A recommendation that fails any of the four properties is an observation, and observations don't get funded.
What are the four properties of a recommendation?
Specific means the reader can tell whether it's been done. "Improve unit economics in the new stores" cannot be checked. "Raise average daily orders in the 72 FY26 stores from 310 to 357 by 31 March" can be checked on a Tuesday by one query.
Owned means a named role carries it, and the role has the authority to do it. Hand a density programme to a finance team who can't change delivery routes and you get a plan that reports but doesn't act.
Dated means there is a date on which someone will ask whether it happened. A recommendation without a date is a preference.
Costed means the reader knows what they're approving in money, and knows the return that makes the money worth spending. Analysts stop short of it more often than of the other three, because the cost falls outside the analysis they were asked to do. That's unlucky, because it's also the property that most often decides whether a recommendation is adopted.
Advertisement
Worked example: a weak and a strong recommendation
Here's the Turant recommendation written badly, then written properly, on identical evidence.
Start with the weak version. "We recommend focusing on improving the performance of newer stores and taking a more disciplined approach to the expansion pipeline while we work through the current cohort's ramp." It has no number, no owner, no date and no cost. It also hedges the one decision the board came to take, which is whether to sign off forty more stores.
The strong version, as it appears on the recommendation slide:
Pause the 40 planned openings for two quarters, which defers ₹13.60 crore of capital spending, and spend ₹5.70 crore on a density programme across the 72 FY26 stores. Target 357 orders a day per store by 31 March 2027 and 420 by 30 September 2027. Owner: the Chief Operating Officer, reporting monthly against store-level density. At 420 orders a day the 72 stores add ₹13.11 crore a year of contribution, so the programme repays in 5.22 months, and 32 of the 72 stores reaching the target repays it inside a year on their own. Stores below 357 at month twelve close.
Every figure in that paragraph came from the spine. Programme cost is ₹3.02 crore of one-time store work plus ₹2.68 crore of six months of incremental local marketing. At 420 orders a store earns ₹2,850.36 a day against today's −₹2,139.02, so the gain is ₹4,989.38 a store a day and ₹13.11 crore a year across 72 stores. Payback is ₹5.70 crore over ₹13.11 crore a year.
What makes a strong recommendation credible?
Two features of the strong version deserve attention, because they make it credible. Detail alone wouldn't.
It states a conservative assumption out loud. Its gain calculation holds contribution per order at today's ₹45.36, even though delivery cost per order falls as density rises. So the ₹13.11 crore is a floor, and you shouldn't present it as a forecast. Saying so costs one sentence and buys the reader's trust in the other figures.
And it carries its own stopping rule. A sceptical director will test the closure trigger at month twelve harder than anything else, and it's also what makes the ask reasonable, because it bounds the downside. A plan with no stopping rule asks for an open-ended commitment, and boards that have been asked for one before will decline this one.
Why does the costing include the option not taken?
The costing also has to include the option not taken. Forty stores at the current cohort's run rate add ₹3.12 crore a year of losses and 12,400 orders a day, which is 10.2% of current volume. That trade is the actual decision: 10.2% of volume against ₹3.12 crore a year and the ₹13.60 crore of capital. Put both sides of it on the slide and the recommendation becomes a choice the board can make. Leave one off and it's a request they can only approve or refuse.