Competing on Analytics: The New Science of Winning

1. 1Name the failure you are replacing: the Analytics 1.0 'order taker' pattern, where analysts sit in the back office, take requests, and return an answer a month later that the manager ignores. Your target is data as the guide to decisions, not decoration.
2. 2Move analysts out of the back office and into working relationships with the managers who own decisions. In 1.0 the tell of a broken culture is that 'managers didn't typically have a close relationship with quantitative analysts.' Assign analysts to decision owners, not to a request queue.
3. 3Set a combined objective for analytics: both shaping decisions and shaping products/services. The 3.0 posture treats data and analysis as standing behind decisions and behind new offerings (as with GE's sensor-driven service model or Monsanto's Climate Pro prescriptive planting).
4. 4Embed analytics into operational workflow rather than running them as one-off studies. The 3.0 example: inventory analytics are 'incorporated into a supply chain management system, so that the right number of products is always held in the warehouse'—not a separate analytics run.
5. 5Control the volume of descriptive reporting and push self-service to users, freeing analytical professionals for predictive and prescriptive work. Sophisticated 2017 companies 'try to control their volume and they try to get users (rather than analytical professionals) to create them.'
6. 6Import the experimental temperament from Analytics 2.0—an 'impatient, experimental culture' (Facebook's 'Move fast and break things')—but calibrate it, since that mode 'perhaps would not fit well within many large organizations.'
7. 7Prove value with measured payoff and require decisions to actually change. Calculate savings in 'typical analytical fashion' as UPS did with ORION, and reject the 1.0 outcome where results were produced but decisions still ran on 'intuition and gut feel.'

UPS's ORION driver-routing project is the culture in action. Some people in the company 'thought no computer could improve on what experienced UPS drivers already did so well'—the intuition-and-gut-feel resistance the culture must overcome. Rather than run a study and file it, UPS embedded a thirty-page algorithm into daily operations: ORION bases each of 55,000 drivers' routes on the actual pickup and drop-off addresses, gives a different route every day, and will eventually re-route in real time on weather, traffic, and pickup calls. Leadership committed real money (several hundred million dollars over about a decade) and, in analytical fashion, quantified the return: roughly 7–8 fewer miles per driver per day, 100 million miles and 10 million gallons of fuel saved annually, and about half a billion dollars a year in labor and fuel. Because the output was operational and its payoff was measured, the organization acted on the data instead of deferring to driver instinct. The culture then extended the pattern, applying analytics 'in many other facets of our business' beyond the physical network.

• If analysts report requests through a queue and return answers weeks later, then you have an 'order taker' culture—reassign them to decision owners so results get used.
• If a proposed analysis produces a report but changes no decision or workflow, then treat it as decision 'support' theater and either embed it in an operational system or kill it.
• If descriptive reporting volume is growing, then cap it and hand routine visuals to self-service users so professionals focus on predictive and prescriptive work.
• If the initiative touches a large organization, then borrow 2.0 experimentation but not 'move fast and break things' wholesale—that Silicon Valley mode 'would not fit well' at scale.
• If leaders cannot state the measured payoff (as UPS did for ORION), then you have not yet made facts the guide to the decision—require the number before commitment.

• Producing analytics 'in spades' while decisions still run on intuition. Tell: managers praise the analysis but the choice they make is unchanged.
• Analysts staying 'in the back office' with no close relationship to managers. Tell: work arrives as tickets, not conversations, and turnaround is 'a month or so.'
• Confusing volume of descriptive reporting with an analytical culture. Tell: proliferating dashboards nobody uses and 'multiple versions of the truth' in spreadsheets.
• Treating analytics as a one-off study instead of embedding it. Tell: the model runs 'a separate analytics run' rather than living inside the operating system that acts on it.
• Mistaking better technology or more data for the point. Tell: heavy tool investment with little change in how aggressively data is exploited to create new or better ways of doing business.

• ▸Decision or workflow: the specific choice this analysis will change, and who owns it.
• ▸Current basis: is it made today on intuition/gut feel or on data?
• ▸Analyst-to-owner link: named analyst paired with named decision owner (not a queue).
• ▸Objective type: shaping a decision, a product/service, or both.
• ▸Embed point: the operating system/process the analytics live inside (not a standalone study).
• ▸Descriptive vs. advanced: what routine reporting moves to self-service; what predictive/prescriptive work this frees.
• ▸Experiment scope: what you will test fast, and the limit that keeps 'break things' safe at your scale.
• ▸Measured payoff: the quantified savings/benefit expected, stated ORION-style before commitment.
• ▸Action commitment: what the organization will actually do differently once the data comes back.

