A dispatcher in 2026 sits between two boards. The first is the one on the wall — tech tiles, color-coded job cards, the red "EMERGENCY" tag that flashes when an old lady's furnace dies in January. The second board is invisible. It lives inside ServiceTitan Dispatch Pro, Sera Systems' scheduling engine, or FieldEdge's auto-router. It re-solves the entire day every 10 minutes using predicted job revenue, tech historical close rate, travel cost, and capacity — math the human dispatcher has been doing in their head for 20 years with imperfect information and a 4-hour caffeine half-life. The shop owners who lift dispatch yield 12-18% in 60 days are not the ones who turn Dispatch Pro on and walk away. They are the ones whose dispatchers learn to read the invisible board like they already read the wall — knowing what it sees, knowing what it cannot see, and writing override criteria themselves so the AI's mistakes get caught before they cost $4,000 in install margin. This lesson is the dispatcher's manual for the invisible board: what Dispatch Pro is actually optimizing for, what Sera's profit-aware flip changes, how FieldEdge weighs travel against capacity, and the override-criteria document every dispatcher should write in their own handwriting by Friday of week one.

The Four Variables on the Invisible Board

The first thing to understand about ServiceTitan Dispatch Pro is that it is not magic and it is not a calendar app. It is fast regression on four variables, run every 10 minutes across every open call and every available tech. The four variables are predicted job revenue, tech historical close rate, travel cost, and capacity. Strip the marketing language off Dispatch Pro and that is what it is. A dispatcher who reads the board through those four lenses can predict 80% of what the model will propose before the screen refreshes.

Predicted job revenue is what this call type has historically generated at this specific shop, adjusted for system age, equipment type, customer history, and time-of-day patterns. The model knows that "no cool" calls in July at a 14-year-old condenser address at this shop close at $8,400 on average. It knows that drain calls at apartment complexes close at $312. It knows the variance — the no-cool call has a fat tail (some close at $24K because the homeowner replaces the whole system) and the drain call has almost none. When a Bel Air homeowner calls at 11:45 a.m. with a no-blower complaint on a 14-year-old furnace, predicted job revenue is around $4,800 with a tail to $14K. When the same dispatcher takes a basic drain call from an apartment-complex management contract at 11:50 a.m., predicted job revenue is $290. Dispatch Pro sees both numbers; the wall-board does not.

Tech historical close rate is the documented close on this call type for this tech at this shop. Jose closes no-cool calls at 38% MPR and $9,200 average ticket — high replacement attach, strong financing close. Marco closes them at 22% MPR and $4,100 average ticket — solid on linesets, weaker on the kitchen-table replacement conversation. The rookie closes at 14% MPR and $1,800 average ticket. The model multiplies predicted revenue by close rate to get expected revenue per assignment: send Jose to the Bel Air no-blower and expected revenue is $3,500; send Marco and expected is $902; send the rookie and expected is $252. The math is dispassionate. The dispatcher's gut "Jose closes Bel Air" is, in fact, the model's math.

Travel cost is the weighted cost of driving time from the tech's current location to the next stop. Weight components: fuel, lost billable time at the tech's fully-loaded hourly cost ($80-$120 in 2026), customer-impact (a 90-minute travel window means a missed time-window commitment). Dispatch Pro reads tech GPS and Google Distance Matrix in real time and updates travel weight every 10 minutes. A tech 45 minutes out gets penalized; a tech 6 minutes away gets boosted.

Capacity is the tech's remaining hours, the install crew's stacked dependencies, the recall return windows, and time-window commitments already booked. Capacity prevents the model from over-booking the highest-EV tech — Jose may have the highest expected revenue on five different calls, but he only has eight hours and three are committed. Capacity forces the model to route at least some calls to the rookie and Marco.

The output is a ranked recommendation per open call. Send Jose to the Bel Air no-blower. Push the apartment drain to the 2 p.m. slot. Hold the warranty callback for tomorrow because every available tech today is revenue-disadvantaged. The recommendation refreshes every 10 minutes as reality changes. The dispatcher decides whether to accept or override.

