Section 1
The five challenges at a glance
Five failure modes define the data-driven trap for service businesses. First, purpose-finding: starting from the data you have and asking what it can tell you, the core inversion de Langhe and Puntoni documented, which systematically biases firms toward questions that are answerable rather than important (de Langhe and Puntoni, 2020). Second, dashboard proliferation: metrics accumulate faster than decisions, until reporting becomes an end in itself and the weekly review inspects numbers no pending decision needs. Third, the lamppost problem: available data describes the past of your existing business, while growth decisions concern alternatives you have not tried, for which by definition no internal data exists. Fourth, advocacy analytics: data recruited after a preference forms, where analysis functions as ammunition rather than illumination, a pattern long documented in confirmation-bias research and amplified when leaders ask analysts to support a case (Kahneman, Sibony and Sunstein, 2021). Fifth, precision theater: small-sample service businesses reading signal into noise, treating a three-client uptick as a trend and re-deciding strategy monthly on numbers too thin to mean anything. The table summarizes all five; the following sections examine purpose-finding, the lamppost problem, and precision theater in depth, because those three most directly suppress decision quality and velocity in firms under roughly 50 people.
Section 2
Challenge one: purpose-finding, the quiet inversion at the heart of data-driven culture
De Langhe and Puntoni open with the uncomfortable observation that despite record data volumes, many executives report their analytics initiatives fail to produce actionable insight (de Langhe and Puntoni, 2020). Their diagnosis is sequence, not competence. In a data-driven culture, the existing data sets the agenda: analysts explore what the warehouse can answer, surface the most interesting findings, and leadership reverse-engineers significance from them. The result is decisions shaped by data availability, which is a function of which tools the firm happened to buy, not which questions matter. Their alternative is a discipline: instead of finding a purpose for data, find data for a purpose. The purpose is always a decision with alternatives. For a service firm, the live decisions are concrete: raise prices or hold; specialize the niche or stay broad; hire the strategist or subcontract; double the winning channel or diversify. Each of these defines, in advance, what evidence would change the answer. Purpose-finding feels productive because dashboards are full and meetings have charts, but it optimizes the measurable past while the decisions that determine growth go under-evidenced. The first corrective step costs nothing: list the firm's five most consequential open decisions, then audit how much of last month's analytical attention touched any of them. For most operators the honest answer is close to none.
Section 3
Challenge two: the lamppost problem, your data describes the road already taken
The drunkard searches under the lamppost because the light is better there, and operators analyze their existing funnel because that is where the data is. The deeper point in decision-driven analytics is that a decision is a choice among alternatives, and the data that matters is the data that discriminates between those alternatives (de Langhe and Puntoni, 2020). Internal data almost never does this, because internal data is a record of the single path already taken. Your CRM can describe conversion on your current offer at your current price to your current audience; it is structurally silent on the offer you have not launched, the segment you have not addressed, and the price you have not tested. Firms that miss this end up extremely confident about marginal optimizations and evidence-free on strategic moves, which inverts the stakes. The remedy is to treat decision-relevant data acquisition as its own activity: structured win-loss interviews, a price test on the next ten proposals, a pilot cohort for the unlaunched offer, competitor proposal teardowns. Eisenhardt's fast-deciding firms behaved exactly this way, relying on real-time information generated close to the decision rather than retrospective reports (Eisenhardt, 1989). The discipline pairs naturally with the 70% threshold from the velocity pillar: define the discriminating evidence, acquire just enough of it, decide, and let the reversible decision itself generate the next round of data.
Section 4
Challenge three: precision theater and noise in small-sample judgment
Service firms in the 5-7 figure range run on small numbers: a dozen proposals a quarter, a handful of clients per cohort, channel data measured in dozens of leads. At that scale, monthly metric movements are dominated by noise, yet operators routinely re-decide strategy on them, killing a channel after one weak month, declaring an offer validated after three sales. This is precision theater: the charts have decimal points, so the conclusions feel rigorous. The judgment research says intuition about variance is reliably wrong. Kahneman, Sibony and Sunstein report that when an insurer asked underwriters to price identical cases, executives predicted around 10% variance between judgments; the median was 55% (Kahneman, Sibony and Sunstein, 2021). If trained professionals disagree by half on identical inputs, a founder reading a 15% swing in a 20-lead month as signal is decoding static. The decision-driven correction has two parts. First, set decision thresholds before looking: define in advance what magnitude of change, sustained over what period, would alter the decision, which converts data review from Rorschach test to tripwire check. Second, reduce noise in the judgments themselves: independent written reads from each leader before discussion, and a standing rule that strategy changes require either a pre-set threshold being crossed or genuinely new information. Velocity improves because the firm stops oscillating; quality improves because decisions track signal.