1. 1Name the one or few capabilities that are strategically central to how you make money — revenue management, customer loyalty, supply chain optimization, location selection, M&A screening, or hiring — rather than trying to be analytical at everything. The passages define an analytical competitor as one that 'selected one or a few distinctive capabilities' to base strategy on.
2. 2Confirm the chosen capability is a genuine point of differentiation: check that older bases (geography, protective regulation, proprietary technology, product breakthroughs) are eroding in your industry, leaving execution and decision quality as what remains.
3. 3Instrument the process to generate the data. UPS 'kludged together' regional cellular networks into a proprietary US-wide network to gather information on the movement of every package, which fed 'one of the largest data warehouses in the world.' No data collection at the point of activity, no analytics.
4. 4Move up the maturity ladder from descriptive to predictive to prescriptive. UPS went from pushpins on maps, to predictive Package Flow Technologies (2003), to prescriptive ORION that tells drivers the exact route to run.
5. 5Quantify the payoff in the metric that matters to the capability and track it 'in typical analytical fashion': ORION shortens each route 7–8 miles, saves 100 million miles and 10 million gallons of fuel yearly, and yields ~$400M+ (calculated near half a billion) in annual savings.
6. 6Budget for a long, expensive build and defend it against skeptics. ORION took about a decade to develop and roll out across 55,000 drivers, cost several hundred million, and had internal doubters who 'thought no computer could' beat their best drivers.
7. 7Extend the same analytics outward: open predictive engines to customers so they interact with operations and personalize their experience, turning the data you collect into something 'as valuable to the customer as the package itself.'
8. 8Treat the soft factors as the hardest part — analytical culture, leadership commitment, and focusing analytics on a pressing business problem. The authors note these were the toughest to pull off a decade ago and remain so.

UPS chose delivery-route optimization as its distinctive capability. Historically managers laid out routes with pushpins on maps, and drivers ran roughly the same route daily, delivering demanding commercial stops before residential ones. The e-commerce shift meant residential customers wanted control over timing, so drivers criss-crossed the same paths, wasting time and fuel — and leadership foresaw a point where even great drivers 'could not wring one more second of efficiency out of their routes.' UPS built the data foundation first (a proprietary national cellular network feeding a massive data warehouse), then layered predictive Package Flow Technologies (2003, cutting 85 million miles/year), then the prescriptive ORION model. ORION is a thirty-page algorithm computing the optimum daily route for each of 55,000 drivers, now varying routes daily and eventually reacting in real time to weather, pickup calls, and traffic. It shortens each route by 7–8 miles, saving 100 million miles and 10 million gallons of fuel annually and generating over $400 million in yearly savings. The authors call it possibly 'the largest commercial analytics project ever undertaken.'

• If a capability is central to how you make money AND you can perform it better than rivals through data and analysis, then make it the strategic basis; if it's not strategically central, keep it as ordinary support analytics, not a competitive bet.
• If your traditional differentiators (geography, regulation, proprietary tech, product breakthroughs) are being copied or eroded, then shift the basis of competition to execution efficiency and decision quality — where analytics wins.
• If descriptive reporting no longer moves the metric, then advance to predictive, and if prediction still leaves human judgment as the bottleneck, then go prescriptive (tell the operator exactly what to do, as ORION does).
• If the data you collect is as valuable to customers as your core product, then productize it and open your predictive engines to them.
• If you can't state the payoff as a calculated number tied to the capability's cost driver, then you haven't yet built the analytical case to justify or defend the investment.

• Complacency — 'staying the course' until winners become case studies. The tell: leaders argue that what built past success will sustain it, as film, newspaper, and video-rental incumbents did against digital, Craigslist, and Netflix.
• Spreading analytics thin across everything instead of picking one or a few distinctive capabilities. The tell: no single process where you're demonstrably better than anyone in your industry.
• Mistaking tools for strategy. The passage warns 'analytics themselves don't constitute a strategy' — the tell is a data-platform initiative with no named business capability it's meant to optimize.
• Skipping the data-collection layer and jumping to modeling. The tell: you want ORION-style routing but never built the instrumentation (UPS needed the handheld devices and proprietary network first).
• Underestimating the human and cultural work. The tell: the model is technically sound but stalls because leadership, culture, or a focused business problem is missing — the authors call these the hardest things, then and now.
• Letting internal skeptics kill the long build. The tell: 'no computer could' beat our experts — the same objection UPS faced before ORION proved out over a decade.

• ▸Chosen capability: name the one process/decision you'll be best at (e.g., route optimization).
• ▸Strategic centrality test: state how this capability drives revenue or cost.
• ▸Eroding differentiators: list which old advantages (geography, regulation, tech, product) no longer protect you.
• ▸Data instrumentation needed: what you must capture and where (e.g., every package's movement).
• ▸Maturity target: mark current and target stage — descriptive / predictive / prescriptive.
• ▸Payoff metric + calculation: the cost driver and expected quantified benefit (miles, fuel, dollars).
• ▸Investment & timeline: expected spend and rollout horizon (UPS: hundreds of millions, ~a decade).
• ▸Skeptic response plan: how you'll counter 'no computer could beat us' objections.
• ▸Customer-facing extension: how you'll open the analytics to customers as a product.

1. 1Identify a repetitive, high-volume operational decision currently made by intuition or a fixed rule (e.g., a driver following the same route daily). These are the highest-payoff targets because analytics scales across thousands of instances.
2. 2Build the algorithm around the actual data inputs, not the abstraction: ORION routes on the specific addresses where packages are picked up or dropped, not on a generic template. Ground the model in the operational variables that change each cycle.
3. 3Embed the analytic output into the operational system rather than running it as a separate analysis. Instead of a standalone forecasting run, incorporate it into the supply chain management system so the right inventory is always held automatically.
4. 4Quantify the payoff in the operating currency—miles, gallons, labor hours, dollars—so decisions are validated against measured savings (ORION: 7-8 miles shortened per driver, 100 million miles and 10 million gallons saved yearly, $400M+ in cost savings/avoidance).
5. 5Move from static to real-time: start with a route that changes daily, then plan to re-optimize in real time on live factors like weather, a pickup call, or traffic.
6. 6Where data volume and adaptation needs exceed human modeling capacity, shift to machine learning so the machine builds and tests thousands of models; use it to become more granular per customer/market and react to fast-changing data.
7. 7Extend beyond internal decisions: recognize the data flowing through your operations is often as valuable to customers as the core product, and package it as a data product or service.

