A project executive opens the Monday dashboard and the schedule performance index has dropped from 0.97 to 0.91 in three weeks, the cost performance index is holding at 1.02, and the critical-path float has fallen from twelve days to four. The predictive analytics layer, riding on the Primavera and Procore data, has flagged the project amber and projects a nineteen-day overrun if the trend holds. The instinct is to read that as a verdict, the project is failing, and to start a recovery scramble. But a declining SPI is a signal, not a sentence: it says progress is lagging the baseline, and it does not say why, whether a single trade is stuck behind a late submittal, whether the baseline itself was optimistic, whether a weather month is distorting the curve, or whether the update data is simply stale because the last three two-week reports were never properly statused. This lesson is about the discipline of reading SPI, CPI, and float consumption as the predictive-ML engine's leading indicators of schedule risk, the metric-as-signal that surfaces where to look while the human interprets the cause and decides the response, on data the human keeps honest, because an indicator read as a verdict drives the wrong scramble and an indicator built on stale data points at nothing at all.

Earned Value Is the Predictive-ML Engine on the Schedule

The program named four engines, and the third is predictive ML, which forecasts from patterns in data. Earned-value analytics is that engine applied to the schedule and the budget. The mechanics are old and well-codified: you measure the budgeted cost of work performed (the earned value), compare it to the budgeted cost of work scheduled (the planned value) to get the schedule performance index, SPI equals earned value over planned value, and compare it to the actual cost of work performed to get the cost performance index, CPI equals earned value over actual cost. An SPI of 1.0 means you are exactly on schedule, below 1.0 means behind, above 1.0 means ahead; the same logic runs for CPI on cost. Float consumption tracks how fast the schedule is spending its buffers, the total float on the paths that feed the critical path, because a project can hold an SPI near 1.0 while quietly burning the float that protects the finish date.

What makes this predictive rather than merely descriptive is the projection. A current SPI of 0.91 is a description of where you are, but the analytics platforms, Procore's analytics, InEight, nPlan's forecasting trained on thousands of historical schedules, project that index forward to estimate the completion date if the current performance continues, turning the present metric into a forecast. nPlan in particular is explicit about being a predictive engine: it learns from a large corpus of past project schedules to forecast where the current one is likely to land and to flag the activities most likely to slip, which is the predictive-ML engine in its purest form, a forecast from patterns in a body of historical data.

These are called leading indicators for a reason. A finish-date slip is a lagging fact, you learn it when you miss the date, too late to act. SPI, CPI, and float consumption move before the date is missed: the SPI starts declining and the float starts burning weeks or months before the finish slips, so they are early warnings that give the project the lead time to respond while response is still possible. The value of the predictive engine is precisely this earliness, the same earliness that made procurement risk tracking valuable, surfacing the schedule risk while there is still time to act on it, which is why a project runs these indicators rather than waiting for the finish date to tell the truth.

A Declining SPI Is a Signal, Not a Verdict

The program established the metric-as-signal discipline in the RFI closure and procurement lessons, and it governs the earned-value indicators completely. A declining SPI signals that progress is lagging the baseline; it does not deliver a verdict on why, or on what to do. The number is a symptom, and a symptom can have many causes, so reading the symptom as a diagnosis is the error the discipline exists to prevent. An SPI of 0.91 could mean a single critical trade is stuck behind a late owner decision, a problem that is real but contained and recoverable; or it could mean the baseline was padded with optimistic durations and the project is actually fine relative to a realistic plan; or it could mean a procurement delay has stalled a whole front; or it could mean nothing real at all because the schedule update was sloppy and the earned value is understated. The same 0.91 carries all these possibilities, and the metric cannot tell them apart.

This is why the human interprets the cause in context rather than reading the number as a sentence. The interpretation is where the project knowledge enters: the executive who knows that the late owner decision is about to be resolved reads the 0.91 differently from one who knows the procurement delay is structural and worsening. The metric directs attention to the project, says look here, something is lagging, and the human looks and determines what the lag actually is and whether it is recoverable, which the index alone cannot say. Treating the index as a verdict, declaring the project failing because the SPI is below some threshold, skips the interpretation and risks the wrong response, a recovery scramble for a problem that was a baseline artifact, or complacency about a 0.97 that is quietly burning all the float on the path that matters.

A schedule index is a thermometer, not a diagnosis. It tells you a temperature is high and it tells you nothing about the disease, so the number is the start of the inquiry, not the end of it, and the project that treats the index as a verdict treats the symptom and misses the cause.