The 10-Minute Cadence and Why It Matters

The re-evaluation cadence is the differentiator and the part that surprises veteran dispatchers the most. Old-school dispatch logic re-runs at major events — when calls land, when techs finish, when emergencies hit. Dispatch Pro re-runs the entire board every 10 minutes regardless of whether an event occurred. That distinction matters because the shop's reality changes every 10 minutes whether the dispatcher noticed or not.

At 10:30 a.m. the rookie finishes a tune-up 12 minutes early because there was nothing to fix. The wall-board doesn't update; the dispatcher is on the phone with a parts supplier; the rookie sits in his truck for 18 minutes scrolling Instagram. The invisible board re-evaluates at 10:30, sees the rookie's location update, sees an open drain call 7 minutes away, sees the rookie's drain close rate, and proposes the assignment. The dispatcher gets a flag; she approves; the rookie is on a paying job at 10:32 instead of 11:15. 43 minutes of recovered labor. At $90/hour, that is $65 of recovered cost on one tech on one transition.

Multiply across a 7-truck day. Three or four of those reshuffles per truck per day, each worth $50-$100 of recovered labor and incremental billable revenue. That is the source of the 12-18% dispatch yield lift in published 2026 deployment data. The model doesn't make smarter individual decisions than a great dispatcher; the model makes them more often. Cadence beats peak intelligence on a board that changes every 10 minutes.

The implication for the dispatcher's day is structural. Pre-Dispatch-Pro rhythm: 6:30 a.m. board build, 7 a.m. tech huddle, then react. With Dispatch Pro live: 6:30 a.m. board build (Dispatch Pro proposes, dispatcher reviews), 7 a.m. tech huddle, then read the AI's proposed reshuffles every 10 minutes and approve or override. The dispatcher stops reacting to events; she starts reviewing model proposals. Pace calmer, output sharper. Mistakes from "I didn't notice the rookie was free" disappear.

When the Dispatcher Overrides, and How

The 12-18% dispatch yield lift is not the lift from Dispatch Pro alone. It is the lift from Dispatch Pro plus disciplined human override. Shops that turn on Dispatch Pro and accept 100% of its recommendations see a lift on day one and a decline by day 30 as the model's blind spots compound. Shops that override without logging see no model improvement over months. The 12-18% lift lands only at shops where the dispatcher overrides 5-15% of recommendations with a one-line documented reason that feeds back into the model and the comp-plan review.

Five categories of override sit permanently in human hands. Comp plan dynamics. Marco is on a PIP — one more low-revenue day and he is gone. Dispatch Pro doesn't see PIP status. Jose is one close away from his bonus tier kicking in this month; the comp curve isn't in the model. The dispatcher overrides to give Marco a winnable mid-revenue call to rebuild him, or to push Jose toward the high-EV no-cool call that triggers his bonus. This is humans managing humans, not metrics managing humans, and it is permanent.

Install crew gross margin. Dispatch Pro proposes pulling Marco off the Marin Park install at 1 p.m. to handle a service call at 1:30 because the predicted-revenue × close formula says service > install on that hour. The dispatcher knows Marco's hands on the lineset save the install crew 90 minutes of rework risk, and that the install's lineset margin alone is $3,800 on the day. Pulling Marco costs more than the service call earns. The model can't see install-crew opportunity cost because install-crew margin lives one layer below job-level revenue data; the dispatcher protects the install crew.

Recall risk by originating tech. The recall on the Bel Air condenser needs to return to Marco because Marco installed it and he remembers the install context. Sending the rookie to a recall doubles re-callback probability because the rookie doesn't know what the original install looked like before the part failed. Dispatch Pro sees the call as "service capacity available, route to closest available tech" — it doesn't know the originating tech. The dispatcher routes recalls by origin, not by capacity, every time.

