July 2024 [V]. Document NIST AI 600-1 names confabulation as a risk: "the production of confidently stated but erroneous or false content" [V]

The descriptions of simulation, pattern learning, and rules come from their definition and mechanism. This version attributes to them no capability or limit with a source. The table omits vendor brands. No family is presented as superior to another outside a specific problem.

9 · The problems that enter a Decision System would have the same structure in every industry (decision-maker, objective function, and shared constraints), and the analytical architecture would be selected, on solid grounds, from that structure

The structure would have 4 elements, and writing them down would be the starting point for choosing methods [H].  They are who decides, what that person seeks, which constraints the decision shares with other decisions, and at what cadence it is decided again. The following questions guide the selection, and each points to a family:

1. Is the choice among alternatives under constraints? Optimization

2. Is it necessary to know what changes if one intervenes? Causal inference and econometrics

3. Does the result depend on interactions, queues, or rare events that are not observed often? Simulation

4. Is it necessary to anticipate a quantity? Time-series forecasting

5. Is the task classifying or detecting in volume? Pattern learning

6. Is the criterion already known, and must it be auditable? Rules and heuristics

7. Is the input or the output text, or does the task chain steps with tools? Language models and agents

The formulation establishes what has to be solved. The final choice also depends on the mathematical structure of the problem, on uncertainty, and on the available data. It depends as well on the explainability required and on computing [H].  A real decision would almost always combine several families, in the order the problem dictates [H]. In a bank's lending, risk models estimate the probability of default. The price and the limit are set with that estimate as an input. They are set within the capital charge issued by the balance-sheet core [H, SF-3 §4, §9]. Those models can be pattern learning, and the decision that uses them is optimization [H]. The family that appears first in the data flow can be the last in the order of the decision.

10 · Banking, energy, and hospitality already have decisions formulated by InsightIA, and Oil & Gas, Industrial, and Infrastructure repeat the structure with 1 example each

Each row is written as decision, decision-maker, objective function, and analytical family. The 6 Decision Systems functions appear in the last column.  The banking, energy, and hospitality rows come from pieces InsightIA has already published or has in production. The Oil & Gas, Industrial, and Infrastructure rows are derivations of this piece, with no real case behind them [H].

Sector	Decision	Decision-maker	Objective function and main constraints	Analytical families	Function
Financial System	Mix and price of liabilities, liquidity buffer, duration, and allocation of balance-sheet capacity	The bank's asset and liability management unit	The one set by the targets system; regulatory liquidity and capital constraints	Optimization, with forecasting of deposit behavior	Capital Allocation
Financial System	To whom to originate credit, with what amount, price, and limit	The credit area, within the capital charge issued by the core	Margin adjusted by the transfer curve and the capital charge	Pattern learning as an input; optimization of price and limit	Capital Allocation
Financial System	Through which channel, how often, and with which offer to reach each customer	The commercial area	Customer value, with installed service capacity as a shared constraint	Causal inference, pattern learning, optimization of the allocation	Latent Growth
Financial System	Where to set fraud thresholds, and how much to pay for the evidence that calibrates them	Whoever approves the fraud policy	Total economic cost: loss, customer friction, and investigation capacity	Pattern learning; causal inference with a controlled sample	Execution on Decision Systems
Financial System	Appetite, allocation of ambition among businesses, and plan assumptions	Senior management and the board	Produces the objective function of the other systems; reopened by performance	Scenario simulation and multi-objective optimization	Institutional Systems
Power, Energy & Environmental Transition	Which combination of contracts, capacity, and backup to buy to cover the critical hours	The buyer that awards: the distribution company or the retailer	Total cost of covering its critical hours, including replacement if a contract fails	Hourly simulation and optimization of the contract portfolio	Capital Allocation
Power, Energy & Environmental Transition	Which storage configuration to build: technology, duration, node, and date	The developer, the fund, and the financier	Return compared with the alternatives, with system benefit and capturable revenue measured separately	Hourly and nodal simulation of the system, and financial evaluation	Capital Projects
Hospitality	Whether to reposition an asset, and against which source market	The owner or the asset committee	Asset performance against plan, with the decision reopened when its reference series diverge	Time-series forecasting and reopening rules triggered by divergence	Execution on Decision Systems
Oil & Gas	How to allocate the drilling and maintenance budget among fields	The operator's investment committee	Value of production, with budget, available rigs, and production commitments as constraints	Optimization with simulation of price and production scenarios	Capital Allocation
Industrial	How much to produce at each plant and with which energy, month by month	The chief operating officer of a cement or steel company	Contribution margin, with kiln capacity, inventory, and energy contracts as constraints	Demand forecasting and optimization of the plan	S&OP
Infrastructure	Which sections of a concessioned highway network to work on, and when	The concessionaire	Life-cycle cost, with the service levels of the concession title and the budget as constraints	Deterioration forecasting and optimization of the program	Capital Projects

