--- title: "Validated vs. Directional: The Labor Savings Number That Survives an Executive | WFM Labs" description: "Most labor savings numbers die in the room they are presented in. The discipline that makes one hold up: two ledgers, re-derived standards, and nothing summed." url: https://wfmlabs.ai/blog/labor-savings-number-survives-an-executive source: wfmlabs.ai — independent UKG Pro WFM engineering practice --- [Blog](https://wfmlabs.ai/blog) / Validated vs. Directional: The Labor Savings Number That Survives an Executive # Validated vs. Directional: The Labor Savings Number That Survives an Executive Jeff Bugbee · August 10, 2026 - Labor analytics - Labor standards - Workforce analytics - UKG Pro WFM The short answer A labor savings number survives executive scrutiny when what the actuals prove is kept in a separate ledger from what a model estimates, and the two are never added together. Re-derive standards from your own timestamps, exclude confounded periods instead of explaining them, backtest before presenting, and let the smaller validated number carry the decision. The number dies in the room. Somebody presents $6M of labor opportunity, an operator in the third row says “that period included the relocation,” and within ninety seconds the conversation has stopped being about labor and started being about whether the analysis is any good. Nothing gets funded. Everyone agrees to “revisit the methodology.” I’ve watched versions of that meeting more than once, and the failure is almost never arithmetic. It’s that a single soft assumption was load-bearing for the entire figure, and one person who knows the operation could find it in a breath. The fix isn’t a better model. It’s refusing to blend two different kinds of claim into one number. ## What actually separates a validated number from a directional one? **Validated** means your own actuals already demonstrated it. Sites in your network, under comparable conditions, in your own history, ran the process in fewer hours. The savings figure is the gap between what your better periods proved possible and what the rest of the network spent. There is precedent for every hour in the number, and it lives in data the operator can go look at. **Directional** means a model produced it. A labor standard, a queueing assumption, a capacity simulation, a vendor benchmark. Directional numbers are legitimate and often larger — they’re how you size an opportunity nobody has captured yet. But they inherit every assumption underneath them. Both belong in the analysis. What kills credibility is adding them together. The moment $1.4M of demonstrated savings is summed with $4.6M of modeled savings into “$6M of opportunity,” the whole thing carries the confidence level of its weakest component. **Two ledgers, always. Report the validated number as the commitment and the directional number as the exploration, and never let a slide show one total.** Executives are good at this distinction, in my experience, and they’re grateful when you make it for them. What they resent is having to find the seam themselves. ## The six rules that make the number hold I derive these numbers in PySpark against a client’s own punch, schedule, and volume history — [not against a benchmark deck](https://wfmlabs.ai/services/analytics-data-platforms). Six practices do most of the work of making the output survive contact with the people who run the sites: **1\. Re-derive the standards. Don’t inherit them.** The earned-hours standards a company already has were usually set years ago by someone who left, and they encode a process that has since changed. Rebuild time factors from actual process timestamps in the timekeeping and operational data. When the re-derived standard disagrees with the documented one, that gap is itself a finding — and it’s better to surface it yourself than to have an operator surface it for you. **2\. Exclude the confounded period. Don’t explain it.** A remodel, a system cutover, a hurricane week, a site that changed its operating model in March. The instinct is to keep the data and annotate it. Don’t. Every annotation is an invitation to argue, and a finding built on 46 clean weeks is stronger than one built on 52 weeks with a footnote. I’d rather hand back a smaller number with nothing in it anyone can pull on. **3\. Segment by percentile, not fixed thresholds.** “Sites above 1.15 earned-to-actual” is a rule that expires. Networks change — sites open, volumes shift, the mix moves — and hardcoded bands quietly reclassify half the estimate a year later. Percentile-based segmentation defines performance relative to the network as it currently exists, so the analysis stays true when the network moves. **4\. Compare like with like.** A site’s earned-to-actual ratio is meaningless against a site with different volume, different hours of operation, and a different labor mix. Residual performance against comparable peers is the honest comparison, and it’s the one that pre-empts “you can’t compare us to them” — which is otherwise the second question you’ll get. **5\. Model the wind-down before you count the savings.