UPS's ORION shows the full arc. Engineers built a thirty-page algorithm that determines the optimum route for each of 55,000 drivers every day, basing routing on the actual pickup/drop-off addresses rather than a fixed daily route. It took roughly a decade to develop and roll out, and UPS spent several hundreds of millions on it. The payoff was measured in operating terms: each driver's route shortened by an average of seven to eight miles, collectively saving 100 million miles and 10 million gallons of fuel a year, and UPS calculated about half a billion dollars a year in labor and fuel savings. ORION currently gives a different route each day and will eventually re-route in real time based on weather, pickup calls, or traffic. UPS also discovered that the network data itself was as valuable to customers as the packages, letting customers improve their own processes and making UPS a more valuable partner.

• If a decision recurs across thousands of near-identical instances (routes, inventory levels, service intervals), then embed analytics into the operational system rather than producing standalone reports.
• If the payoff can't be quantified in operating units (miles, gallons, dollars, labor), then treat the project as unproven—UPS validated ORION by calculating ~$500M annual savings.
• If your data volume and need for personalized, rapidly-adapting models exceeds what human analysts can hand-build, then shift to machine learning, which builds thousands of models in the time one artisanal model took.
• If model interpretability matters more than raw accuracy for the decision, then favor hypothesis-based analytics, because machine-learning models are often not transparent to their human users.
• If the data flowing through your operations would help customers make better decisions, then productize it—GE's prediction-of-need service, Monsanto's Climate Pro, and Optum's $67B data business all did this.

• Treating analytics as 'decision support' that stays in the back office—the tell: analyses arrive a month later, the manager doesn't understand the method, and the result is ignored while the decision is still made on gut feel.
• The 'order taker' culture where analysts wait to be asked, then round up data slowly—the tell: reactive, slow turnaround and executives who 'look data-driven' but don't act on the analysis.
• Running analytics as a separate step disconnected from operations—the tell: someone still has to manually apply the forecast, so the right inventory isn't automatically held.
• Assuming more data or newer tools equals value—the tell: heavy tech investment with little change to how the business actually operates; value comes from aggressive exploitation, not tool count.
• Deploying machine learning where accountability requires explanation—the tell: the model outperforms but no human can explain why it decided as it did, and users can't defend the recommendation.

• ▸Recurring decision: name the operational decision currently made by intuition or fixed rule
• ▸Instance volume: how many times per day/cycle this decision repeats
• ▸Live inputs: the specific data that changes each instance (e.g., actual addresses, weather, traffic)
• ▸Host system: the operational system the analytic will be embedded into (not a standalone run)
• ▸Cadence target: static → daily-refresh → real-time; state current and next step
• ▸Payoff metric: the operating unit you'll measure (miles, gallons, labor hours, dollars)
• ▸Method choice: hypothesis-based vs. machine learning, with interpretability need noted
• ▸Data-product angle: whether the output data has resale/service value to customers

1. 1Move up the analytics value ladder before claiming competitive advantage: descriptive (historical access, alerts, reporting), then predictive (propensity, segmentation, forecasting), then prescriptive (optimization, recommendation engines), then autonomous. Potential competitive advantage rises as sophistication rises, so target the highest tier your data and processes can support.
2. 2Identify the single process where analytics can reset the economics of your business, then instrument it end-to-end. UPS instrumented package movement — handheld devices in every driver's hands (1990), a proprietary national cellular network, and one of the world's largest data warehouses — before building predictive and prescriptive models on top.
3. 3Convert insight into a named, deployable model rather than a one-off analysis. UPS built Package Flow Technologies (predictive, deployed 2003) and then ORION (prescriptive, a thirty-page algorithm optimizing routes for all 55,000 drivers), which produces recurring, quantified savings.
4. 4Attach hard financial and operational metrics to the model and track them continuously: ORION shortens each driver's route by 7–8 miles/day, saving 100 million miles and 10 million gallons of fuel per year, generating over $400 million in annual cost savings and avoidance. Report savings in the currency your CEO reads.
5. 5Budget the majority of effort for change management, not the math. On ORION, the most expensive and time-consuming factor by far was teaching drivers and getting them to accept the new routing — expect the same when embedding analytics into operational processes.
6. 6Extend advantage outward: open your predictive engines to customers so they interact with your operations directly. UPS gave customers the same information it had, letting them personalize delivery and improve their own business processes — the data became as valuable to the customer as the package, making UPS a more valuable partner and deepening loyalty.
7. 7Guard the advantage against complacency. Treat 'constructive dissatisfaction' (Jim Casey's term) as policy: keep reinventing the analytical capability because the path of least resistance turned film, newspaper, and movie-rental leaders into case studies.

UPS's analytical advantage compounds across the value ladder. It started descriptive — handheld devices given to every driver in 1990, stitched onto a kludged-together proprietary national cellular network, feeding one of the world's largest data warehouses that captured the movement of every package. It moved predictive with Package Flow Technologies, deployed in 2003, which reduced 85 million miles annually. It reached prescriptive with ORION, a thirty-page algorithm computing the least-time, least-fuel route for each of 55,000 drivers; skeptics inside the company believed no computer could beat experienced drivers. ORION now shortens each route by 7–8 miles a day, saving 100 million miles and 10 million gallons of fuel per year and generating more than $400 million in annual cost savings and avoidance. UPS then opened its predictive engines to customers, who used the network data to improve their own decisions — turning an internal efficiency play into a stickier partnership and sustained competitive advantage.

