Operations2 min read
Cost to collect, computed honestly
Most cost-to-collect numbers are a commission rate with some overhead added. The useful version tells you which rupees you are spending to chase rupees you were never going to get.
Cost to collect sounds like a single number and is really a family of them. Computed at the portfolio level it is almost useless, because it averages the effortless recoveries with the hopeless ones and produces a figure that cannot guide a decision. Computed per cohort it becomes the most actionable number in the operation.
What belongs in the numerator
- People: telecaller and field officer salaries, supervision, quality, training, attrition and rehiring.
- Contact: telephony minutes, SMS and WhatsApp charges, and the ones people forget, failed and unanswered attempts.
- Technology: platform, dialer, recording storage, integrations.
- Third parties: agency commission, Alternate MetaData, legal, field vendors.
- Compliance: certification, audits, and the cost of handling complaints and disputes.
The item most often missing is the cost of unproductive contact. A floor that dials four times to reach once is paying for four dials, and if the number that finally connects was wrong all along, it paid four times for nothing. Alternate MetaData looks expensive until you price the alternative.
Compute it per cohort, not per portfolio
Cost to collect per bucket, per product, per vintage, per ticket size. The pattern that emerges in almost every book is the same: a band of accounts where recovery is cheap, a band where it is expensive but positive, and a tail where the operation reliably spends more than it recovers and continues to do so because nobody has drawn the line. Finding that tail is the entire point of the exercise.
The comparison that matters is marginal, not average
Average cost to collect tells you what the operation costs. Marginal cost, what the next unit of effort on this cohort yields, tells you what to do tomorrow. A cohort with a good average and a terrible marginal return is one you should stop working harder, and an average will never show you that.
Where automation actually changes the number
It does not change the cost of the hard conversations, which still need people. It changes the cost of the easy ones, which is most of the volume, and it changes what your people are doing with the hours it frees. The honest way to evaluate it is not cost per contact, where automation always wins, but recovery per rupee of total operating cost, where it only wins if the freed hours were pointed somewhere useful.