Field Collections • Chief of Risk business case

Scale field collections selectively, with a gated hybrid model and explicit stop-losses.

This decision memo synthesizes the attached pilots, planning pages, economics workbook, and operating insights into one risk-aware recommendation. The narrative is designed for an executive discussion: first the decision, then the proof, then the economics, then the controls required before broader scale.

Decision in one sentence

Approve a staged hybrid rollout: internal in dense 31-60 pockets, Kobra in proven or experimental 91-120 pockets, and do not approve broad statewide or national scale until productivity, incrementality, and data reconciliation gates are cleared.

3 / 3 Pilots showed positive cure uplift in the treatment arm.

MXN 300k Monthly payroll for one internal 10+1 field team.

10/day Internal productivity level that roughly breakevens CDMX on the conservative case.

6.8% Cumulative contact rate after revisits versus 3.1% on first visit only.

1. Executive recommendation

Recommendation: staged hybrid, with internal bias only where density is proven.

The evidence does not support a simple binary answer. In-house is not yet proven as a broad operating model; Kobra is not attractive as a broad answer at current or expected pricing. The strongest path is a hybrid that uses each model where it is structurally advantaged and keeps the first approval narrow.

Go now: CDMX 31-60 internal proof Go now: Chihuahua 91-120 with Kobra Test now: Jalisco 91-120 Kobra core Next internal replication: Jalisco 31-60 Prepare next wave: selective Estado de México 31-60 Do not approve: broad CDMX 91-120 Do not approve: broad 91-120 Kobra scale

Key message for the Chief of Risk: field collections appears to be a viable incremental channel, but only selectively. The right approval is a gated rollout with clear governance, starting with a narrow internal CDMX proof and a tightly scoped Kobra Pilot 4.

Why this recommendation holds up under scrutiny

Positive pilot evidence exists, but it is uneven by geography and bucket.
Broad Kobra expansion destroyed economics in several states despite real gross recovery.
Internal economics look attractive only if productivity stays near 10+ visits per day and all-in costs do not rise materially above payroll-only assumptions.
Jalisco is the cleanest repeat-positive geography across the attached materials.
Chihuahua is the cleanest vendor-backed 91-120 case.
CDMX has enough density to justify an internal proof in 31-60, but the evidence does not support a broad 91-120 launch.

2. Pilot results and proof of concept

The pilots proved that field visits can work. They did not prove that scale should be broad.

For the CRO version, the evidence is shown in two layers. First comes the gross field view: cash recovered by field visits and cash spent on field visits. Then comes the net decision view: treatment-group recovery minus the remote-only recovery baseline minus field spend. The two recovery fields are not the same in the attached source files, so they are shown separately.

Pilot / bucket	Field recovery	Field spend	Gross field result	What this gross view means
Pilot 1 91-120	MXN 357,609	MXN 220,000	MXN 137,609	Operational workbook view only: field cash recovered minus field spend.
Pilot 2 91-120	MXN 541,144	MXN 400,000	MXN 141,144	Gross recovery still exists, but this does not mean the rollout is attractive after the remote baseline is included.
Pilot 3 31-60	MXN 392,900	MXN 267,604	MXN 125,296	Earlier bucket shows the strongest gross signal, but still needs net validation.

Pilot / bucket	Recovery from field-treatment group	Recovery from calls & notifications baseline	Field spend	Final delta = field-treatment recovery - baseline recovery - field spend	Decision read
Pilot 1 91-120	MXN 331,964	MXN 115,559	MXN 220,000	MXN -3,595	Channel proved viable, but economics were only roughly breakeven at pilot level.
Pilot 2 91-120	MXN 558,693	MXN 300,495	MXN 400,000	MXN -141,802	Late-stage broad Kobra expansion is not supported.
Pilot 3 31-60	MXN 1,037,916	MXN 731,912	MXN 267,604	MXN 38,400	Earlier bucket looks more promising, but still needs selective geography.

