Business Intelligence · October 11, 2026 · By Andrew Neculai

Real Estate Agent Productivity Metrics: The Eight Numbers That Predict Who Leaves

Real estate agent productivity metrics that work as warning lights, not scorecards: eight leading numbers from your MLS and CRM, with thresholds and the churn arithmetic.

Two brokerage managers reviewing an agent performance dashboard on a laptop in a sunlit coastal office overlooking palms, dunes, a fishing pier and the Atlantic

Most brokerages learn an agent is leaving when the termination letter arrives, and by then the decision is months old. The real estate agent productivity metrics on the average broker's dashboard are closing sides and gross commission income, both describing work won a quarter ago. They are scorecards, not warning lights.

The gap is expensive. A 12-month analysis of 184,097 productive agents across four MLS regions by Recruiting Insight with Lone Wolf Technologies, reported by HousingWire on February 2, 2026, put 2025 agent turnover at 6.8%, up from 6.0%, with $15.72 billion in annual production changing brokerages. It also found internal transfers held 89% 12-month retention against 76% for agents recruited from another brand, and averaged $1.516 million in production against $1.218 million. Keeping a producer you already have beats replacing them on both cost and odds.

NAR's 2025 Profile of Real Estate Firms, released November 19, 2025, found agent retention was a top two-year challenge at 18% of firms overall, but 44% of three-office firms and 45% of those with four or more offices. The bigger you get, the less you see by walking the floor. What follows is the instrumentation that replaces the walk.

Table of Contents

  1. Why Real Estate Agent Productivity Metrics Miss the Thing That Matters
  2. The Eight Numbers, and What Each One Tells You First
  3. Scoring a 42-Agent Brokerage: The Arithmetic
  4. What Production Data Cannot Tell You
  5. A 90-Day Rollout, and What Not to Automate
  6. Frequently Asked Questions
  7. Ready to See Which Agents Are Drifting?

Why Real Estate Agent Productivity Metrics Miss the Thing That Matters

Closing volume is a lagging indicator: it reports a decision made 60 to 120 days earlier, so by the time it drops the pipeline that fed it is already empty. Useful agent productivity metrics measure what creates future closings, meaning appointments, signed agreements and database work, not sides. That is the same discipline that separates AI projects that pay from the ones that do not, covered in AI for real estate brokerages.

The second problem is that production distributions are not normal, so averages mislead. Loft47's 2026 State of Real Estate Commissions, drawn from its own back-office payout data covering 52,256 closed transactions at 204 brokerages and 6,729 agents with a deal in calendar 2025, reports a median US agent payout of $31,152 against a mean of $70,909, with the top 10% at $179,536 or more. Median solo production was five to six deals a year. When the mean is more than double the median, a brokerage average describes almost nobody.

Third, published benchmarks are contested, so measure against your own prior-year cohorts. A widely shared graphic claiming 71% of Realtors closed zero deals was addressed by NAR deputy chief economist Jessica Lautz in HousingWire on June 26, 2026: it is not NAR data, and NAR's 2026 Member Profile puts zero-transaction Realtors at 6% of those operating individually and 2% of those on teams. External figures swing by an order of magnitude depending on whether they count licensees, Realtors or agents with MLS activity.

The Eight Numbers, and What Each One Tells You First

Eight metrics, all of them already sitting in systems you pay for: the MLS, the CRM, the transaction management platform and the back office. Four lead production, four lead departure, and the overlap is the point.