• If the analysis only supports a decision that gets ignored, then you are still in the weak 'decision support' 1.0 mode — push toward prescriptive/automated actions that drive decisions rather than decorate them.
• If your target process is core, high-volume, and repeatable (like every package on every route), then invest in prescriptive optimization; the per-unit savings multiply into large, defensible numbers.
• If you cannot state the advantage in profit, miles, gallons, or dollars, then treat it as unproven — quantify cost savings and avoidance the way ORION does before calling it a competitive advantage.
• If experienced staff insist the machine can't beat their judgment, then front-load change management and pilot proof, because acceptance — not the algorithm — is your bottleneck.
• If your model produces data valuable to customers, then expose it to them; sharing predictive engines converts an internal capability into external loyalty and partnership value.

• Treating analytics as IT. The tell: budget and org chart put analysts in the back office with no close relationship to managers — decisions keep getting made on gut, and results are produced 'in spades' but never used.
• Underfunding change management. The tell: the model is technically finished but adoption stalls; frontline workers (like drivers) resist the new way of working.
• Stopping at descriptive reporting. The tell: leaders 'look data-driven' with dashboards and bar charts but never move to predictive or prescriptive, capping the potential advantage.
• Complacency after early success. The tell: the organization keeps doing what made it a winner and doesn't hear the footsteps — the film/newspaper/movie-rental pattern of winners becoming also-rans.
• Claiming advantage without measurement. The tell: enthusiasm and vocabulary but no per-unit, aggregate, or dollar figures attached to the initiative.

• ▸Target process: name the single high-volume process to instrument
• ▸Analytics tier: mark descriptive / predictive / prescriptive / autonomous being pursued
• ▸Data infrastructure: list the capture, network, and warehouse assets in place or needed
• ▸Named model/deliverable: give the model a name and deployment date (e.g., ORION, 2003)
• ▸Unit-level result: state the per-unit gain (e.g., 7–8 miles/driver/day)
• ▸Aggregate result: state annual totals (miles, gallons, dollars saved/avoided)
• ▸Change-management plan: name who must accept it and how you'll teach them
• ▸External extension: describe how customers/suppliers get access to the model
• ▸Complacency check: date and owner for next constructive-dissatisfaction review

1. 1Locate primary sponsorship at the CEO level; the passages show Netflix, Caesars, Capital One, and UPS proceed 'full steam ahead' precisely because they have CEO-level sponsorship and passion, while firms lacking it must take the slower 'prove-it' path.
2. 2If you lack executive passion now, run the prove-it path: pick a narrow, high-value domain with adequate data and hardware/software already in place, deliver measurable results, and use them to convert a senior sponsor—then escalate to enterprise-wide commitment.
3. 3Move analytics out of the back office and into the boardroom: have the CEO tout analytical activity to every stakeholder in the annual report and press, so it is visible to customers and shareholders rather than marginal and departmentally managed.
4. 4Commit real resources at scale, as UPS did with ORION—spending several hundred million over a decade to reap roughly half a billion dollars a year in labor and fuel savings—signaling that leadership backs analytics with budget, not rhetoric.
5. 5Fund change management as a first-class line item, not an afterthought; at UPS the most expensive and time-consuming factor in ORION was teaching and getting drivers to accept the new routing.
6. 6Set the cultural tone deliberately: 86% of firms surveyed tried to build a data-driven culture but only 37% succeeded, because middle managers won't adopt analytics if no leader sets the tone for how it improves their jobs and results.
7. 7Push analytical decisions through to execution everywhere in the organization—a leader's commitment fails if a single frontline manager overrides the analysis, so verify the sponsorship spreads into operational behavior.

The Boston Red Sox show senior commitment done right. Owner John Henry (a quantitative hedge fund manager) and Tom Werner provided high-level sponsorship that enabled a full-steam-ahead approach. They hired a management team that shared the vision: Theo Epstein as the youngest GM in baseball history plus assistant GM Jed Hoyer, both committed to ignoring conventional baseball beliefs in favor of data. They then brought in the best analytical talent, sabermetrics pioneer Bill James, whom no other team had bothered to hire. But 2003 exposed the limit of top commitment alone: analysts had proven Pedro Martínez became far easier to hit after ~105 pitches (opponents batted .231 on pitches 91-105 versus .370 on 106-120) and warned manager Grady Little, yet the manager left him in—showing that executive commitment must spread everywhere to change actual decisions.

• If the CEO is personally passionate and vocal about fact-based decisions, then take the full-steam-ahead path; if not, then take the prove-it path first and treat winning a sponsor as the goal.
• If your analytics live only in the actuarial, quality, or marketing department and are invisible to the CEO, then they cannot drive competitive strategy—treat that as a not-yet-committed state.
• If you're launching operational analytics that embed into a core process, then budget change management as the largest cost line, because that is where adoption actually fails.
• If a manager overrides a validated analytical recommendation in execution, then the commitment has not yet spread deep enough—escalate to close the gap before scaling further.
• If leadership expects middle managers to adopt analytics without setting the tone on how it helps their results, then culture change will stall—require the sponsor to own the message.

