Telecom company
Sales Opens the Account and the Argument Is Already There
The moment a new campaign list lands in the CRM, the system enriches every prospect from public registries and writes a ready-to-use sales argument back into the record. The team opens the account and the opening line is already there.
Days of prospect research, now filled in before the first call.
What changed
- ~1,700+ companies / run: enriched automatically per campaign list (estimate)
- hours saved: days of manual prospect research, now automatic (estimate)
- ↓ skipped research: every prospect gets an argument, not only the ones with time (estimate)
01The challenge
Where it started.
For every prospect in an outbound campaign, the B2B sales team needed company and financial background plus a tailored opening argument to start the conversation. Done by hand for hundreds to thousands of companies per campaign, this was slow and uneven.
When the team was busy, the research was often skipped entirely and the call went out cold, or the argument was thin and generic because there was no time to dig into the registries.
The quality of the opening depended on who had time, not on the prospect, so the same campaign list produced very different first impressions.
Quote
“The AI writes only the argument. Finding the facts and counting what is left are jobs for code, and code does them the same way every run.”
02What we built
Agents and workflows, on one platform.
The goal was not to replace the seller. It was to have every prospect arrive with its homework already done, the same way, every time.
Inputs
AI agents
State
Output
Why the Elevon platform
Detected by the shape of the data
A change-detector notices a new campaign not by a timestamp, but by the shape of the data: a batch of rows whose sales-argument field is still empty. Nothing needs to be scheduled or announced.
A batch loop that respects the limits
A self-triggering loop processes the list in windows with a deliberate delay between them to stay within rate limits. It counts how many companies remain and re-triggers itself until the list is done, with a clear stop condition.
Restartable and idempotent
State is held in a spreadsheet on shared storage, so the loop can be interrupted and resumed. Each company is completed exactly once, even across restarts, so nothing is done twice and nothing is dropped.
03Rollout
Step by step.
When a new list appears, the loop starts, takes a window of companies, and for each one, identified by its business registration number, a worker enriches it from public company and financial registries plus a light web look. Then it writes a concise, ready-to-use sales-argument card of up to about 2,000 characters back into the CRM.
Between windows the loop pauses, counts the companies still without an argument, and re-triggers itself. When the count reaches zero it stops. A run covers about 1,746 companies, and the sellers see the arguments filled in as the batches complete.
In practice
Enrichment and counting happen in code; the AI writes only the argument text. Branching is cheap and happens before the expensive steps: has this company already been done, are we still inside the window, so the system asks the free questions first and only then spends on lookup and writing.
04Results
What changed.
The team stops researching prospects by hand and starts opening records that already carry a tailored argument.
- ~1,700+ companies / runenriched automatically per campaign list (estimate)
- hours saveddays of manual prospect research, now automatic (estimate)
- ↓ skipped researchevery prospect gets an argument, not only the ones with time (estimate)
How we estimate: the per-run count reflects one observed campaign list; time saved is the manual research effort the loop removes. The figures are illustrative and should be replaced with the client’s real numbers.
- Every prospect in the list gets a tailored sales argument, not just the ones there was time for
- Company and financial facts pulled from public registries for each business registration number
- About 1,746 companies enriched per run, in windows that respect rate limits
- Each company completed exactly once, even if the run is interrupted and resumed
- A consistent opening across the whole campaign instead of quality that depends on who had time
Why it worked: The work splits cleanly along its natural line: code handles the facts and the bookkeeping, the AI handles the wording. That keeps the numbers trustworthy and the arguments readable, and it makes the whole loop safe to restart.
Detecting work by the shape of the data, not a schedule, means the team just drops a list into the CRM and the system takes it from there, with no new button and no new habit to learn.
Quote
“Prospect research used to wait until someone had a free afternoon. Now the argument is in the record before the call is even planned.”
Next step
Your process could be the next case.
A 30-minute call. You tell us about one process, and we say whether it fits the platform, a product build or a consulting project.