So the indicators are read as signals that initiate an inquiry, not as verdicts that conclude one. The SPI drop is the prompt to ask why, the float burn is the prompt to ask which path and whether it matters, the CPI movement is the prompt to ask whether the cost trend is a productivity problem or a scope change. The metric earns its keep by surfacing the question early; the human earns the answer by interpreting the cause and deciding the response, which is the metric-as-signal division applied to the schedule indicators that most tempt a reader to mistake the number for the meaning.

The Predictive-Attention Posture: The Indicator Surfaces Where to Look

The program calls the correct stance toward predictive indicators the predictive-attention posture: the indicator does not make the decision, it directs the human's attention to where a decision may be needed. SPI, CPI, and float consumption are attention-direction instruments. On a project with hundreds of activities and dozens of trades, a human cannot watch everything, so the indicators do the watching and surface the places that warrant the human's limited attention, the activity nPlan flags as most likely to slip, the path whose float is collapsing, the cost account whose CPI is drifting. The human then spends their attention where the indicator pointed, investigating the flagged place and deciding what, if anything, to do.

This posture resolves the tension between the volume the indicators can watch and the judgment only the human can apply. The indicators watch the whole project continuously, which no human could, and they surface the few places that matter, which is their value. The human applies judgment to those few places, which the indicators cannot, which is the human's value. Neither substitutes for the other: an executive who watches no indicators and relies on gut misses the early signals the indicators catch, and an executive who acts on every indicator as a verdict scrambles at every fluctuation and burns the team's trust. The posture is to let the indicators direct the attention and to apply the judgment where they direct it.

The predictive-attention posture also disciplines the response to a flag. When the float-consumption indicator flags a path collapsing toward zero, the posture is not to immediately crash that path but to look: is this the critical path or a near-critical one that does not yet drive the finish, is the float burn a one-time event or a trend, is the path's collapse a real productivity problem or an artifact of a logic change in the last update. The flag is the start of the look, and the response follows the look, not the flag, which keeps the project from spending recovery effort on a flag that the look would have shown did not need it, and concentrates the effort on the flags that the look confirms.

The Indicators Are Only as Good as the Update Data

The earned-value indicators are computed from the schedule update data, the percent-complete statusing, the actual costs booked, the activity progress reported, so the indicators are only as good as that data, which is the data-quality dependency the program has taught for every predictive and analytic step. If the schedule is not statused candidly, if the superintendents report optimistic percent-complete to look good, or if the last two updates were skipped and the data is stale, the SPI and CPI are computed from fiction and the forecast is fiction. A project can show a comfortable SPI of 0.98 that is pure illusion because the field is reporting ninety-percent-complete on activities that are sixty-percent-complete, and the indicator, faithfully computing from the bad data, will reassure the executive right up to the missed date.

This makes the data-quality dependency the central vulnerability of the schedule indicators, and it cuts in both directions. Optimistic statusing inflates the SPI and hides the real risk, the dangerous direction, because it produces a false negative, a comfortable indicator over a project that is actually behind. Pessimistic or stale statusing depresses the SPI and manufactures a phantom risk, which wastes recovery attention on a problem that is a data artifact. Either way, the indicator is reporting the data, not the reality, when the two diverge, so the project must invest in the data quality, the honest, timely statusing, as the precondition for the indicators to mean anything. The discipline established for procurement tracking applies exactly: the prediction is only as good as the data, so the step is a prediction-plus-data-quality discipline, not a prediction alone.

The human's interpretation must therefore include the data behind the indicator. Before reading a 0.91 SPI as a real schedule lag, the executive asks whether the update is current and the statusing honest, because a 0.91 from clean data is a real signal worth investigating and a 0.91 from stale or gamed data is noise worth correcting at the source. The skilled reader of these indicators is as attentive to the data quality as to the numbers, discounting indicators built on suspect data and trusting indicators built on clean data, because the indicator inherits the data's reliability and a confident forecast on bad data is the most dangerous output the engine can produce, a false precision that the project mistakes for knowledge.

Float Consumption: The Quiet Indicator That SPI Can Hide

SPI and CPI get the attention because they are single, legible numbers, but float consumption is often the more telling indicator, and it is the one a project most often misreads, because a project can hold a healthy SPI while burning the float that protects the finish. SPI measures progress against the planned value across the whole schedule, so a project can be on track in aggregate, SPI near 1.0, while one path is consuming its float far faster than planned, eating the buffer that stood between a delay on that path and a slip of the finish date. When that float reaches zero, the path becomes critical, and the next delay on it moves the finish, but the SPI gave no warning because the aggregate progress looked fine.

This is why float consumption is the leading indicator the discipline most insists the human watch alongside the headline indices. The float burn on the near-critical paths is the early warning of a future critical-path problem, surfacing the path that is about to start driving the finish before it starts driving it. A project that watches only SPI and CPI can be blindsided by a path that was never critical becoming critical, because the aggregate indices smoothed over the path-level float burn. The float-consumption indicator surfaces that path-level risk, directing the human's attention to the path that is spending its buffer, which is exactly the kind of early, specific signal the predictive engine is for.

