How to run a two-week practicum for collections agents

Spend ten working days on the AI agent's real handovers and holds: payment plans outside its bounds, settlements, disputes and customers who showed signs of vulnerability. A certified team leader reviews every decision and listens to a set of calls. The collector passes with at least 95 decisions in every 100 right and no missed red line, such as chasing a customer who disclosed serious hardship.

What is the practicum, and who owns it?

A practicum is supervised practice on real work, judged against a fixed bar. In collections, the person on the phone has two kinds of decision to make: what to do with the AI agent's held work, and how to handle the customer the agent handed over. The practicum tests both. The specific role it plays in agent training is covered in what a practicum in AI agent training is.

The collections team leader owns the practicum and supervises, or names a certified supervisor. The head of collections sets the lender's own limits that define what the trainee may agree. A conduct or quality reviewer sets the red lines for vulnerability. The collector owns their decisions and call notes.

How do you build the queue?

Collections decisions mix quick approvals with long calls. Around 15 decisions a day, plus four or five reviewed calls, gives a fortnight of roughly 150 decisions. Include:

• payment plans the AI agent could not agree because they fall outside its bounds;
• settlement offers and write-off proposals, held at depth 4;
• accounts where the customer disputes the debt;
• handovers triggered by vulnerability signals: illness, bereavement, job loss, confusion;
• contacts the agent refused because the contact limit was reached;
• non-standard letters the agent drafted for review.

Seed anonymised past cases for any type missing from the live queue. For example, a seeded handover might show a customer who told the agent "I've just come out of hospital and can't deal with this". The correct decision is to pause collection activity, offer support routes and record the need, not to agree a smaller plan on the spot.

The steps

1. Person (team leader): brief the collector on the bar, the red lines and the log.
2. Agent, depth 1: route held items and handovers to the practicum queue, tagged by type.
3. Agent, depth 2: present each account with its draft, the deciding bound and the facts, showing which came from the ledger and which the customer declared.
4. Person (collector): decide or call, with a reason and call notes.
5. Person (team leader): decide independently and listen to selected calls.
6. Person (team leader): authorise any settlement, write-off or plan outside standard bounds before it is confirmed to the customer.
7. Agent, depth 1: log both decisions, the match and any red-line flag.
8. Person (team leader, then assessor): review the log at the end.

How do you set and apply the bar?

A decision is correct if it matches the team leader's or the leader agrees it was sound. For calls, correctness covers what was agreed, what was recorded and how the customer was treated.

The red lines: continuing collection activity after a customer disclosed serious vulnerability without offering support; contacting a customer outside the lender's contact rules; agreeing a settlement or write-off without authority; pressing a disputed debt before the dispute is reviewed. One miss fails the practicum. In the UK, the FCA's Consumer Duty expects good outcomes for retail customers, and these red lines reflect that.

Plans count as correct only if they are affordable on the evidence. A plan the customer agrees to on the call, at an instalment the income and spending figures cannot support, is logged as incorrect even if the customer was happy. The team leader checks the figures, not the customer's tone.

Which controls matter, and what evidence do they leave?

• Leader authorisation of non-standard outcomes. Evidence: the leader's name on each in the append-only record.
• Call review. Evidence: call scores and notes.
• Independent leader decision. Evidence: logged before comparison.
• Daily debrief. Evidence: a line per mismatch.
• Full log. Evidence: the practicum log template for collections agents.

What should you measure?

Correct decisions out of the total, by type. Red-line misses (zero). Plans agreed during the practicum that broke within a month, measured afterwards. And call quality scores on vulnerability handovers, which matter more than speed.

Where does it usually go wrong?

The fortnight has no vulnerability cases. The collector passes without having handled one. Seed them.

The team leader reviews calls only when time allows. Fix the call sample in advance.

Collectors learn to accept the AI agent's summary and skip their own questions. A summary is a starting point; customers in difficulty often say the important thing on the second question.

The practicum runs at month start, when payment volumes peak and supervision thins. Move it. IT and support teams run the same format; compare running a practicum for customer support agents.

Where the Academy fits

Pryme Intelligence Academy's Certified Pryme Operator ends with a practicum on these terms: two weeks of real review decisions, at least 95 in every 100 correct and no missed red line, signed off by a Pryme assessor. The internal log becomes the evidence the assessor reviews. More in the guide to AI literacy and certification for people who work with agents.