A solar floor is a small piece of telecom infrastructure with a sales team attached. Most of the ways it fails are configuration problems that present as people problems — connect rates fall, managers coach harder, and the actual cause is a caller ID that has been flagged for three weeks or a list loaded in an order nobody chose.
This is the build order. It is an order rather than a list because each layer constrains the one above it, and floors assembled out of sequence end up rebuilding.
Layer 1 — Telephony
Everything downstream inherits from this. The decisions worth making deliberately:
| Setting | What to choose | Why it bites later |
|---|---|---|
| Carrier relationship | One primary with a documented path for reporting mislabelled numbers | When numbers start getting tagged, the remedy runs through the carrier. A reseller with no escalation path leaves you rotating numbers forever |
| Call attestation | Confirm your traffic is signed at the highest attestation your carrier supports | Unsigned or low-attestation traffic is the raw material for spam-likely labelling on the receiving network |
| Number inventory | A managed pool, sized to volume, with per-number health tracked | Untracked numbers make it impossible to distinguish a bad list from a burnt caller ID |
| Recording | On, with retention set to your policy and consent handled per jurisdiction | Recordings are the only way to audit dispositions, and the only evidence when a complaint arrives |
Layer 2 — Dialing mode
Three modes, and the choice is mostly a function of floor size and record value.
- Preview. The agent sees the record, then the system dials. Slowest, and correct when records are expensive or the conversation needs preparation — aged files, callbacks, anything where the agent should know the history before the homeowner says hello.
- Power. A fixed number of lines per available agent. Predictable, controllable, and the right default for most solar floors. The ratio is a dial the manager owns and should move deliberately, not a number set once during install.
- Predictive. The system paces based on estimated agent availability. Genuinely more efficient at scale, and genuinely hazardous below it, because the estimate needs volume to be an estimate rather than a guess.
Whatever the mode, abandonment is the number that has consequences beyond efficiency. Set a ceiling, monitor it as a first-class metric, and treat any pacing change as an experiment with that ceiling as the stop condition.
Layer 3 — Caller ID
The homeowner decides whether to answer based on ten characters on a screen. Three things matter, in this order:
- The number is not flagged. Everything else is irrelevant if the handset says spam. Check your numbers against the major analytics providers on a schedule — not when someone complains that connects dropped.
- Caller name is registered. An unregistered number displays as nothing, and nothing looks like a robocall.
- The area code is plausible. Local presence works because people answer familiar prefixes. It stops working when the same prefix is used at a volume no local business could produce, so treat it as a resource to manage rather than a trick.
Rotation is a symptom management strategy, not a fix. A floor rotating numbers weekly has a volume-per-number problem, a list problem, or a complaint problem, and rotation only hides which.
Layer 4 — List plumbing
This is where dialer setup meets data, and where most of the recoverable performance actually lives. The dialer needs to know three things about every record that arrives: where it came from, when it was last touched, and what happened.
- Load in a chosen order. Not the order the file arrived in. Sorting by anything correlated with reachability — recency of the underlying event, phone type, prior activity — changes the shape of the first hour.
- Segment by source, permanently. A blended file cannot be evaluated. If two vendors are mixed at load, you can never fire the worse one.
- Attempt limits per record, per campaign, per day.Enforced in the platform, not by policy. Anything enforced by policy eventually is not enforced.
- Callback windows that respect local time. A national list dialed on the floor’s local clock will call the west coast too early and the east coast too late, every single day.
The hygiene routine that keeps those files dialable — dedupe order, suppression, dead-number replacement, reload cadence — is a discipline in its own right and is covered in list hygiene for call centers. The regulatory side of who may be dialed at all is covered separately in DNC compliance for outbound solar; set it up before the first campaign rather than after the first complaint, and take your own legal advice.
Layer 5 — Dispositions
Disposition design is the most consequential ten minutes of the whole build, and it is usually done by accepting a vendor default.
The rule: one disposition per outcome, and no disposition that two agents would use differently. A list with both “not interested” and “no” produces two piles that mean the same thing and can never be compared. A list without a distinct code for a wrong party makes it impossible to measure data quality separately from agent performance — which is the single most useful split a floor can have.
Minimum viable set:
- Contacted — right party — appointment set
- Contacted — right party — declined
- Contacted — right party — callback scheduled
- Contacted — wrong party
- No answer / machine
- Bad number
- Do not call requested
Seven codes an agent can pick correctly under pressure beat thirty that get chosen by proximity on the screen.
Layer 6 — The four numbers
Everything else is reporting. These four are management.
| Metric | What it tells you | What it does not |
|---|---|---|
| Connect rate | Whether the numbers are live and the caller ID is trusted | Whether the person who answered is the homeowner |
| Right-party rate | Whether the data actually matched the household | Anything about the agent |
| Set rate off right party | Agent and script performance, isolated from data quality | Whether the appointment will hold |
| Held rate | Whether the sets are real | Nothing about dialing — it is a downstream number and belongs to confirmation process |
Connect rate on its own is the most over-watched number on a floor. Split from right-party rate, it stops being a mystery: one of them is a telephony problem and the other is a list problem, and they get fixed by different people. What happens after the set is a separate craft entirely — see solar appointment setting for the qualification and handoff that decide whether a set survives to a sit, and how to price a bought lead for what the whole chain can afford to pay per record.
The order matters
Floors built top-down — script first, dispositions later, telephony whenever the connects drop — spend their first quarter unable to tell which layer is broken. Built in this order, each layer is measurable before the next one is added, and a bad week has a diagnosis instead of a mood.
Frequently asked questions
Predictive or power dialing for a solar floor?
Power dialing until the floor is large enough for predictive pacing to have anything to predict. Predictive algorithms estimate agent availability from throughput, and on a small floor the estimate is noise — which shows up as abandoned calls, the one failure mode with regulatory consequences attached. Most solar floors are better served by a fixed ratio they control than by pacing they do not.
How many numbers does an outbound floor need?
Enough that no single number carries a volume that looks automated, and few enough that each one accumulates a reputation you can actually monitor. The failure mode at both extremes is the same — calls stop connecting — but the diagnosis differs: too few numbers get flagged, and too many make it impossible to tell which ones were.
Should answering-machine detection be on?
It is a trade, and floors that treat it as free lose contacts without noticing. Detection works by listening to the first moments of an answered call and deciding whether a human said hello. Tuned aggressively it discards live humans who answer slowly, quietly or with an unusual greeting — and those calls never appear as a loss, they appear as a machine. Audit it against recordings before trusting the disposition counts.
What is the first thing to fix on an underperforming floor?
The list, nearly always. Dialer configuration changes how efficiently you reach whoever is in the file; it cannot put someone reachable in the file. Managers reach for pacing and scripts first because those are the controls they own, which is why so much effort gets spent optimising the delivery of calls to numbers that no longer belong to the homeowner.