• Relegating analytics to the back office—managed at the departmental level and invisible to senior executives, customers, and shareholders; the tell is that analytics never appear in the boardroom, annual report, or CEO's remarks.
• Assuming top-level buy-in is enough while frontline execution ignores the analysis; the tell is the Grady Little pattern—a documented recommendation overridden at the moment of decision.
• Declaring a data-driven culture without leading it; the tell is the survey gap—86% attempt culture change but only 37% succeed, and 41% blame middle management resistance.
• Treating change management as a minor task; the tell is a stalled rollout where users won't accept the new way, as nearly derailed UPS's ORION driver adoption.
• Relying on a lone 'prove-it' champion without escalating to CEO-level sponsorship; the tell is a permanently slow path that never reaches enterprise scale because passionate executive sponsorship was never secured.

• ▸Primary sponsor & title: name the executive advocate (ideally CEO; COO/CFO/CIO/president acceptable).
• ▸Path selection: mark full-steam-ahead (CEO passion present) or prove-it (must earn sponsorship first).
• ▸Visibility check: list where analytics appear externally—boardroom, annual report, press, CEO speeches.
• ▸Resource commitment: state budget committed and expected annual payoff (model it, ORION-style).
• ▸Change management allocation: percent of project effort and owner assigned to adoption.
• ▸Culture-tone message: the one sentence the sponsor uses to show how analytics improves managers' results.
• ▸Execution-depth test: name the decision points where analysis must override convention, and who enforces it.
• ▸Distinctive data & talent: the strategic data source and the best analytical hire securing the capability.
• ▸Escalation trigger: the measurable result that converts a prove-it win into CEO-level commitment.

1. 1Treat analytical IT as necessary but not sufficient: acquire the software stack (SAS/IBM for full descriptive-predictive-prescriptive, Tableau/Qlik/MicroStrategy for visualization, R/Python/RapidMiner open source, Hadoop/Spark frameworks), but invest equally in the human and organizational aspects, which the passages call the truly differentiating factor.
2. 2Consolidate the data layer into one managed environment before scaling analysts: build an enterprise data warehouse (as UPS did) that captures transaction-level data — every package movement — rather than leaving data fragmented across ERP, POS, web, and mobile silos, so the whole organization works from one version of the truth.
3. 3Select one or a few distinctive capabilities (profitable-customer identification and optimal pricing, supply-chain/inventory optimization, store-location choice, M&A candidate selection, hiring/retention) and aim analytical resources at those, rather than spreading effort thin across every function.
4. 4Acquire and manage information end-to-end — acquire, transform, manage, analyze, and act on both external and internal data — treating information management as a prerequisite for good analytical capability, not an afterthought.
5. 5Move up the analytics maturity spectrum from descriptive/monitoring toward predictive and prescriptive (the higher-value, more proactive end where competitive advantage increases), and embed models directly into products, services, and operational processes.
6. 6Open analytical engines to customers where it strengthens the capability: UPS gave customers the same predictive information it used so they could interact with operations directly and personalize delivery.
7. 7Decide where to scale beyond artisanal modeling: if you already have traditional analytical skills, evaluate autonomous 'model factory' machine-learning approaches to move from tens or hundreds of models to thousands, since transitioning is easier for organizations with existing capabilities.
8. 8Provide broad access and translation: employ people to translate and evangelize models to business functions (sales, marketing) so enterprise-wide analytical output is actually adopted, not stranded in a technical group.

UPS built its enterprise analytics capability incrementally rather than as scattered projects. It designed one of the first handheld devices and put it in every driver's hands in 1990, then kludged regional cellular networks into a proprietary nationwide network so it could gather data on the movement of every package. That coordinated data flow fed one of the largest data warehouses in the world — a single enterprise data source rather than functional silos. On that foundation it layered predictive tools (Package Flow Technologies, deployed 2003, saving 85 million miles annually) and then the prescriptive ORION routing model, which the authors call the 'granddaddy' of real-time applications and possibly the largest commercial analytics project ever undertaken. Critically, UPS also opened its predictive engines to customers, extending enterprise-wide access beyond its own walls. The progression — devices, network, warehouse, predictive, prescriptive — shows enterprise coordination compounding over time rather than isolated wins.

• If your products and technology look like competitors', then compete on high-performance business processes and smartest decisions, because geography, protective regulation, and proprietary technology are no longer defensible bases.
• If you already have traditional artisanal analytical skills, then favor a transition to autonomous/machine-learning modeling to scale volume, because organizations with existing capabilities transition more easily.
• If a decision is high-stakes (M&A, store location, pricing), then back it with systematically assembled data and fact-based analysis rather than intuition, because good decisions usually have data behind them.
• If software or hardware availability is your stated constraint, then re-examine the assumption — software and cloud/compute are no longer scarce; the ability to use analytical software well remains the differentiator.
• If data is too big, volatile, and unstructured for traditional tools, then adopt big-data frameworks (Hadoop, Spark) and machine-learning approaches rather than forcing it through legacy technology.