Customer request specificity. "Send Jose. I want Jose." Customers who have had positive experiences with specific techs ask for them by name. The retention math on customer-tech matching at request shows up in second-year revenue per customer; the model doesn't see customer-tech history at the request-respect layer. The dispatcher honors the request.

Unseen variables. The dispatcher knows the weather (a 3 p.m. thunderstorm will cut outdoor diagnostic time), the traffic (the I-40 closure is bottlenecking south-territory routes), the rookie's mood (he had a bad call at 9 a.m. and needs a quick win before lunch), the supplier dynamic (the parts truck delivers at 11 a.m., so Marco's 11:30 install can start on time only if he's already at the shop), and the construction project blocking Lane Street. None of those are in the model. All of them affect the day.

Writing the Override-Criteria Document Yourself

Every dispatcher should write their override-criteria document themselves by Friday of week one with Dispatch Pro live. The document does two jobs: it disciplines the dispatcher's override reasoning (forcing them to articulate the rules), and it feeds the model the shop-specific patterns the model needs to learn. The document lives in a shared Google Doc or a pinned Slack post. It gets updated every Friday at the operations huddle when override patterns reveal new categories.

The document has a simple structure: one section per override category, with three subsections per category — what triggers the override, what the dispatcher does, what the model should be learning. Here is what the install-crew section looks like at a 7-truck shop after week three of Dispatch Pro:

Install Crew Override. Trigger: Dispatch Pro proposes pulling a named install-crew member (lineset specialist, ductwork lead, controls integrator) off a stacked install for a service call where the service call's expected revenue is less than $1,800. Action: Override; protect the install. Document reason as "install crew margin protection — pulling [name] off [install address] costs [estimated $] on lineset/duct/controls margin." Model learning: install-crew composition refreshes daily; opportunity-cost coefficient for each install role is shop-specific and rises during heavy install seasons (Apr-Jun, Sep-Nov in this market). Re-train model awareness via daily 7 a.m. install-crew tag in ServiceTitan dispatch notes.

That paragraph, written in the dispatcher's own words, is more valuable to the model and to the shop than any vendor-published playbook. It captures shop-specific variables — install-crew composition, seasonal install patterns, named crew roles — that no vendor knows. After 90 days the document has five categories built out, the override rate has dropped from 15% to 9% as the model learns the documented patterns, and the dispatcher's Friday review is no longer a brain dump but a structured walk through five sections with new trigger examples added under each.

Sera Systems and the Profit-Aware Flip

Sera Systems' 2026 dispatch AI runs a different bet than Dispatch Pro. Where Dispatch Pro optimizes the four-variable formula and accepts the expected-revenue ranking that results, Sera optimizes against a single target metric: revenue per tech. The bet underneath is that call-count optimization tolerates low-margin calls — the AI fills the day with billable activity, the day's margin is thin, and the shop runs profitable but not exceptional. Revenue-per-tech optimization explicitly de-prioritizes filler work when a higher-revenue alternative is reachable, even at the cost of leaving a truck temporarily idle.

In practice the difference shows up on slow days. At 1:30 p.m. on a slow Tuesday with three available techs and four open calls — one Bel Air replacement quote, two service tune-ups, one warranty callback — Dispatch Pro's call-count instinct is to fill the slots: each tech runs a call, the day is full at three completed calls. Sera's profit-aware instinct may hold one tech for a higher-revenue call expected to land in the afternoon based on call-volume patterns at this shop, dispatch only the Bel Air to the highest-close tech, and route the two tune-ups in a clustered single-tech run for routing efficiency. The fewer-calls-but-higher-revenue outcome is the Sera bet. The shop's quarterly RPT is the metric that proves the bet right or wrong over a 60-day window.

The dispatcher in a Sera shop has a different conversation with the truck. "Marco, holding you 45 minutes — expecting a no-cool call in the south based on volume pattern. Catch up on the Bel Air write-up." Marco's instinct is "send me on a call, I want to be working." The dispatcher's discipline is to defend the hold because the shop's revenue-per-tech metric requires it. Sera's bet only works when the dispatcher defends the hold and the owner sponsors the bet against the "why is Marco sitting" reflex from the floor.