  1. Signed-agreement rate, trailing 90 days. New listing and buyer representation agreements signed, not closings. The earliest honest read on next quarter. Signal: a drop of 40% or more against the same 90 days last year, which controls for seasonality on the Grand Strand and anywhere else with a season.
  2. Days since last new signed agreement. Counted from the agreement date, not the closing date. Signal: any agent past 60 days who closed business last year. Pending closings keep a drifting agent looking busy for two more months.
  3. Appointment-to-agreement conversion. Listing presentations and buyer consultations divided into agreements signed. Signal: a personal rate that falls by a third while appointment volume holds, which is a coaching problem, not a lead problem.
  4. Company-lead dependence. Share of closings sourced by the brokerage rather than the agent. Loft47 found agents self-source 86% to 95% of their business, with company leads at 5.5% to 14.3% of deals. Signal: agents above 50% dependence cut both ways, being the easiest for a competitor to outbid and the first to blame the house when leads slow.
  5. Database coverage with a next action. The share of an agent's CRM contacts carrying a scheduled next step and a date. Signal: coverage under 20%. NAR's 2026 Member Profile attributes 28% of business to repeat clients and 22% to past-client referrals, so an unworked database is half a book of business sitting idle. The number is only as honest as the data under it, which is why CRM data hygiene comes first.
  6. Listing-side share. Listings as a percentage of the agent's sides. Signal: a buyer-only mix in a slow market, where cycle times stretch and income stops. The income cliff for newer agents is steep. NAR's 2026 Member Profile reports median gross income of $59,200 in 2025, rising to $88,500 at 16 or more years of experience and falling to $8,000 at two years or less.
  7. Fall-through rate. Executed contracts that do not close, by agent, with a coded reason. Signal: a personal rate at 1.5 times the office rate across six or more contracts, which usually points at qualification or expectation-setting.
  8. File cycle time and compliance exceptions. Days from executed contract to complete file, plus items kicked back. Signal: a rising exception count, often the first visible symptom of an agent who has checked out, and a direct read on the admin load each side costs you.

Leading versus lagging, explicitly. Numbers 1, 3, 5 and 7 predict production; 2, 4, 6 and 8 track departure or distress. Sides and GCI stay on the report as outcomes to explain, never as the trigger for a conversation.

Scoring a 42-Agent Brokerage: The Arithmetic

Consider a hypothetical three-office brokerage with 42 producing agents, Loft47's mean US brokerage-retained revenue of $13,800 per agent (roughly $580,000 retained a year), and the 6.8% turnover rate above. Every assumption is replaceable with your own, and should be.

At 6.8%, expect about three departures a year. Replace all three externally and, at that study's 76% 12-month retention for external recruits, roughly 0.7 of the replacements leaves inside the year too. Call it 3.7 agent-years of disruption: about $51,000 at $13,800 retained per agent, or nearly 9% of retained revenue, on churn nobody forecast. That is the number worth a dashboard.

Now score. Give each agent one point per triggered signal from the eight above, then sort. A first run in an office this size usually splits three ways: a majority at zero or one point, a middle group at two or three, and a handful at four or more. The handful is the managing broker's conversation list for the month. The middle group gets a question in a one-on-one, not an intervention. Everyone else needs nothing, which is worth knowing, because the alternative is a manager spreading attention evenly and landing on no one.

The honest caveat. Three of 42 agents will leave, a base rate near 7%. If a five-name flag list catches two of those three, it also holds three who were never going anywhere: 40% precision. That is acceptable when the consequence is a conversation, and unacceptable the moment the list drives compensation, desk assignment or a performance plan. Low precision is only safe when the stakes are low.

What Production Data Cannot Tell You

The strongest known predictors of someone voluntarily quitting are not in your CRM, and modeling will not put them there. Rubenstein, Eberly, Lee and Mitchell's "Surveying the forest" in Personnel Psychology (2017) pooled 316 published articles and roughly 1,800 effect sizes across 57 predictors. Withdrawal cognitions, meaning the person is already thinking about quitting, corrected to ρ = .56, the strongest in the set. Job search behavior came in at ρ = .40, organizational commitment at ρ = −.29, job satisfaction at ρ = −.28.

Notice what that list is made of: things people tell you, in a survey or a conversation. Activity data is a weak proxy for them. So treat vendor claims of individual "flight risk scores" built from system logs carefully, and prefer cohort forecasting to individual scoring. Projecting that your newer-agent band will lose four people this year is a planning input. Telling an agent a model flagged them is a conversation you cannot win.

