Do not rank the people who check the sales
Forced distribution with bottom elimination remains common in Chinese financial companies. Whatever its merits elsewhere, it should never be applied to people whose job is to find and report problems, because it teaches them that accurate reporting is a risk to their careers.
My position is narrow and I hold it firmly. Forced ranking with bottom elimination, the practice of placing a fixed share of staff in the lowest grade each year and moving them out, should never be applied to people whose job is to check other people's work. In an insurer that means return-visit callers, underwriters, claims reviewers, quality inspectors, compliance staff and internal auditors. I am not claiming that forced ranking is always wrong, although I have doubts about it elsewhere. I am claiming that in verification roles it attacks the one thing the role exists to produce, which is an accurate report.
The practice is familiar in Chinese companies, including many financial ones. Its defenders say it prevents grade inflation, forces managers into difficult conversations and keeps large operations efficient. Chinese labour law does not allow a low ranking on its own to justify ending an employment contract, so in practice elimination usually means a lower grade, reassignment, reduced pay or a contract that is not renewed. The pressure is no less real for being indirect.
What the return-visit files showed
In September 2021 our internal audit team sampled 6,000 recordings of return-visit calls from four of Songbai's provincial service centres. A return visit is the call an insurer makes to a new policyholder during the cooling-off period to confirm that they understood the product, received the documents and bought it willingly. It is one of the few moments when someone other than the seller hears the customer's own account of the sale.
In 412 of the sampled calls marked as successfully completed, the recording showed no such conversation. Some lines had not been answered. Some calls had reached a relative who could not speak for the policyholder. In a few cases the person who answered was the agent who had sold the policy. When we re-contacted the 38,000 policyholders those centres had handled in the period, about 2,900 had never had a genuine call. More than a thousand had questions, mostly about participating policies whose dividends had been described as if they were fixed deposit interest.
The callers were not, on the whole, dishonest people. They were ranked each year on completion rate and average handling time, and the bottom 5% of the operations division received the lowest grade. A caller who spent fifteen minutes explaining a dividend illustration to a confused pensioner fell behind one who marked a call complete after a minute. A caller who reported that the agent had answered the telephone created work for a supervisor whose own standing depended on the centre's completion rate. The system had told them what to do, and many of them did it.
A curve that punishes accurate reporting
The defence of forced ranking assumes that output can be compared across people and that more is better. In a verification role neither assumption holds. The valuable output of a return-visit caller, an underwriter or an auditor is often a finding: a customer who did not understand, an application that does not fit the medical evidence, a branch whose expenses do not match its filings. Findings take longer, create disputes and make the people around the finder uncomfortable. A ranking system that rewards speed and penalises friction will, over time, select for people who find less.
There is a second effect that took me longer to see. Forced ranking makes supervisors complicit. A supervisor who must place a fixed share of the team in the lowest grade will protect the members who keep the team's figures up, and the figures easiest to keep up are the ones nobody checks. In two of our four centres, supervisors had known that some calls were not genuine. They had not reported it because doing so would have pushed their teams down the ranking.
After the audit, our chairman and our executive vice president for operations wanted to keep the curve and add monitoring. I understood the instinct. Monitoring was visible and quick, and the curve had served the operations division in other ways. But monitoring a population that is ranked on the very figure being monitored teaches people to falsify more carefully. The president agreed to remove forced ranking and bottom elimination for 4,700 staff in verification and control roles, and to keep it elsewhere while we learnt from the change.
Calibration without a quota
What replaced the curve is ordinary, and it requires discipline. Each verification role is now assessed on accuracy, measured by sampling carried out by a team outside the reporting line; on findings that were later confirmed; and on the quality of the record a reviewer leaves for the next person. Completion rate is still tracked for capacity planning, but it appears in no individual's assessment. Managers calibrate grades across teams in a meeting with HR and the relevant control function present, and no grade has a fixed share.
The obvious risk is drift. In our first calibration round in 2022, 61% of verification staff received one of the two highest grades, and managers stopped trusting the result. We now send every manager the grade distribution for their team each quarter, alongside the distribution for comparable teams, and ask any manager whose team stands far from the others to explain why at calibration. That has brought the distribution back to something people believe, without imposing a quota.
Removing the curve did not mean keeping everyone. Where sampling shows that a caller or reviewer is persistently inaccurate, the manager now sets out the evidence, agrees a period of coaching and, if nothing improves, moves the person to work that suits them better or ends the arrangement properly. That route is slower than a ranking. It is also easier to defend, both to the employee and to a labour arbitration panel.
The results are encouraging, and still young. Independent sampling of genuine policyholder contact rose from 83% in 2021 to 98% by 2024. Complaints about misrepresentation per 10,000 new policies fell by 44%, and cancellations within the cooling-off period rose from 1.1% to 2.4%. I count that rise as a gain: a customer who cancels during the cooling-off period recovers the premium, while one who surrenders in the third year loses much of it. Voluntary attrition in return-visit teams fell from 29% to 13%. The change costs RMB 41 million a year, mainly in additional callers.
I would ask any insurer's board to find out which of its control and verification staff are subject to forced ranking, and what each of them is ranked on. If the answer includes completion rates, speed or the absence of disputes, the board should assume that some of what it hears about the quality of sales has been shaped by that ranking.