The shop-by-shop pick between Dispatch Pro's call-count bet and Sera's revenue-per-tech bet often follows shop economics. Stable-ticket high-volume residential service with 60-80 calls/day and average ticket $450-$600 — Dispatch Pro's call-count optimization captures more of the available volume at predictable margin. Replacement-heavy shops with significant revenue spread (service $400 to replacement $14K, average ticket $1,200 with high variance) — Sera's revenue-per-tech holds techs for higher-revenue calls and de-prioritizes the thin-margin filler. The bake-off question lives in L4 Ch 2; the role-level mental model is here.

FieldEdge Auto-Routing and the Windshield-Time Equation

FieldEdge's auto-routing AI operates on a third axis: travel time and geographic clustering. FieldEdge's bet is that on a service-heavy day with 5-7 trucks across a metro service area, the difference between a thoughtful route and a chaotic route is 60-90 minutes of windshield time per truck per day. At a $80-$120 fully-loaded hourly cost per tech, 5 trucks × 75 minutes/day × 250 days = ~$200K of annual recoverable labor cost — before counting the additional billable calls a 75-minute time savings enables. FieldEdge's 2026 outcome data shows 15-20% windshield-time reduction at shops with disciplined override protocols.

The mechanic: FieldEdge clusters calls geographically, weights against tech skill tags (a tech who is fast on heat pumps gets the heat pump cluster), respects time-window commitments (the 10 a.m. confirmed slot must run at 10 a.m. regardless of clustering optimization), and re-routes when the day's mix changes. The model recomputes the route every time a call lands, completes, or shifts time. The dispatcher's override authority sits in two places explicitly: stacked install crews (do not break the install crew for service routing efficiency, same rule as Dispatch Pro) and recall returns (the recall goes back to the originating tech, period). Both overrides are documented in FieldEdge's job notes the same way Dispatch Pro overrides are documented in ServiceTitan.

FieldEdge's clustering is most powerful in geographies with high service density — metro areas, dense suburban corridors, multi-zip service zones. It struggles in rural shops where calls are 25-40 minutes apart by definition and the clustering math has little to work with. Rural shops lean toward Dispatch Pro's expected-revenue ranking; suburban metro shops lean toward FieldEdge's clustering because travel savings dominate call-mix variance.

The shop-by-shop pick between Dispatch Pro, Sera, and FieldEdge often follows FSM platform: ServiceTitan shops use Dispatch Pro; Sera-platform shops use Sera natively; FieldEdge shops use FieldEdge auto-routing. The 59% in-software AI preference in ServiceTitan's 2026 State of AI in the Trades report shows up most strongly in dispatch — bolt-on dispatch AI is unusual because the integration depth required is hard to maintain across a vendor boundary.

Reading the Board in Trades English

The dispatcher's reflex when Dispatch Pro lands is "the AI is taking my job." The honest answer is the role evolves — the dispatcher becomes the exception handler, the override authority, and the comp-plan-vs-yield interpreter. Reading the board in trades English means knowing what each variable on the screen translates to in real money. "Jose to Bel Air, expected revenue $3,500" reads as "this is the call with a 38% chance of being a $9,200 no-cool replacement and Jose closes 38% of those — Marco closes 22%, sending Marco costs $1,200 today." "Rookie to apartment drain, expected revenue $290" reads as "this is a $290 call no matter who runs it, and it teaches the rookie the apartment playbook for the next 200." "Hold Marco for the afternoon" reads as "Sera-style profit-aware hold against a predicted high-revenue call, justified by 30 Tuesdays of volume data." The translation is what turns the AI from a black box into an apprentice: not compliance ("I did what the AI said"), not rebellion ("the AI is wrong"), but collaboration ("the AI sees this, I see that, here is the decision"). Shops that train the translation in week one capture the 12-18% lift; shops that drop dispatchers into the screen stay at baseline.