• Equating analytics with analytical information technology — the tell is a budget dominated by software/hardware purchases with no investment in analysts, organization, or process, which the passages flag as a huge mistake.
• Running analytics as narrow, disconnected functional activities (like early DSS) — the tell is multiple functions each with their own tools and no shared data warehouse or single version of the truth.
• Complacency after early wins — the tell is 'misguided satisfaction' and staying the course; Casey countered this with 'constructive dissatisfaction,' and film, newspaper, and rental firms became case studies by not reacting.
• Trying to achieve modern granularity with artisanal methods — the tell is armies of statisticians (e.g., 35 offshore statisticians producing only 150 models a year) where autonomous approaches could produce thousands with fewer than 2.5 FTEs.
• Building models nobody uses — the tell is analytical output that never reaches sales/marketing because no one translates or evangelizes it to business users.

• ▸Distinctive capability targeted: name the one or few processes/decisions where you must be better than anyone in your industry.
• ▸Single data source status: describe your enterprise data warehouse and which silos (ERP, POS, web, mobile) still feed it separately.
• ▸Information lifecycle gaps: mark which of acquire/transform/manage/analyze/act on external and internal data is weakest.
• ▸Maturity level per capability: rate as descriptive, predictive, or prescriptive and state target level.
• ▸Software/framework stack: list tools in use (e.g., SAS/IBM, Tableau, R/Python, Hadoop/Spark) and any real availability gap.
• ▸Artisanal vs. autonomous modeling: state current model volume and whether machine-learning 'model factory' scaling applies.
• ▸Analyst and translator staffing: count model builders plus people who translate/evangelize models to business functions.
• ▸Access breadth: note who inside (and which customers/suppliers) can reach analytical outputs directly.
• ▸Complacency check: name the competitive threat you are currently under-reacting to (the 'constructive dissatisfaction' prompt).

1. 1Separate your talent into two populations: analytical professionals (data scientists, quantitative analysts) who design and build, and analytical amateurs (decision makers, business users) whose data literacy determines whether analytics get used. Staff and train both.
2. 2Hire for the cumulative skill stack. A 1.0 quantitative analyst needs statistics, data integration and cleaning, business understanding, communication about data, and the ability to inspire trust in decision-makers. Add 2.0 skills (experimentation, transforming unstructured data, open-source tools like Hadoop/Python, visual analytics) and 3.0 skills (product/engineering knowledge, change management) as you advance.
3. 3Screen recruits for quantitative reasoning ability the way GE, Accenture, Capital One, and P&G do — make it an explicit filter in hiring, not an afterthought.
4. 4Bring in scarce, best-in-class outside talent as consultants or leaders when you have a capability gap, as the Red Sox did by hiring Bill James and installing Epstein and Hoyer.
5. 5Push descriptive-reporting work off analytical professionals and onto users via self-service and visual analytics tools, freeing scarce experts for predictive and prescriptive work.
6. 6Budget change management as a first-class workstream when embedding analytics into operations — teaching and getting frontline people to accept new methods is often the most expensive, time-consuming factor.
7. 7Spread analytical decision authority everywhere it must be executed: an analytical recommendation is worthless if the person acting on it overrides it. Ensure decision-makers understand and trust the analysis enough to follow it under pressure.
8. 8Attach metrics to talent management itself — build indices or globally consistent measurement platforms to evaluate how each unit invests in and manages human capital.

In the 2003 ALCS deciding game against the Yankees, Red Sox analysts had proven conclusively that Pedro Martínez became far easier to hit after ~105 pitches — opposing batting average against him was .231 for pitches 91–105 but jumped to .370 for pitches 106–120. They warned manager Grady Little not to leave Martínez in past that point. But the analytical insight lived with the analysts, not with the decision-maker at the moment it mattered: Little let Martínez keep pitching into the eighth, even against his pitching coach's advice, and the Yankees shelled him. The organization had elite analytical professionals but a gap in workforce capability at the point of execution — the manager lacked the data literacy or trust to act on the finding. The lesson is that analytical talent has to spread everywhere within an organization for analytical decisions to actually be implemented; concentrating it in a back office defeats it.

• If the value of analytics depends on frontline acceptance (operational analytics like UPS's ORION), then fund change management as the largest line item, not an add-on.
• If you have a genuine capability gap and the market has underemployed top talent (like Bill James before the Red Sox hired him), then acquire that talent as a first mover before competitors close the gap.
• If most competitors have already adopted basic analytics (as MLB teams had by 2011), then analytical talent alone won't differentiate you — pair it with time, resources, and a rebuild mandate as the Cubs did.
• If a decision-maker can override an analytical recommendation at the point of action, then invest in their data literacy and trust, not just in producing the analysis.
• If descriptive reporting is consuming your scarce professionals' time, then move those users to self-service/visual tools and redeploy experts to predictive and prescriptive work.

• Concentrating analytical talent in a specialist group while decision-makers stay analytically illiterate — the tell is documented insights (like the Martínez pitch-count data) that get ignored at the moment of execution.
• Hiring for the newest skills only and neglecting the cumulative base — the tell is data scientists who can't clean and integrate data or communicate results and inspire trust.
• Treating change management as free — the tell is a technically successful analytics build (like ORION) stalling because frontline workers won't accept the new way.
• Assuming analytical talent alone confers advantage after the field has matured — the tell is late adoption yielding no edge because most competitors already have analysts (the Cubs recognized data-driven insights 'would not be enough' by 2011).
• Leaving top analytical talent unhired because it looks unconventional or underemployed — the tell is a rival capturing a first-mover advantage; fivethirtyeight found teams with at least one analyst in 2009 outperformed expected win percentage by 44 points over 2012–14.

