How a working-capital-intensive packaging manufacturer partnered with Source Code Lab AI to deploy two agentic systems: one for receivables, one for sales, running on live data inside its own environment in under six weeks.
Potential uplift in qualified pipeline: more ICP-fit brands in the funnel, converting into new-logo revenue as the queue is worked.
Potential reduction in the sales team's prospecting time: discovery, enrichment and scoring are automated, so reps spend their hours selling.
Canpac Trends is a packaging manufacturer serving growing FMCG, D2C, pharma and quick-commerce brands, with operations spread across multiple plants and acquired entities. Like most manufacturers in the category, its business is working-capital-intensive: every day an invoice sits unpaid is cash that can't go back into production.
On the finance side, receivables follow-up was largely manual: matching invoices against POs, dispatch and payment records, tracking aging, and chasing customers one email at a time. Consistency depended on how much time the team had that week.
On the sales side, Canpac's growth plan carried a clear mandate: scale up sales capabilities. But prospecting new brands meant hours of research, enrichment and list-building, and every new name had to be checked against an existing base of 500+ accounts.
The priority wasn't another dashboard. It was to shorten the cash-conversion cycle and grow the pipeline, without adding proportional headcount.
Working from the AI Agents & Automation Blueprint it had already delivered for Canpac, Source Code Lab AI selected two high-priority, fast-payback agents, one quick win in Finance and one major project in Sales, and took both from design to live deployment in a fixed-scope, 4–6 week engagement.
Ingests invoices from Canpac's source system, runs a rules-based 3-way match against PO, dispatch and payment records, and segments every open invoice into aging tiers with a risk flag. A tiered reminder engine then sends templated, escalating reminders on a configurable cadence, alerts the collections owner on high-risk accounts, and writes status back to the ledger.
Continuously discovers growing FMCG, D2C, pharma and quick-commerce brands likely to need packaging, enriches decision-maker contacts, and scores each account against an Ideal Customer Profile codified with Canpac's sales team. The result is a prioritised, de-duplicated outbound queue in the CRM, each account with a fit score and the reasoning behind it.
This wasn't automation bolted onto spreadsheets. Both agents were built human-in-the-loop by design: reminder templates, escalation thresholds and any customer-facing send are approved by the finance team, and the sales team owns every outreach decision, the agent prepares and prioritises, people send.
Both agents were deployed in parallel and run on live data in Canpac's environment, measured against a baseline of DSO and prospecting effort captured at kick-off. The targets below come from the blueprint.
More ICP-fit brands reach the sales team every week, and revenue grows as the queue is worked, at a ~40% lower cost per lead than manual research or bought lists.
Discovery, enrichment and ICP scoring run automatically, giving the sales team back the hours it spent on research to spend on conversations with buyers.
Every open invoice now gets a timely, consistent follow-up, freeing working capital from faster receivables recovery.
Matching, aging and reminders run automatically, the finance team steps in only where escalation calls for it.
See which AI agents would pay back fastest in your business. Most engagements go from blueprint to live agents in 4–6 weeks.
Invoices from different plants and entities are often checked by hand against purchase orders, dispatch and payment records. Reminders then depend on someone remembering to send them, so overdue accounts slip through. For a business that runs on working capital, each extra unpaid day is cash it can’t use.
DSO measures the average number of days it takes to collect payment after a sale. The collections agent checks each invoice, sorts overdue accounts by age and risk, and sends reminders on a schedule finance controls. Canpac’s target is 8 to 15 days shorter. This is a target, not a guarantee, and it depends on data quality and how the team uses the agent.
The agent never moves money or processes payments. Every customer-facing message uses templates finance has approved, and risky accounts are escalated to a person, who decides what happens next.
The agent finds decision-makers from public and licensed sources and scores each brand against an ideal-customer profile your sales team defines. It also checks every account against existing CRM records, so a current customer doesn’t appear as a cold lead. Sales decides who to contact and sends every message.
Without a starting point, any improvement is just an impression. In week one we recorded Canpac’s current DSO and how prospecting worked, and both owners signed off on what “working” meant. The sales team’s reported research-time drop of around 40% is measured against that baseline.