Three failure modes, and their early signals. The dashboard becomes a ranking: once agents can see each other's scores, the flags measure who games the CRM. Watch for a jump in next-action dates with no change in appointments. Pending volume masks the drift: a report built on closings shows a disengaged agent as productive until the pipeline empties, which is what metric 2 exists to prevent. The data is wrong and nobody says so: reconcile sides against the back office monthly, and expect the first three passes to find real errors.

What not to automate: the conversation. AI enhances human judgment here; it does not replace a managing broker. Use it to compile the weekly flag list and summarize six months of activity in a paragraph a manager can read before walking into the room. Never let it send the message, set a score an agent sees, or decide who gets support.

A 90-Day Rollout, and What Not to Automate

Build the measurement before the model, the process before the tooling. Process first, technology second is not a slogan here: a scoring rubric laid over inconsistent data produces confident nonsense.

  1. Days 1 to 30, define and reconcile. Write one-sentence definitions for all eight metrics and get two managers to agree on each. Pull 12 months of history and reconcile sides and GCI against the back office until they tie. Expect gaps in appointment and agreement dates, which most brokerages have never required.
  2. Days 31 to 60, instrument four. Start with signed-agreement rate, days since last agreement, company-lead dependence and fall-through rate, all four from systems of record rather than agent self-reporting. Run them weekly for a month with nothing attached, as a calibration period.
  3. Days 61 to 90, add the rubric and the ritual. Add the remaining four metrics, set thresholds from your own prior-year cohorts rather than published averages, and put a 20-minute monthly review on the calendar with one named owner. New agents get the same instruments on a shorter clock, pairing with the 90-day ramp.

Keep out of it: data entry agents get nothing from, composite scores nobody can decompose, leaderboards, and thresholds copied from a conference slide. If a manager cannot say in one sentence why an agent is on the list, the metric is not ready.

Frequently Asked Questions

How small is too small for this? We have nine agents. Nine agents is small enough that you already sense who is drifting, so build only metrics 1, 2 and 7: signed-agreement rate, days since last agreement, and fall-through rate. Those three take a spreadsheet and an MLS export. The dashboard earns its keep somewhere around 25 to 30 agents, or when you add a second office and lose line of sight.
Should agents see their own productivity scores? Show them the underlying metrics, not a composite score or a ranking. An agent who can see that they have gone 70 days without a new signed agreement has useful information. An agent who sees a 73 out of 100 next to a colleague's 81 learns only to manage the number. Transparency on inputs builds trust; transparency on rankings builds gaming.
Can AI predict which agents will leave? Not reliably at the individual level from activity data alone. The strongest predictors of voluntary turnover in Rubenstein and colleagues' 2017 meta-analysis of 316 studies are things like withdrawal cognitions and job search behavior, which live in conversations, not CRMs. Cohort-level forecasting works better: predicting how many agents in a tenure band will leave is a planning number you can act on.
What data do we need before any of this works? Three things: agreement dates as a required field, one agent identifier that matches across MLS, CRM and back office, and a coded reason on every fall-through. Most brokerages have the third nowhere and the second inconsistently. Fixing identity matching usually takes a week and unblocks everything else, so do it before buying any reporting tool.
How often should the flag list be reviewed? Weekly to generate, monthly to act on. Weekly keeps days-since-last-agreement current, because that metric degrades fast. Monthly action prevents managers reacting to normal week-to-week noise, which is most of what short-interval production data contains. Pair it with a quarterly review of the thresholds themselves against your own updated cohort data.

Ready to See Which Agents Are Drifting?

Most brokerages already hold every number in this article, in three systems that do not talk to each other. The work is defining the metrics, matching the identities, and building a review a manager actually runs. ANtreliX is a strategy-first AI and business intelligence consultancy in North Myrtle Beach, South Carolina, working with brokerages and teams along the Grand Strand and remotely. See our case studies or Talk to ANtreliX.

Business IntelligenceReal EstateData & AnalyticsPredictive Analytics

← All insights

Put these ideas to work in your business.

A conversation costs nothing. It starts with your business problem — not with technology.

Start a conversation