Source: Financial System, SF-3 §4 and §7–§9; Power, Energy & Environmental Transition, Perspectives PE-1 and storage, v0.10, checked on October 15 and not certified; Hospitality, Decision Signal DS-2026-004; Oil & Gas, Industrial, and Infrastructure, InsightIA derivation [H]. The assignment to each function and the choice of families are InsightIA formulation [H].

Two boundaries in the table need explanation.  In Capital Projects, the configuration of a physical system is decided: which asset, of what size, where, and when [Fundamentos de InsightIA (InsightIA Fundamentals)]. In Capital Allocation, finite capital is distributed among alternatives that compete for it [Fundamentos de InsightIA]. That is why the storage configuration goes in the first and the buyer's contract portfolio in the second. Institutional Systems is the sixth Decision Systems function, the one that constitutes the institution as a decision system. In banking, it would correspond to the system that sets the targets of the others [H]. The table also shows that the families cut across the functions. Optimization appears in 5 of the 6 [D]. No sector, by itself, determines a family [H].

Behind InsightIA's rows are public facts that show why the formulation carries weight.  In banking, the liquidity coverage ratio of Mexican commercial banks has a required minimum of 100% [V, SF-3]. It is a constraint that must be written into the formulation before the deposit price is set. In energy, the Colombian energy product for the 18:00–22:00 peak was offered [V, PE-1]. In the 2026 long-term auction it went unawarded because of the regulatory competition limit [V, PE-1]. Bids covered 9.3% of the quantity requested [V; D, PE-1]. The hours the buyers requested were left without a long-term contract in that product [D]. The hybrid product in the same auction did contract delivery from 18:00 to 22:00 [V, PE-1]. In storage, the arbitrage margin of a 4-hour battery at the Chilean Crucero bus was calculated [D, storage Perspective]. It fell over 12 months [D, storage Perspective, with the traceability of the figure still open]. The business case changes while the asset is in service. In hospitality, tourist arrivals in Tulum fell 8.6% cumulatively from January to July 2026 [D, DS-2026-004]. They fell 2.4 times faster than arrivals in their region [D, DS-2026-004]. The decision depends on which of several series anchors the plan.

11 · Advanced analytics or AI would be used in every Decision System, and the right method, which may not be AI, would come after the problem, as a consequence of the conversation about the decision

The 11 rows of the previous section use at least one advanced analytics method: optimization, simulation, forecasting, or pattern learning [D]. In none of them could the tool have been chosen before the decision was written down [H].  Several do not require AI in the strict sense. The plan for the hotel asset calls for forecasting and rules, without optimization or learning [H]. Pattern learning appears as an input in lending and as the basis of the policy in fraud. A Decision System is designed to solve the problem and uses AI only when the problem calls for it [H]. Choosing the tool first would force a search for a problem to fit it. Formulating first would allow the tool to be deduced [H].

Here the closing line of InsightIA's universal banking architecture becomes a thesis.  Which analytical family suits each system is deduced from the problem and comes afterward. Technology comes last, as a consequence of the conversation about the decision [H, SF-3 §12]. The 2 transitions would be ordered as follows. The digital transition gets the data ready. The analytical transition formulates the decision. The tool, already chosen by the problem, runs on what was built in the first [H].

12 · Whoever paid for the Digital Transition can check next Monday, with 3 questions, which business decision improved with it

This piece describes an architecture and leaves open how your company decides. You can check in your own company whether the argument applies:

1. Which business decision can you name that improved with your Digital Transition, with its owner and the baseline against which the improvement was measured?