• ▸Professional talent inventory: list current data scientists/analysts and their 1.0/2.0/3.0 skills
• ▸Skill-stack gaps: which cumulative skills (stats, data cleaning, experimentation, open-source, change mgmt) are missing
• ▸Amateur data literacy: rate decision-makers' ability to interpret and act on analytics by unit
• ▸Recruiting screen: is quantitative reasoning an explicit hiring filter? (yes/no + how)
• ▸External talent needs: which scarce/best-in-class capability to bring in as consultant or leader
• ▸Execution-point risk: where can a decision-maker override an analytical recommendation, and is trust in place?
• ▸Change-management budget: resources allocated to teaching/adoption for each embedded-analytics project
• ▸Self-service shift: which descriptive/reporting tasks can move from professionals to users via visual tools
• ▸Talent-management metric: index or measure used to evaluate each unit's investment in analytical human capital

1. 1Inventory what you have against the Analytics 1.0 baseline: a relational data warehouse fed by extract-transform-load (ETL), with all data forced into rows and columns before storage. Note where ETL is consuming disproportionate time and resources and where warehoused data has leaked into production applications.
2. 2Consolidate analytics resources out of departmental silos. Map every place analytics lives today (actuarial pricing, manufacturing quality management, marketing lifetime-value analysis) and pull data, technology, and analysts into coordinated enterprise-wide management rather than disconnected functions.
3. 3Assess data against the DELTA 'Data' standard: integrated, high-quality, easily accessible about your business and markets. Push toward distinctive sources and types of data — the kind competitors cannot cheaply replicate — not just cleaner versions of common data.
4. 4Match hardware and software to the analytical method. Traditional artisanal modeling runs on the warehouse; machine learning and deep learning need in-memory architectures or cloud-based environments that expand on demand plus large volumes of labeled training data. Decide per workload.
5. 5Choose your software stack deliberately: proprietary vendor tools (more support) versus open-source libraries like those released by Google, Microsoft, Facebook, and Yahoo! (less support). Add Hadoop-class tooling for managing and analyzing data that won't fit the row-and-column model.
6. 6Embed analytics into operational systems rather than leaving them in the back office as 'decision support.' Move outputs into production processes and add visual analytics so results reach the boardroom and annual report, not just technicians.
7. 7Instrument model monitoring as a permanent function. Since deep learning and neural networks are 'black boxes' you cannot fully interpret, assign analysts to watch outputs for relevance and reasonability and to flag when predictive performance decays.
8. 8Escalate to enterprise scale only under executive sponsorship. Confirm a manager with commitment exists; if there's no CEO-level passion, run a 'prove-it' path to demonstrate value before making the full architecture investment.

UPS built ORION on top of the data flowing through its physical package network. Rather than treating that data as exhaust, the architecture captured and analyzed it to optimize routing, generating more than $400 million in annual cost savings and avoidance. The distinctive-data principle then showed up: UPS discovered the network data was often as valuable to customers as the packages themselves, so customers used extracted data to improve their own processes and make better decisions. This turned an internal optimization architecture into a customer-facing data product and made UPS a more valuable partner. The prerequisite was CEO-level sponsorship — UPS is named as a 'full steam ahead' company with executive passion — which let it move fast rather than through a slow prove-it path. UPS then expanded the same architecture beyond physical-network optimization into many other facets of the business.

• If ETL time and resources dominate your data effort and warehoused data has bled into production systems, then you are still architected for Analytics 1.0 and need modern storage/compute (Hadoop, in-memory, cloud), not a bigger warehouse.
• If a workload is a machine learning or deep learning model running on large data volumes, then provision in-memory or expandable cloud hardware and secure labeled training data (e.g., an ImageNet-style source) before building; the warehouse alone won't support it.
• If models are 'black box' (neural nets, deep learning) and used in a regulated industry, then don't rely on interpretability — instead instrument continuous monitoring and retire/retrain any model whose predictive accuracy drops.
• If analytics resources sit in departmental silos invisible to senior executives, then they cannot drive competitive strategy; consolidate to enterprise management before expecting strategic impact.
• If there is no passionate CEO-level sponsor, then take the slower prove-it path to demonstrate value first, because full-scale architecture investment without sponsorship stalls.

• Treating the data warehouse as the finish line. Tell: every new data type requires forcing it into rows and columns via ETL, and no one can catalog what's actually in the warehouse anymore.
• Running analytics as an 'order taker' back office. Tell: analysts return a month later with an answer the manager doesn't understand and doesn't use — analytics 'support' decisions but get ignored.
• Assuming you can fully interpret machine learning models. Tell: teams demand explanations for thousands of models and tens of thousands of variables and stall, instead of monitoring whether the models still predict accurately.
• Leaving capabilities departmentalized (actuarial here, quality management there, marketing lifetime-value elsewhere). Tell: activities are valuable to a function but invisible to executives, customers, and shareholders and never touch competitive strategy.
• Buying open-source ML libraries without accounting for reduced support. Tell: analysts get the algorithm running but have no vendor backstop when it breaks or needs tuning, and adoption stalls.