Reading float consumption well requires the same interpretation discipline as the indices. A path burning float is a signal to look at that path, not a verdict that the path is failing, because the burn may be a planned consumption the baseline anticipated, or a one-time event, or a real trend that will reach zero and turn the path critical. The human looks at the flagged path, determines whether the burn is anticipated or alarming, and decides whether to act, the same metric-as-signal discipline applied to the indicator that the headline indices most often hide, which is why the disciplined reader treats float consumption as a first-class indicator and not a footnote to the SPI.

The Applied Problem: A Schedule-Risk Reading Protocol

Here is the exercise. Build the schedule-risk reading protocol your project will use to read SPI, CPI, and float consumption as leading indicators, with the interpretation and the data discipline the indicators require. The protocol's purpose is to convert the predictive engine's signals into correct responses, surfacing the schedule risk early while keeping the human's interpretation of the cause and the data behind the indicator between the signal and the action, so the project acts on real risks and not on data artifacts or misread aggregates.

Produce two artifacts. First, the reading protocol: which indicators the project tracks (SPI, CPI, float consumption on the critical and near-critical paths, and the platform's forecast such as nPlan's projected completion), how often it reads them, the thresholds that move an indicator from green to amber to red as attention-direction signals rather than verdicts, and the interpretation step that follows every flag, the look that determines the cause before the response. Second, the data-discipline analysis: the statusing practices the indicators depend on, the honest, timely percent-complete and actual-cost booking, the checks that detect optimistic or stale statusing before it corrupts the indicators, and the rule that an indicator from suspect data is discounted and corrected at the source rather than acted on as a real signal, because the indicator inherits the data's reliability.

The deliverable is the reading protocol and the data-discipline analysis, and the lasting product is a project that reads its schedule indicators as the leading-indicator signals they are, getting the early warning the predictive engine provides, while interpreting the cause in context and keeping the data honest, so the indicators surface the real schedule risks early and the project responds to the causes rather than the symptoms. The executive who masters this gets the predictive engine's lead time without inheriting its two failure modes, the verdict-mistake that scrambles at every fluctuation and the data-mistake that trusts a confident forecast built on optimistic statusing, because the protocol places the human's interpretation between the signal and the response and the human's data discipline between the field and the indicator, which is where the metric-as-signal and data-quality disciplines together must live.

Key Takeaways

• Earned-value analytics is the predictive-ML engine (the third engine) applied to the schedule and budget: SPI (earned value over planned value) measures schedule performance, CPI (earned value over actual cost) measures cost performance, and float consumption tracks how fast the schedule spends its buffers, with platforms like nPlan forecasting the completion date from the current performance.
• These are leading indicators: a finish-date slip is a lagging fact learned too late, but SPI, CPI, and float consumption move weeks or months before the date is missed, giving the project the lead time to respond while response is still possible, the same earliness value as procurement risk tracking.
• A declining SPI is a signal, not a verdict (the metric-as-signal discipline): a 0.91 could be a stuck trade, an optimistic baseline, a procurement stall, or sloppy statusing, and the same number carries all these causes, so the human interprets the cause in context rather than reading the index as a sentence on the project.
• The predictive-attention posture: the indicator does not make the decision, it directs the human's limited attention to where a decision may be needed (the activity likely to slip, the path burning float), and the human applies the judgment the indicator cannot, neither substituting for the other.
• The indicators are only as good as the update data (the data-quality dependency): optimistic statusing inflates the SPI and hides real risk (a false negative, the dangerous direction), stale or pessimistic statusing manufactures phantom risk, so the project must keep the statusing honest and timely as the precondition for the indicators to mean anything.
• The human's interpretation must include the data: before reading a 0.91 as a real lag, the executive asks whether the update is current and the statusing honest, discounting indicators built on suspect data and trusting indicators built on clean data, because a confident forecast on bad data is the most dangerous false precision the engine produces.
• Float consumption is the quiet indicator SPI can hide: a project can hold a healthy aggregate SPI while one path burns the float that protects the finish, so the disciplined reader watches the path-level float burn as a first-class leading indicator of a future critical-path problem, not a footnote to the headline indices.
• The artifact: a schedule-risk reading protocol (which indicators, how often, the amber/red thresholds as attention signals not verdicts, the interpretation look after every flag) plus a data-discipline analysis (the statusing the indicators depend on, the checks for optimistic or stale data, the rule that suspect-data indicators are corrected at the source rather than acted on).