Rebuilding a national homebuilder's sales operation

Starting from what management could not see

The engagement began with an onsite discovery across three sessions — wholesale and packages, sales management, and marketing — because the presenting complaints were symptoms and it was not yet clear what the operation actually looked like.

What came out of it was a picture of a business managing itself by hand on top of a system it had already paid for. To prepare a Monday one-on-one, a manager opened a consultant's record and read calls and deal bins individually; there was no single page that showed a week. The consultant KPI board was produced from the CRM, exported, and re-keyed into a spreadsheet, and that spreadsheet — not the CRM — was what consultants checked morning and evening.

The root cause of the visibility gap was structural rather than cosmetic. Teams existed in the CRM but were being used only to drive an online-lead rotation, with people added and removed depending on how busy their display centre was. It was a routing list, not a hierarchy, so no manager rollup could exist on top of it.

The stated management principle was "if it is not in the CRM it did not happen". That is the right principle and it was not yet safe to enforce, because several things that had happened were not in the CRM and several things in the CRM had not happened.

Finding out the routing had been decorative for years

A complaint arrived that leads were not reaching the right regional consultants, alongside a visibility issue where two people could suddenly see records they should not. The easy move was to reverse a recent team change, declare it fixed and move on.

It was not the cause. An audit of all ten lead-assignment workflows, plus the user, role and team model, showed the visibility came from a permission set scoped to all contacts — unrelated to teams and predating the engagement entirely. A separate view setting was exposing 76 ownerless deals.

The routing finding was larger. All three round-robin workflows branched four ways by territory and then rotated into only two pools. Two territories fed one pool and two fed another. Regional routing had never worked the way the business believed, for as long as anyone could remember.

It was rebuilt by deriving the territory map from actual record ownership over the current year rather than from the documented territory list — which corrected two entries the documentation had wrong — then repointing nine rotate actions with the existing pools kept alongside the new ones so rotation never emptied mid-change. The full technical account is written up separately in Four branches, two pools, and nobody knew.

Controls that were enabled and had never run

Three workflows policing deposit-stage data completeness were switched on, correctly configured by inspection, and had never enrolled a single record.

The cause is a platform behaviour with no error surface: association absence cannot be expressed the way it had been written. One API rejects the filter outright; the other accepts it, stores it, and never runs it. Nothing anywhere reports a problem.

They were rebuilt against a pattern that does work, verified with throwaway lists showing 453 of 662 deposit deals carried the required association, and date-bounded so that enabling them did not retro-blast the entire history. That detail matters more than the fix — a corrected control switched on against years of backlog is its own incident. The technical write-up is in Three workflows, enabled and enrolling nobody.

Scorecards, once the data could carry them

Management wanted weekly scorecards, and the sequence agreed was deliberately enablement before enforcement: service levels first, then scorecards tied to KPIs, then self-serve visibility — rather than managers nagging.

Each candidate metric was tested against live data before acceptance, and six of eight failed. One field was populated on exactly one deal portal-wide. New-deal completeness had no room to move, because 165 deals created in thirty days had zero missing region and zero missing sale type. Contact hygiene was 99.6% complete only because three workflows write it automatically, so it measured the automation rather than the person.

Two data problems were surfaced rather than absorbed. Meeting counts run at roughly double reality, because consultants create a meeting in the CRM for the reminder and then send a separate calendar invite. No-shows are deleted rather than outcome-marked, which destroys the record of the attempt. Any meeting-based KPI in the business is overstated until those habits change, and saying so was more useful than building a scorecard on top of them.

What shipped was two components rather than eight, ramped from a 15.5% baseline rather than set at an aspirational bar, cross-checked against a hand-built figure before launch, and filtered on the one field that reliably separates human work from machine-generated work.

The integrations, and the ones that fail silently

Two partner lead sources were brought into the CRM properly during the programme — inbound enquiries through the forms API, and status updates through a custom object with authenticated webhooks.

One of them produced the most instructive failure of the engagement. Submissions were being accepted with a success response and silently discarded, because the payload carried a page reference on a domain the platform did not recognise and treated the whole submission as spam. A first theory about rate limiting was wrong, and was corrected against evidence rather than defended.

That pattern — an integration reporting success while dropping data — is the one worth taking away. It is why every integration on this programme is verified by reading the record at the far end rather than trusting the response at the near one.

What ongoing ownership looks like here

The programme is sequenced across six phases with an exit gate on each: access and quick wins, management visibility, process definition and pipeline redesign, automation and enablement, marketing, then the wholesale channel. Phases overlap deliberately, and process mapping starts early because most of the automation depends on it.

Some decisions were to not build. A forecast submission tool was rejected because it adds a manual task and the consultant-set rating is not trusted internally — deal scoring became the forecasting mechanism instead. Parking stages are being removed on principle so that time-in-stage measures a real stall rather than a filing convention.

Four deal stages and four workflows are locked by an upstream system and require third-party sign-off. Two API capabilities are unavailable on this portal's connection, so certain work is manual by necessity. Those constraints are documented as standing constraints rather than rediscovered each time somebody proposes a change.