2. Where is the objective function of that decision written down, and who can change it?

3. Of your analytics and AI use cases, which changed a decision criterion with a recorded result, and which ended in a dashboard that someone consults?

If the first answer takes time, what is missing would lie in the formulation of the decision, on data the company has already paid for [H].

What this piece does not claim

• That the Digital Transition was a mistake or a waste: it is the condition for everything the piece proposes

• That any occupation is redundant: the data engineer, the data scientist, and the operations research analyst do work the Analytical Transition needs

• That one analytical family outperforms another outside a specific problem

• How many companies have their decisions formulated, nor any data project failure rates without a published method

• Anything about any particular company

Sources

1. ISO · ISO/IEC 38505-1:2026, Information technology — Governance of data — Part 1: Application of ISO/IEC 38500 to the governance of data, August 2026, public abstract: https://www.iso.org/standard/87195.html

2. Gartner · Press release of July 20, 2026, survey of 204 finance leaders: https://www.gartner.com/en/newsroom/press-releases/2026-07-20-gartner-survey-shows-45-percent-of-cfos-say-their-ai-investments-lean-towwards-productivity-while-20-percent-say-these-investments-lean-towards-decision-quality

3. Gartner · Press release of September 1, 2026, survey of 1,303 respondents: https://www.gartner.com/en/newsroom/press-releases/gartner-survey-finds-only-22-percent-of-organizations-have-successfully-scaled-ai-across-multiple-business-units

4. Gartner · Top Emerging Adoption Trends for Generative AI, Vibha Chitkara, June 22, 2026: https://www.gartner.com/en/articles/emerging-adoption-trends-for-genai; the labels, the key, and the quotation in section 3 are read from the image of its figure, "GenAI use-case adoption trends across industries", code CTMKT_5230645 printed at the foot of the image

5. Gartner · Press release of May 7, 2024, survey of 644 respondents: https://www.gartner.com/en/newsroom/press-releases/2024-05-07-gartner-survey-finds-generative-ai-is-now-the-most-frequently-deployed-ai-solution-in-organizations

6. Gartner · Press release of June 30, 2025, survey of 432 respondents: https://www.gartner.com/en/newsroom/press-releases/2025-06-30-gartner-survey-finds-forty-five-percent-of-organizations-with-high-artificial-intelligence-maturity-keep-artificial-intelligence-projects-operational-for-at-least-three-years

7. O*NET · Database Architects, 15-1243.00: https://www.onetonline.org/link/summary/15-1243.00; Data Scientists, 15-2051.00: https://www.onetonline.org/link/summary/15-2051.00

8. U.S. Bureau of Labor Statistics (BLS) · Occupational Outlook Handbook, Operations Research Analysts, last modified August 27, 2026: https://www.bls.gov/OOH/math/operations-research-analysts.htm

9. Boyd and Vandenberghe · EE364a, Convex Optimization, Stanford, introductory lecture: https://web.stanford.edu/class/ee364a/lectures/intro.pdf

10. Hernán and Robins · Causal Inference: What If, Chapman & Hall/CRC, chapter 2: https://www.hsph.harvard.edu/miguel-hernan/wp-content/uploads/sites/1268/2024/04/hernanrobins_WhatIf_26apr24.pdf

11. Hyndman · "Why are some things easier to forecast than others?", Hyndsight, September 18, 2012: https://robjhyndman.com/hyndsight/hardforecasts/

12. NIST · NIST AI 600-1, Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile, July 2024: https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.600-1.pdf

13. InsightIA · Perspective SF-3 v0.9, Donde el precio interno no llega, las áreas de un banco pueden decidir por separado (Where the internal price stops, a bank's business areas can decide separately), October 14, 2026, certified; Decision Thesis DT-2026-001; Decision Signal DS-2026-004; Fundamentos de InsightIA, August 24, 2026; Perspectives PE-1 v0.10, with the round 4 review opinion, not certifiable and with 4 open evidence findings in sections this piece does not use, and storage v0.10, in production with open traceability reservations; what this piece takes from them was checked on October 15, 2026