Ask the article who resigned in October what September looked like: forty calls a day asking for ledgers, maintaining a tracker Excel at midnight, being shouted at by clients they were chasing on a partner's behalf. They signed up to learn audit. They spent the season as a call centre. Multiply that by every article and junior who has left in the last three years, and you have a retention problem most firms keep trying to fix with stipend hikes and pizza on the 30th, when the actual lever was sitting in the workflow the entire time.
The bargain articles actually accept
Articleship's deal was never about the stipend. It's exposure in exchange for effort. Long hours on a stock count, a tricky 3CD clause, a real assessment: that's the deal working. Long hours re-asking Kalpana Textiles for their fixed-asset register for the fifth time: that's the deal broken. People leave when the effort stops buying learning.
This is easy to miss from the partner's chair, because the hours look identical from the outside. Twelve-hour days in September, twelve-hour days the following September. What changed is the ratio inside those hours: how much of the day is spent thinking like an accountant versus how much is spent behaving like a collections agent for people who don't report to you and have no reason to reply to you.
A week in September, two years apart
Take an article we'll call Rohan, in his second year, at a firm we'll call Advani & Co. In his first September, a Tuesday looked like this: in by 9, straight into the WhatsApp backlog from twelve clients who hadn't sent GSTR-1 source documents. Forty minutes of calls before 11, mostly voicemail and one client who wanted to argue about why the firm needed his purchase register again when he'd "sent it last month." An hour updating the tracker Excel, who's pending, who promised what, who to escalate to the partner. Lunch at his desk, because the afternoon was three more rounds of the same calls for a different set of clients. By 7pm he had touched exactly one actual working paper, for eleven minutes, before a partner asked him to chase two more clients before EOD.
The following September, same firm, same GSTR-1 deadline, same twelve clients on his list. Rohan's Tuesday: in by 9, the tracker is already current because the reminders went out automatically over the weekend and the responses were sorted against the checklist overnight. Four clients still need a nudge; those go out as templated follow-ups without him touching a keyboard. His morning is a partner walking him through a client's input tax credit mismatch. His afternoon is drafting the first pass of a reconciliation, with actual feedback on what he got wrong. He leaves at 7 having learned something he'll use again. Nothing about Rohan changed between those two Septembers. The job did.
What an exit actually costs the firm
Firms tend to price attrition as "we'll hire someone else," as if the cost stops at the recruiter's fee. It doesn't. Run the arithmetic on a single exit and the number gets uncomfortable fast, not because any one line is huge, but because they all land in the same six-month window and none of them show up on a P&L as "attrition cost."
| Cost | Illustrative estimate | Who absorbs it |
|---|---|---|
| Recruiting and interviewing a replacement | 3 to 6 weeks of partner and HR time | Partners |
| Time to productive (basic tasks, no supervision) | 3 to 6 months | The bench |
| Institutional memory: which client sends what, where, in what format | Rebuilt from scratch, imperfectly | Every remaining senior |
| Partner review hours redirected because the bench got junior overnight | 2 to 4 extra review hours per week, for months | Partners |
| Reputation loop: the next batch hears why this article left | Fewer strong applicants next cycle | The firm, long-term |
None of those lines is dramatic on its own. Stacked together, they mean a firm that loses two articles a year after season is running a permanent training deficit, always a few months behind where it would be if the last cohort had simply stayed. That deficit is invisible in the accounts and very visible in how long every review takes.
What exit interviews actually say (and don't)
Exit interviews rarely surface the real reason cleanly, because "the work was mostly repetitive admin" sounds like a complaint about effort, and articles have spent three years being told effort is the whole point. So the stated reasons cluster around the stipend, or "better opportunity," or "family reasons," all true in a narrow sense, all missing the mechanism. Below is a composite of the pattern that shows up across dozens of these conversations, paraphrased and combined, not a real transcript from a real person:
I didn't mind the hours. I minded that most of them were calling the same twelve people, over and over, about the same missing document. I learned more about audit in one month once the season ended than I did in the two months during it, because that's when I finally sat with a senior instead of a phone.
That line, "I learned more once the season ended," is the tell. The busiest weeks of the year, the ones with the most partner attention and the most real technical problems on the table, are also the weeks the least learning happens, because the article's day has been rerouted into logistics. The season that should be the best training ground becomes the reason they leave.
The fix is operational, not motivational
Pizza on the 30th doesn't retain anyone. Removing the chase does. When intake, tracking, and follow-up run on software, the same season hours go into reconciliations, audit judgment, and client exposure, the work articles came for. Firms that made that switch report the same deadlines with a different October: no resignation letters, and a bench that walks into the next season already ahead instead of already burnt.
Will removing the chase reduce the learning value of articleship?
No, it reverses what's currently happening. Chasing documents was never the learning; it's the prerequisite admin that has to happen before the learning can start. An article can chase a client for a fixed-asset register for two weeks and learn nothing about fixed assets in the process. The learning is in reconciling the register once it arrives, questioning an odd entry, seeing how a partner treats a borderline capitalisation call. Automating the chase doesn't remove a training exercise; it removes the wait before the training exercise.
What do we do with existing juniors' time once the chasing workload disappears?
Point it at the backlog every firm has and no one has time for: cleaning up prior-year working papers, building out checklists for compliance events the firm handles informally, giving articles a first pass at reconciliations with a senior reviewing rather than drafting from zero. Firms that free up ten to fifteen hours a week per junior during season almost never run out of higher-value work to hand them, they've just never had the hours available to find out.
How do we retain institutional knowledge about clients if we're not relying on the article who 'just knows' who sends what?
That reliance is the risk, not the safeguard. "Ask Priya, she knows how Kalpana Textiles sends things" is a single point of failure that walks out the door every time Priya does, which is exactly the problem this article is about. A checklist and an intake history attached to the client, not to a person's memory, is what survives an exit. The irony is that firms cite "institutional knowledge" as a reason to keep the manual process, when the manual process is precisely what makes that knowledge disappear every time someone resigns.
The metric worth tracking
If you want a number that predicts next October's resignations before they happen, don't track billable hours during season, track how many of an article's hours went into chasing versus reviewing. Ask at the mid-season check-in, not the exit interview. A ratio that's flipped toward chasing is a resignation with a two-month delay on it. Fix the ratio and you've fixed the retention problem without spending a rupee on stipends.
Written by Team DocBox, Founding team, DocBox. General guidance on practice operations, not professional or legal advice for a specific matter.