State / bucket / pilot	Field recovery	Field spend	Gross field result	Recovery from field-treatment group	Recovery from calls & notifications baseline	Final delta	Read
Jalisco 91-120 • Pilot 1	MXN 109,730	MXN 42,822	MXN 66,908	MXN 110,630	MXN 22,441	MXN 45,367	Strong positive late-stage proof.
Chihuahua 91-120 • Pilot 2	MXN 70,398	MXN 37,157	MXN 33,240	MXN 100,715	MXN 21,265	MXN 42,293	Cleanest current Kobra case.
CDMX 91-120 • Pilot 2	MXN 192,444	MXN 170,518	MXN 21,926	MXN 179,244	MXN 103,955	MXN -95,229	Gross-positive but net-negative; not a wave-1 answer.
Jalisco 31-60 • Pilot 3	MXN 115,159	MXN 68,987	MXN 46,172	MXN 265,676	MXN 170,937	MXN 25,752	Best internal replication candidate.
Estado de México 31-60 • Pilot 3	MXN 110,145	MXN 88,357	MXN 21,788	MXN 337,160	MXN 223,619	MXN 25,184	Attractive, but after internal operating proof.
CDMX 31-60 • Pilot 3	MXN 167,596	MXN 110,260	MXN 57,336	MXN 435,080	MXN 333,114	MXN -8,294	Enough signal for a narrow proof, not enough for citywide approval.

Strong enough to support scaling decisions

Jalisco is the strongest repeat-positive geography. Chihuahua is the strongest clean Kobra case. Broad 91-120 rollout through Kobra is not supported. Micro-geography matters enough that borough/city packs are better decision units than full states.

Strong enough to support operating changes, but not cash claims

Revisits materially improve borrower contact and PTP formation. The first-visit direct contact rate in the attached image is 3.11%; second visit rises to 10.10%; third to 19.77%. This supports revisit discipline, but it does not by itself prove proportional incremental cash.

Tentative findings that should be framed as directional

Late-afternoon visits appear better than earlier-day visits; raw contact improved after March 25; agent performance varied materially. These are useful for rollout design, but they are not decision anchors on their own.

3. Economics and operating model

Gross economics tell us where to look. Final delta tells us whether to scale.

The current materials contain two different economic views. The pilot readout lets us calculate final delta because it includes treatment-group recovery, remote-only recovery baseline, and field spend. The launch-sizing file usually gives only projected field recovery and projected spend, so it supports gross prioritization but not a final net decision on its own.

Model	Field spend logic	Where it fits	What can break it
In-house 10 collectors + 1 supervisor	MXN 300k monthly team cost. Cost per visit = 300,000 / (10 × 22 × visits/day).	Dense, repeatable 31-60 pools where routing can stay tight.	Productivity below 10/day, thin inflow, or omitted all-in costs such as transport, devices, recruiting, insurance, and shrinkage.
Kobra	Expected decision price in the materials: MXN 185/visit.	Proven 91-120 pockets and still-unanswered experiments.	Vendor price reset, low-balance tail, diluted geography, or poor address quality.
Hybrid	Internal fixed-cost cell for dense pools + variable vendor cost for selected late-stage pools.	Best overall fit to the evidence.	Governance complexity if the program is not managed on one economic dashboard.

CDMX internal sensitivity, shown with all parts of the delta

This is the clearest monthly internal proof in the materials because it explicitly shows all three parts: projected field recovery, remote-only baseline recovery, and field spend.

Visits / day	Cost / visit	Projected field recovery	Projected calls & notifications baseline recovery	Field spend	Final delta
8	MXN 170.45	MXN 398,606	MXN 158,552	MXN 300,000	MXN -59,946
10	MXN 136.36	MXN 498,257	MXN 198,189	MXN 300,000	MXN 68
12	MXN 113.64	MXN 597,909	MXN 237,827	MXN 300,000	MXN 60,081

Risk note: the internal case shown in the attached HTML intentionally excludes several cost lines. That is acceptable for an early directional model, but it is not yet a final investment-grade cost view.

Internal team cost breakdown

The internal base case uses one field cell with 10 field agents + 1 supervisor. The current model reflects payroll and labor burden only.

Cost component	Formula	Monthly amount	What it means
One field agent	(MXN 10,000 fixed + MXN 12,000 variable) × 1.20 burden	MXN 26,400	Per-agent monthly cost including the labor burden used in the current model.
Ten field agents	10 × MXN 26,400	MXN 264,000	Total monthly cost for the collector team.
One supervisor	MXN 30,000 compensation × 1.20 burden	MXN 36,000	Monthly supervisor cost in the base case.
Total internal 10+1 team	MXN 264,000 + MXN 36,000	MXN 300,000	Base monthly payroll cost used in the internal economics.

