Method · 16 of 18
Every number on this console, defined
Written to be read by somebody who will be asked to defend it to an analyst on a results call. If a figure cannot be explained in a sentence, it should not be on the screen.
Measured
11
Counted from a system of record. Reproducible without this console.
Modelled
8
Derived through a stated, fixed, published rule from measured inputs.
Assumed
1
An input we do not have and have chosen a value for. See the ledger.
Metric definitions
What each figure counts, how it is produced, and where it fails
| Metric | What it counts | How it is produced | Class | Where it fails |
|---|---|---|---|---|
| Portfolio occupancy | Occupied desks divided by operational desks, across all centres including those opened this quarter. | Counted from your inventory system. No modelling. | Measured | Blends mature and ramping centres. Always read next to the mature-only figure. |
| 100 basis points | One per cent of operational desks. At 133,600 desks that is 1,336 desks. | Arithmetic on the desk count. | Measured | Moves as capacity grows. Recompute it every quarter, not annually. |
| Value of 100 bps | Desks in one point multiplied by assumed realisation per desk per month. | Realisation is assumed at ₹17,900 and is not disclosed by the company. | Assumed | The single most sensitive input on this console. See the assumption ledger. |
| Enquiry | One inbound expression of interest, from any channel, de-duplicated on contact and company within 30 days. | Counted at ingestion from your CRM and channel feeds. | Measured | The de-duplication window is a judgement. A company enquiring twice in six weeks counts twice. |
| Reached | A two-way conversation took place. Not a dial, not a ring-out, not a voicemail, not a delivered message. | Counted from call and message logs at the session level. | Measured | A ten second conversation counts the same as a four minute one. Read alongside median duration. |
| Live requirement | A stated desk count, a stated micro-market and a stated move-in window, all three captured in the same conversation. | Extracted from the transcript, then confirmed against the structured field written to the CRM. | Measured | Extraction is not perfect. Sampled and human-audited weekly at the pilot stage. |
| Tour attended | The prospect physically walked a centre. Booked and attended are counted separately throughout. | From your tour log, not from the agent’s claim. | Measured | Depends on centre teams marking attendance. Under-recorded where they do not. |
| Unserved demand | Live requirements that could not be filled from current inventory in the requested micro-market and window. | Requirement matched against live availability at the moment of the conversation. | Measured | A requirement unserved in June may be servable in September. A snapshot, not a backlog. |
| Competitive pressure | Times a competing operator was named in a conversation, weighted by the share of those conversations not won. | Named-entity extraction over transcripts, restricted to a fixed operator list. | Modelled | Only counts operators the prospect mentioned aloud. Silent comparisons are invisible. |
| Voice-verified intent | What the member stated about the next term, on a call, in their own words: expanding, flat, contracting or exit. | Classified from the transcript into four categories, with the quotation retained. | Modelled | Stated intent is not outcome. Not yet back-tested against actual renewals. |
| Renewal gap | Accounts carrying a healthy rating on file whose stated intent on the call was contraction or exit. | Set difference between the two fields. | Modelled | Both inputs are judgements. The gap is a prompt to look, not a verdict. |
| Expansion score | A 0 to 100 composite of headcount signal, stated desk need and same-building capacity. | Weighted sum, weights fixed and published, not learned. | Modelled | Deliberately simple so it can be explained. A learned model would score better and be harder to defend. |
| Reported closure | Share of tickets marked resolved by the facilities team within the window. | From your helpdesk. Untouched. | Measured | Measures the workflow, not the outcome. |
| Verified closure | Share of those same tickets where the member confirmed on a callback that the issue is actually fixed. | Outbound callback on a sample, then extrapolated to the centre. | Modelled | Sample sizes below 30 per centre per month are not reported. Extrapolation carries error. |
| Forward occupancy | Modelled occupancy at 30 and 90 days, decomposed into five components. | Three components measured from contracts and the opening schedule, two modelled. See screen 05. | Modelled | Confidence band is ±140 bps at 90 days and is itself an estimate until back-tested. |
| Entity resolution | The company named on the call matched to a CIN or LLPIN on the statutory register, with the match confidence recorded. | Name and jurisdiction matched against MCA master data during the qualification call. | Measured | Common names and recent name changes produce ambiguous matches. These are flagged rather than guessed. |
| Scale band | A to E, from filed revenue, funded status, paid-up capital and contributing headcount. | Fixed published thresholds, applied in order. Not learned and not tuned per deal. | Modelled | A new subsidiary of a large parent scores on the parent’s scale, which is why the parent is shown next to it. |
| Risk band | Low, Watch, Elevated or High, from the count and type of adverse flags on the statutory record. | Rule based. Two or more independent adverse signals moves a record up a band. | Modelled | Coverage of small private companies is thin. Absence of a flag is not evidence of health. |
| Contact role verification | Whether the person on the call appears on a public filing as a director, designated partner or authorised signatory. | Name matched against MCA and GST records for that entity only. | Measured | Verifies a role, never a person. A negative result is a signing process issue, not a credit one. |
| Growth centre | A centre open 24 months or less. Reported separately from mature centres everywhere on this console. | From the opening date. | Measured | 24 months is a convention, not a law. Choose one and never move it. |
Collection
- Inventory and CRM. Read directly. Read-only until you switch writing on.
- Conversations. Every call and message recorded, transcribed and scored. Retained per your policy, not ours.
- Channel feeds. Site, app, portals, broker submissions, walk-ins, the enquiry line.
- Trigger sources. Public filings and announcements, licensed property data. Never scraped personal data.
Processing
- Extraction. Requirement fields pulled from the transcript, then written as structured data, never as free text.
- Clustering. Topics formed from what members actually said, not matched against a fixed keyword list.
- Scoring. Per conversation, in the language it happened in. Kannada, Tamil and Telugu scored natively, not through English.
- Thresholds. Nothing reported below 30 conversations for a centre in a window. The sample cannot carry the claim.
Known limits of this build
- 16 of 31 Bengaluru centres loaded
- Micro-market map is schematic, not geographic
- Realisation per desk is assumed, not disclosed
- Voice-verified intent is not yet back-tested against outcomes
- Confidence band on the 90 day forecast is itself an estimate
Standing statement
Read-only on day one
Read-only on day one. Nothing is written back to your systems until you switch it on.
The layer proves itself against live data before it changes anything. This is on the console rather than in a contract because a contract does not travel into a meeting, and because a commitment written into the interface is harder to quietly relax than one written into a slide.