• ▸Current storage era — name your primary storage (warehouse/Hadoop/cloud) and note where ETL costs and production-data leakage occur.
• ▸Data quality & integration — rate integrated, high-quality, easily accessible (yes/partial/no) per key domain.
• ▸Distinctive data sources — list data you hold that competitors cannot easily obtain (e.g., network flow data).
• ▸Silo map — list every department currently running its own analytics tools and who owns each.
• ▸Workload-to-hardware match — for each analytical method, note required compute (warehouse / in-memory / cloud).
• ▸Software stack decision — proprietary vs open-source per capability, and note the support tradeoff accepted.
• ▸Model monitoring plan — name who watches outputs for relevance/reasonability and the accuracy threshold that triggers retraining.
• ▸Deployment target — mark which analytics are embedded in operational systems vs still back-office.
• ▸Sponsorship level — CEO-level passion (full steam ahead) or prove-it path required.

1. 1Run a maturity assessment against the four attributes of analytical competition — support of a strategic distinctive capability, enterprise-wide approach, passionate/driving executives, and initiatives aimed at substantial results — and score how much of each you exhibit (all four = stage 5; only one or two = lower stages).
2. 2Place yourself on the five-stage continuum: stage 1 (analytically impaired, virtually no capability) through stage 5 (full-blown analytical competitor). Use the stage-attribute table logic: identify which attribute is missing or weakest, because that gap defines your stage ceiling.
3. 3Check whether your analytics are actually bound to strategy: look for analytics (or synonyms) in recent annual reports. A stage 4 firm can have strong analytical activity across the business yet fail this test — the tell that will/strategic focus, not skill, is the blocker.
4. 4Separate the 'skill' question from the 'will' question. If you have skill but the CEO and executive team are supportive-but-not-passionate, you are stage 4 and need executive conviction, not more tools. If you lack the technical/data foundation, you are lower and need capability-building first.
5. 5Sequence investments to the stage: stage 5 firms control the volume of descriptive analytics and push self-service to users; lower-stage firms still must do foundational 1.0 work (getting, cleaning, storing data). Do not skip 1.0 activities — every organization still needs them.
6. 6Focus analytical capability on the single most significant capability your strategy requires before broadening. Stage 5 firms already have theirs focused on that capability and only then expand more broadly.
7. 7Re-assess periodically and benchmark against external maturity assessments (e.g., IIA's Analytics Maturity Assessment on the five-point scale plus DELTA factors) to guard against complacency — the 'constructive dissatisfaction' Jim Casey advocated.

A consumer products firm studied by the authors had strong analytical activities in several areas of the business — enough skill to look advanced. But on assessment it landed at stage 4, not stage 5, because two attributes were weak: analytics were not clearly bound to strategy, and neither analytics nor likely synonyms appeared in the company's recent annual reports. The diagnosis was 'skill but not the out-and-out will to compete on this basis' — the executive team was supportive but not passionate. The remedy was not more analytical projects but tying the existing activity to a distinctive strategic capability and securing driving executive commitment. With only a small increase in emphasis the firm could cross into stage 5. Contrast this with Amazon, which scored highest of any company assessed by IIA — a digital native whose analytics are inseparable from strategy.

• If you exhibit all four attributes (distinctive-capability support, enterprise-wide scope, passionate executives, substantial-results focus), then you are stage 5 and should broaden application while protecting focus; otherwise identify the missing attribute as your ceiling.
• If you have the skill but the executive team is supportive rather than passionate, then you are stage 4 — invest in executive will and strategic binding, not additional tooling.
• If analytics (or synonyms) do not appear in your annual reports despite active analytical work, then treat analytics as not yet bound to strategy and correct that before claiming competitor status.
• If you are early-stage, then keep doing 1.0/descriptive foundational work (data acquisition, cleaning, storage) rather than jumping to advanced techniques, because every organization still needs 1.0 activities.
• If you are a large, relatively sophisticated firm, do not assume you are already a competitor — IIA's 2016 survey of fifty firms found most were not, and only digital natives approached competitor levels on average.

• Mistaking skill for stage 5: doing lots of analytics across the business while it stays disconnected from strategy. Tell — analytics absent from annual reports and no single distinctive capability targeted.
• Complacency — 'staying the course' and not hearing the footsteps behind you. Tell — misguided satisfaction, no constructive dissatisfaction, treating past success as sufficient (the film/newspaper/movie-rental pattern).
• Over-investing in technology as the answer. Tell — belief that value comes from how much data or which tools you have, when it actually comes from how aggressively you exploit resources to create new ways of doing business.
• Assuming size or industry reputation equals maturity. Tell — a large sophisticated firm self-rating as a competitor when systematic assessment places most such firms below competitor level.
• Skipping foundational descriptive work in a rush to predictive/prescriptive. Tell — no reliable view of what happened recently versus the distant past, or uncontrolled descriptive volume flooding analytical professionals.

• ▸Distinctive-capability support (0-3): rate how tightly analytics support one important, distinctive strategic capability
• ▸Enterprise-wide scope (0-3): rate how broadly analytics reach beyond isolated pockets
• ▸Executive passion/drive (0-3): rate whether leaders are passionate drivers vs. merely supportive
• ▸Substantial-results targeting (0-3): rate whether initiatives aim at significant business outcomes
• ▸Strategy-binding evidence: note whether analytics/synonyms appear in recent annual reports (yes/no)
• ▸Assigned stage (1-5): map scores to the five-stage continuum; lowest weak attribute sets the ceiling
• ▸Skill vs. will gap: state whether the primary block is capability (skill) or executive commitment (will)
• ▸Foundational 1.0 status: confirm data getting/cleaning/storage and descriptive coverage are in place
• ▸Next-move priority: the single action that raises the weakest attribute one level