The Friday Override Review

The cadence that makes Dispatch Pro improve over months is the Friday override review. Every Friday at 4 p.m., the dispatcher pulls the week's override log — every time she overrode the AI's recommendation, with the documented one-line reason — and walks through the patterns in a 25-minute session with the service manager or operations manager. The output is updated override criteria, comp-plan adjustments, and model-feedback notes.

The patterns surface quickly. Week one: dispatcher overrode 14 times for install-crew protection, all reasonable. Week two: 11 install-crew overrides plus three new patterns — a tech requested by name three times by repeat customers (customer-request override), one recall misroute caught (recall-by-originating-tech), one weather override on an outdoor diagnostic. Week three: 9 install-crew overrides as the model starts learning install patterns, plus the three week-two categories repeating, plus a new comp-plan override (Marco on PIP). The Friday review captures each new category and adds it to the override-criteria document. By week six, the override rate has dropped from 15% to 11% because the model is learning install patterns; by week 12, it has dropped to 8% as the model learns the comp-plan and recall patterns.

The Friday review also feeds comp adjustments. If the dispatcher overrode 9 times to give Marco rebuild calls and the PIP situation persists, that is the comp plan and the coaching plan, not just the dispatch decision. If the dispatcher overrode 3 times to honor customer name requests for Jose, that is data on customer-tech retention — feeds the photos and bios on the on-my-way text (L2 Ch3 Lesson 3's territory). Shops that skip the Friday review never improve; override stays at 15% forever. Shops that run it on schedule see override drop, dispatcher strategic capacity rise, and dispatch yield compound quarter over quarter. 25 minutes a Friday is the operational difference between a $5M shop and a $7M shop.

Key Takeaways

• Dispatch Pro is fast regression on four variables run every 10 minutes: predicted job revenue, tech historical close rate, travel cost, and capacity. The dispatcher who reads the board through those four lenses predicts 80% of what the model proposes before the screen refreshes.
• Cadence beats peak intelligence. Dispatch Pro doesn't make smarter individual decisions than a great dispatcher; it makes them more often. Three or four 10-minute reshuffles per truck per day, each worth $50-$100, is the source of the published 12-18% dispatch yield lift in 60 days.
• The 12-18% lift requires override discipline, not Dispatch Pro alone. Shops that accept 100% of recommendations decline by day 30; shops that override without logging stay flat; shops that override 5-15% with one-line documented reasons capture the lift and the model learns over months.
• Five permanent override categories: comp plan dynamics (PIP, bonus tier), install crew gross margin (lineset/duct/controls opportunity cost), recall risk by originating tech (return to the install hands), customer request specificity ("send Jose"), and unseen variables (weather, traffic, mood, supplier, street closures).
• Sera Systems' profit-aware bet is revenue per tech, not call count. On slow days, Sera holds techs for predicted high-revenue calls and de-prioritizes filler. Better fit for replacement-heavy shops with ticket variance; Dispatch Pro better fits stable-ticket high-volume residential service.
• FieldEdge auto-routing cuts windshield time 15-20% on service-heavy metro days. Most powerful in dense suburban corridors; weakest in rural shops where call spacing dominates clustering math. Dispatcher overrides on stacked installs and recall returns are the same as Dispatch Pro.
• Write the override-criteria document yourself by Friday of week one. One section per category, three subsections per section: trigger, action, model learning. Captures shop-specific variables (install-crew composition, seasonal install patterns, named crew roles) no vendor playbook knows.
• The Friday override review is the cadence that compounds. 25 minutes, structured walk through the week's override patterns, updated criteria, comp adjustments, model-feedback notes. Override rate drops 15% → 11% → 8% over 12 weeks at shops that run it; stays at 15% forever at shops that skip it.
• Reading the board in trades English is the apprentice frame. Not rebellion ("the AI is wrong"), not compliance ("I did what the AI said") — "the AI sees this, I see that, here is the decision."