Launch-pack sizing view: gross first, because the remote baseline is not attached everywhere

The planning file gives projected field recovery and projected field spend by municipality. That is enough to show gross field result = projected field recovery - projected field spend. It is not enough on its own to show final delta, because the calls-and-notifications baseline is not attached at municipality level in that file.

Launch pack	Model	Projected field recovery	Projected field spend	Gross field result	Calls & notifications baseline recovery	Final delta status
CDMX wave 1 Benito Juárez + Coyoacán + Cuauhtémoc + Venustiano Carranza	Internal	MXN 1,164,930	MXN 246,273	MXN 918,658	Not attached in pack file	Needs matched baseline before final scale approval
Jalisco 31-60 launchable internal pack	Internal	MXN 1,457,471	MXN 799,061	MXN 658,409	Not attached in pack file	Gross-attractive; final delta still needs baseline join
Estado de México 31-60 launchable internal pack	Internal	MXN 1,255,546	MXN 921,766	MXN 333,780	Not attached in pack file	Wave 2 candidate after proof
Chihuahua 91-120 covered Kobra pack	Kobra	MXN 542,266	MXN 292,672	MXN 249,594	Not attached in pack file	Supported by positive pilot delta already
Jalisco 91-120 Pilot 4 core Guadalajara + Zapopan	Kobra	MXN 267,409	MXN 188,885	MXN 78,524	Not attached in pack file	Experiment only; first-cut geography is narrow

4. Geographic / state-level recommendation

Prioritize geographies where three things overlap: pilot proof, density, and manageable field execution.

The map below is intentionally conservative. It ranks places not only by modeled economic upside, but by whether the supporting evidence is pilot-backed or merely inferred from current opportunity sizing.

Wave 1: scale now

Jalisco 31-60 internal, led by Guadalajara and Zapopan.
Chihuahua 91-120 with Kobra, especially Juárez and Chihuahua.

Wave 1: prove now

CDMX 31-60 internal in Benito Juárez, Coyoacán, Cuauhtémoc, and Venustiano Carranza.
Use high-balance prioritization from day one.

Wave 2: conditional

Estado de México 31-60 internal in Ecatepec, Nezahualcóyotl, Naucalpan, Tlalnepantla, and adjacent dense metros.
Only after productivity and conduct controls are stable.

Test only / stop

Jalisco 91-120 Kobra: Guadalajara + Zapopan core, San Pedro Tlaquepaque optional.
Do not scale: broad CDMX 91-120, Nuevo León 91-120, Puebla 91-120, Querétaro 91-120.

CDMX 31-60 launch pack: prioritized areas

This is the recommended first internal pack because it combines good field-fit evidence with attractive gross pack economics.

Municipality	Projected field recovery	Projected field spend	Gross field result	Localized	Direct contact	Action
Benito Juárez	MXN 186,981	MXN 112,581	MXN 74,400	97.9%	8.3%	Start
Coyoacán	MXN 266,975	MXN 160,745	MXN 106,230	89.9%	5.1%	Start
Cuauhtémoc	MXN 284,643	MXN 171,383	MXN 113,260	89.2%	6.5%	Start
Venustiano Carranza	MXN 213,973	MXN 128,833	MXN 85,140	93.2%	5.4%	Start
Gustavo A. Madero	MXN 425,492	MXN 256,188	MXN 169,304	88.6%	4.9%	Reserve / add after proof
Iztacalco	MXN 189,435	MXN 114,058	MXN 75,376	96.1%	3.9%	Reserve / add after proof

Why CDMX 91-120 is not in wave 1

The attached evidence is consistently weak for CDMX 91-120. The state-level pilot result is negative net, and every CDMX 91-120 municipality in the current planning file is already gross-negative before subtracting the calls-and-notifications baseline.

Municipality	Projected field recovery	Projected field spend	Gross field result	Localized	Direct contact	Action
Miguel Hidalgo	MXN 57,691	MXN 68,565	MXN -10,873	88.2%	1.8%	Do not start
Coyoacán	MXN 81,660	MXN 97,051	MXN -15,391	92.8%	2.9%	Do not start
Gustavo A. Madero	MXN 154,959	MXN 184,164	MXN -29,206	83.9%	7.5%	Do not start
Iztapalapa	MXN 186,173	MXN 221,262	MXN -35,089	76.0%	7.5%	Do not start

Simple answer for the CRO

We are not recommending CDMX 91-120 because the bank already has direct CDMX 91-120 evidence, and that evidence is negative after the remote baseline is included: MXN 179,244 - MXN 103,955 - MXN 170,518 = MXN -95,229.

