Hallinan Stroud

Hallinan Stroud's buyer is a controller or director of accounting at a B2B company of 50 to 400 people. Invoices age in the ledger, deals live in HubSpot or Salesforce, and nothing checks before a reminder goes whether sales is in the middle of something with that customer. Finance keeps a do-not-chase spreadsheet instead.

Two pieces make up Hallinan Stroud: a package installed in the CRM, and a scheduler that sends finance's own reminder templates. Before each send, the scheduler checks the CRM's deals and meetings against a written hold rule. On a match, the account owner picks Send it, Hold until a date, or I will raise it myself; silence never sends.

Hallinan Stroud's hold rule fires only on a ledger customer linked to a CRM company, and machine learning proposes the links rules cannot make. Tax ID, billing domain and exact legal name settle clean pairs, but trading names, separately billed subsidiaries and CRM duplicates leave a few hundred open on a ledger of 1,500 customers. An embedding model retrieves candidate companies and a pair classifier scores each.

Every decision in Hallinan Stroud that someone would regret is made by a person. A pair scoring below 0.92 waits in Match review for the AR specialist, and an unmatched customer over finance's balance floor is held, not sent. Each review adds a labeled row to our corpus holding only similarity features and a yes or no.

Connecting a company to Hallinan Stroud means embedding its whole customer ledger and its whole CRM company list at once. After that, a few dozen new customers arrive a day until month end or a bulk CRM reassignment brings the load back. Our buyers will not hand balances and live deals to someone else's model API, so every model stays on cloud infrastructure in Dublin that we operate.

An onboarding backfill at Hallinan Stroud takes hours today, using general-purpose instances on cloud infrastructure in Dublin. By the end of 2026 both open-weight matching models run on cloud GPU capacity we control in Dublin, started per backfill so Match review opens within an hour of install. The pair classifier is fine-tuned on Match review decisions between January and March 2027 and retrained monthly; a third model, for placing billed subsidiaries, joins them between April and June 2027.

Hallinan Stroud Tecnologia Ltda., a Brazilian limited company registered in São Paulo in November 2025, is its three founders and nobody else. They know one phone call from three sides: the controller who took it, the revenue lead who made it, the engineer who builds CRM packages. As of September 2026 no customer has finished our 30-day dry run, so we have raised no invoice.