Section 5
Innovative solutions
Operators applying this research well converge on five practices. One: the decision backlog replaces the metrics review. The weekly leadership meeting opens with the ranked list of open decisions, and data is presented only in service of an item on that list, directly implementing the decision-first sequence (de Langhe and Puntoni, 2020). Two: question-zero discipline. Every analytics request states the decision it serves, the alternatives in play, and what result would change the chooser's mind; requests that cannot answer question zero are declined. This single rule eliminates most advocacy analytics, because data recruited to confirm a foregone conclusion cannot specify what would change the decision. Three: the narrow dashboard. Each tracked metric must map to either a standing decision rule or an active tripwire; metrics that inform nothing are archived, not displayed. Firms typically cut tracked KPIs by more than half and notice nothing missing. Four: cheap discriminating experiments. For lamppost-blind decisions, the firm buys data with small reversible probes, ten-proposal price tests, two-week pilot offers, five win-loss interviews, treating Bezos's two-way-door logic as a data-acquisition strategy (Amazon, 2015 shareholder letter). Five: noise hygiene on judgment calls, with independent written estimates before discussion and pre-registered thresholds for strategy changes (Kahneman, Sibony and Sunstein, 2021). Together these shift analytical hours from describing the past to discriminating between futures.
Section 6
Solution framework
The Decision-Driven Operating Loop runs in five steps and fits a firm with no analyst. Step one: enumerate. Maintain a living list of the firm's open decisions, each stated as a genuine choice between named alternatives, because de Langhe and Puntoni's research locates most analytic waste in questions that were never choices (de Langhe and Puntoni, 2020). Step two: specify. For the top three decisions, write the discriminating evidence: what observation would make alternative A clearly better than B. If no observation could change the answer, the decision is already made and should simply be executed. Step three: source. Match each evidence requirement to its cheapest source, existing data, a quick external pull, or a designed probe, and time-box acquisition using the 70% threshold so evidence-gathering cannot become deferral (Amazon, 2016 shareholder letter). Step four: judge with hygiene. Decision owners collect independent written reads, then decide on the pre-stated criteria, logging the call and its review date in the decision register. Step five: retire and re-arm. Once decided, the supporting analysis is archived, the dashboard sheds any metric that served only that decision, and tripwires from the decision log become the only standing monitors. The loop is the analytics counterpart of the decision velocity system: every step exists to shorten the distance between a question and a committed, evidence-matched answer.
Section 7
Evidence-based action plan
Week one: run the two audits. List every metric currently reviewed weekly or monthly, and separately list the five most consequential open decisions. Draw the mapping. Metrics serving no decision get archived; decisions served by no data get flagged. Most firms discover the McKinsey pattern in miniature, substantial review time producing little decision progress (McKinsey, 2019). Week two: install question zero. Announce that every analysis request, internal or to a vendor, must state the decision, the alternatives, and what result would change the answer. Apply it retroactively to recurring reports; expect at least a third to fail the test. Week three: specify and source evidence for the top three decisions. Write the discriminating-evidence statement for each, then launch the cheapest probes: a price test on live proposals, five win-loss interviews, a two-week pilot. Set the 70% threshold and a decision date for each (Amazon, 2016 shareholder letter). Week four: rebuild the review meeting. The agenda becomes the decision backlog; each item gets its evidence reviewed against pre-stated criteria, independent reads first, and either closes with a logged decision or gets a specific evidence gap and date. Measure two numbers monthly: decisions closed, and tracked metrics retired. A firm that closes more decisions while watching fewer numbers has escaped the data-driven trap, and its analytics finally purchase velocity. For adjacent evidence in this pillar, see [Reversible vs Irreversible Decisions: The Bezos Type 1/Type 2 Framework and the Research Behind It](/blog/growth-reversible-vs-irreversible-decisions) and [The Uncertainty Advantage: What the Research Says About Small-Firm Agility vs Large-Firm Resources](/blog/growth-uncertainty-advantage-small-firm-agility).