** Hours don’t come out of a network the day a report is delivered. There’s a transition period, a training cost, a stretch where a site runs both ways. Netting that out shrinks year-one savings, and it is exactly why the remaining number is believable. **6\. Backtest before you present.** Hold out history, apply the finding, check it against what actually happened. If the method would have predicted the past badly, you’ve learned that privately instead of publicly. ## Doesn’t this just make the number smaller? Yes. That’s the point, and it’s worth being direct about the tradeoff rather than pretending there isn’t one. A validated-only number is smaller than the blended figure a competing proposal will show. It will look less impressive in a bake-off. What it does instead is convert: it’s the number a CFO can put in a plan without personal exposure, and it’s the number that’s still standing in month six when someone asks whether the program delivered. There’s also a compounding effect people underestimate. The first number you present sets the credibility of every number you present afterward. A conservative figure that lands intact buys you the right to bring a larger, more speculative one next quarter and have it taken seriously. ## What this looked like in the field A national biopharma services network with several hundred US sites needed labor savings quantified ahead of a workforce management program. The engagement re-derived earned-hours standards from the sites’ own process timestamps rather than the inherited documentation, segmented sites by percentile against comparable peers, and excluded periods with known operational confounders. The output was **$1.4M in actuals-validated annual savings**, reported separately inside a larger portfolio of modeled opportunity — the validated ledger and the directional ledger side by side, never summed. Roughly **30% of sites carried about 80% of the total opportunity**, which turned “improve labor efficiency” into a prioritized list of where to start. The part that mattered downstream: because the analysis ran in code against real operational data, the findings handed off as configuration requirements for the build — labor standards, thresholds, site groupings — rather than as a recommendation deck someone would have to re-derive later. That’s the same reason the [warehouse layer is worth building properly first](https://wfmlabs.ai/blog/ukg-data-hub-warehouse-analytics): everything here depends on having governed, trustworthy hours to compute against. ## When NOT to do this - **You need a rough order of magnitude to decide whether to look further.** A directional model is the right tool and it’s cheaper. Just label it directional and don’t let it get quoted as a commitment. - **Your timekeeping data isn’t trustworthy yet.** If punches are unreliable, job transfers aren’t captured, or half your sites are on a different process, fix the source. No analytical discipline rescues data nobody recorded. - **The decision is already made.** If the program is funded and the question is “where do we start,” skip the quantification and go straight to prioritization and standards work. - **Nobody will own the number.** A savings figure with no operating leader accountable for it is a slide. Find the owner before you find the number. --- _If you have a labor savings number you’re not sure would survive its first hard question, [send me the methodology](https://wfmlabs.ai/contact) — I’ll tell you straight which parts of it are validated and which are directional._ ## Common questions What is the difference between validated and directional labor savings? Validated savings are demonstrated in your own historical actuals — hours that comparable sites already achieved under comparable conditions, so the number has precedent inside your own network. Directional savings come from a model: a standard, a simulation, a capacity assumption. Both are useful. They answer different questions and carry different confidence, which is why they belong in separate ledgers and must never be summed into one headline figure. How do you quantify labor savings from UKG Pro WFM data? By deriving it in code against your own punch, schedule, and volume history rather than applying a benchmark. Re-derive earned-hours standards from actual process timestamps, compute earned-to-actual by site and daypart, segment sites by percentile rather than fixed thresholds, exclude periods with known confounders, and backtest the finding against held-out history before anyone sees a slide. Why do labor savings estimates fall apart in executive reviews? Usually because one soft assumption is load-bearing for the whole number. An operator names the anomalous quarter, the inherited standard nobody has validated in years, or the frictional cost of actually removing the hours — and because everything was summed into a single figure, the entire estimate loses credibility, not just the weak part of it. ## Working on something like this? I build custom UKG Pro WFM solutions — integrations, analytics, custom apps, and AI-enabled tools, including the ones that reach well past UKG. Let’s talk through your scenario. [Get in touch](https://wfmlabs.ai/contact)