Jalisco 91-120 core

Guadalajara and Zapopan remain the cleanest Kobra core. San Pedro Tlaquepaque is close to breakeven. Tonalá should not be in the first cut.

Targeting inside geographies

The updated planning file supports prioritizing debt 15k+ and credit limit 10k+, with especially strong economics in the higher-balance sub-slices.

5. Improvement levers before or during scaling

The best levers are the ones that improve route quality, target quality, and true incrementality.

Not every idea should be a prerequisite. The right sequencing is to enforce the few levers that most directly change wasted visits, cannibalization risk, and conduct quality before headcount expands.

Lever	Likely value	Feasibility	Should it be a prerequisite?	Recommended stance
Address enrichment bureau + living address confidence	High	Medium	Yes	Highest-value prerequisite. It directly reduces wasted visits and should be built into dispatch gating.
Address prioritization algorithm	High	High	Yes	Implement early. Use the same mindset as third-party-number prioritization.
Image sharing already cleared by infosec / compliance	Medium	High	Operationally yes	Include in rollout. Strong governance and QA benefit, even if cash impact is indirect.
Rapid revisits	High	High	Yes	Operational evidence is strong. Put a 48-72 hour revisit SLA into the launch design.
Visiting-hour adjustment	Medium	High	No	Quick win. Shift harder accounts to late afternoon and evening.
Geo-location check + silent pushes	Medium	Medium / low	Not yet	Interesting, but needs tighter legal and conduct design before becoming a proceed/stop gate.
Credit_limit and total_limit analysis	High	High	Yes	Part of rollout. The current materials strongly support high-balance prioritization.
Recent contact status in last 7 days	Potentially high	Medium	Needs testing	Important missing analysis for incrementality. Use as a suppression test before broad scale.

6. Next steps and expansion path

Immediate next steps, and what happens if the CDMX internal launch works

The rollout should be framed as a sequence, not as a one-time national decision. The safest pattern is: prove the internal operating model in a narrow CDMX pack, run Kobra Pilot 4 in a narrow Jalisco core, deploy the highest-value improvements, then expand only if the gates are hit.

Step 1

Launch CDMX 31-60 internally in Benito Juárez, Coyoacán, Cuauhtémoc, and Venustiano Carranza. Prioritize debt 15k+ and credit limit 10k+ first.

Step 2

Run Kobra Pilot 4 in Guadalajara + Zapopan. Add San Pedro Tlaquepaque only if week-1 volume is insufficient. Keep Tonalá out of the first cut.

Step 3

Put improvements live from day 0: address enrichment, address prioritization, approved image sharing, revisit SLA, visiting-hour adjustment, and recent-contact suppression test.

Step 4

Set a hard decision gate after the first operating cycle and expand only if the internal model is stable on productivity, field fit, and economics.

If CDMX internal launch works, then expand in this order	Why this is the next move	What has to be true first
1. Add reserve CDMX boroughs Gustavo A. Madero, then Iztacalco, then later-wave areas only selectively	Lowest execution risk. Same city, same team, same routing logic, same oversight model.	Internal team sustains roughly 10+ visits/day, address confidence remains stable, and the matched cohort economics are not negative.
2. Extend into adjacent Estado de México dense metros Ecatepec, Nezahualcóyotl, Naucalpan, Tlalnepantla, then adjacent clusters	Best way to deepen utilization using contiguous urban density rather than opening a totally new operating cell immediately.	CDMX proof is stable and the operating dashboard can handle a wider footprint.
3. Open the next full internal regional cell in Jalisco 31-60	Jalisco is still the strongest replication candidate once the internal model itself is proven.	The bank is comfortable funding a second internal cell and the first cell is operationally stable.

Clean answer on expansion sequencing: if CDMX internal goes well, do not jump immediately to broad national rollout. First expand within CDMX reserve areas, then into adjacent Estado de México dense metros, and only then replicate the full internal model into Jalisco as the next regional hub.