title: Roadmap v4: value-driven sprints S1–S6 (18 Oct 2026 – 9 Jan 2027), two revenue tracks plus platform, value model, engineering practices status: draft for Atul (21 Oct); replaces sprints-2026-10-18.md v3 from S2 onwards; numbers in the value model are Atul's working assumptions, to be refined author: Claude (chat, MMDEP-PM) date: 2026-10-21

Roadmap v4¶

1. Value model (working assumptions, Atul 21 Oct)¶

Track	Measure	Working value	Source of the number
R1 Website	Lead (form, call or WhatsApp click recorded in the CRM)	₹10 per lead	CRM plus site events
R1 Website	Search impression (Google Search Console)	₹0.10 per impression	Search Console API
R1 Website	Visitor engaged minute	₹0.10 per minute	Analytics (GA4 engagement time, or our own events)
R2 ULIP renewal	+1 percentage point renewal rate (any policy year)	₹10,000 per point (to refine)	Renewal register, cohort by policy year

Baseline renewal curve: year 1: 90%; year 2: 80%; then −10 points a year (70%, 60%, …). The target is to lift every year's rate. Points gained are measured against the same cohort last year.

Sprint value = Σ(leads × 10) + Σ(impressions × 0.10) + Σ(engaged minutes × 0.10) + Σ(renewal points × 10,000). It's computed automatically each sprint (§4, item V1) and shown on the sprint page. Investment and expectations grow when the value line grows (§6).

2. Sprints (two weeks each)¶

Sprint	Dates	R1 Website (SEO, retention, leads)	R2 ULIP renewals	Platform and engineering
S1	18–31 Oct (running)	SEO basics live (canonicals, titles, sitemaps, breadcrumbs); route crawler and daily route health	Renewal LLD: source of due dates, cohort counts per policy year, consent and channels (questions §5)	Harness on GitHub and GitLab, merge by risk, actions sync, knowledge extraction, memory rule §21, stgapi restored, staging data load, API write-security LLD
S2	1–14 Nov	Measurement first: analytics and Search Console baselines; events for impressions, engaged minutes and leads; lead forms into the CRM; Core Web Vitals check. First fund factsheet page from the production API	Renewal due list (T-45 to T+30), totals only in MMDEP (client data stays in the client-class system). Caller worklist (sheet or CRM view). Message templates drafted. SMS sender/template registration (DLT) and WhatsApp Business setup started	Server (SRV-1) plus LiteLLM (POOL-1); inventory machines wired; C4 and ADRs for all projects (ARCH-1); MkDocs site; API write security on staging; value ledger V1; backups and restore drill (§3)
S3	15–28 Nov	Programmatic fund pages (API data plus factsheet parsers), internal linking, structured data; calculators with lead capture	Automated reminders (T-30, T-15, T-7, due, grace) on WhatsApp/SMS/email; callers log outcomes; ULIP fund performance shown in reminders (fund enhancement)	Hosted MMDEP MCP gateway plus the database decision; API read-route security; production write security enforced; developer onboarding pack
S4	29 Nov–12 Dec	Retention: watchlist, weekly fund digest (opt-in), NFO and fund news pages from MM-Cron; repeat-visit tracking	Renewal risk score (rules first: payment history, contactability, fund returns, policy year); cohort dashboard with renewal rate per policy year against baseline	Entry-level developer starts (trust level T1); performance tests; incident runbooks plus a postmortem template
S5	13–26 Dec	Lead scoring and nurture (start of lead-conversion automation): score from pages and calculators used; follow-up sequences	Upsell and cross-sell prompts for renewing clients (top-up, SIP, health/term), with callers scripted	Second (good) machine in the pool; Aider as default engine if B19 proves it
S6	27 Dec–9 Jan	A/B tests on page layouts and forms; SEO content cadence	Optimise timing and channel by measured results; renewal agent (human plus automation) playbook	Review of investments by value line; plan Q1

Carried throughout: - NAV data quality (API priority 1); - the "APIs only" move (no direct DB reads from the websites); - catalogue APIs (MCRM); - PE on corrected data; - SuiteCRM integration stays lowest.

3. Engineering practices not yet covered (to add)¶

Practice	Why	When
Production monitoring and alerting (uptime, error rates, slow pages, NAV freshness) with alerts to the new phone	The collector sees problems hourly, but nobody is alerted	S2 (collector plus push alerts), S3 (error tracking in the API and site)
Backups and restore drills (databases, WordPress, server configs)	A backup that's never been restored isn't a backup	S2: verify the backups exist; restore drill on staging; then quarterly
Branch protection and 2FA everywhere (SEG-1, still open)	Prevents accidental or hostile pushes	S2
Secret rotation and access reviews (SEC-2: SuiteCRM secret, database users)	Old exposures stay exploitable until rotated	S2, then quarterly
Release management: versions, changelogs, rollback plans, feature flags for risky changes	Safe, reversible releases	S2 onwards (part of workflow i)
Infrastructure as code: server setup scripted (Fabric tasks or a shell script in the repository)	A rebuild after failure takes hours, not days	S2 (new server), S3 (staging)
Privacy and compliance: DPDP Act 2023 consent and retention for leads and renewal contacts; TRAI DLT for SMS; consent for WhatsApp templates and call recording; IRDAI distribution rules	Required before automated renewal messages	S2 (design), S3 (before go-live)
Product analytics and experiments (event plan, A/B)	Without measurement the value model can't run	S2 (events), S6 (A/B)
Performance and Core Web Vitals	SEO ranking and visitor retention	S2 (baseline), S3 (fixes)
Accessibility (WCAG basics)	Reach, plus SEO	S3
Single point of failure: the laptop (collector, harness, runner)	One machine down stops tiers 2 and 3	S2 server plus inventory machines; heartbeat already flags it
Dependency licences	Open-source licence obligations	S3 (licence scan with pip-audit or similar)
Postmortems (blameless, short) for incidents (stgapi down 24 Sep–19 Oct, NAV gaps)	Learn once, not repeatedly	S2 template; the first two written from the TEA and knowledge extraction

4. New items arising¶

V1 Value ledger: mmdep value pulls Search Console, analytics events, CRM lead counts and renewal totals (totals only), applies §1, and writes reports/value/S<n>.md plus a trend on the sprint page. S2.
Fund pages from factsheets: the factsheet parsers (OWP) feed programmatic fund pages, which drive impressions and minutes for R1.
Renewal data stays client-class: due lists, phone numbers and messages run only in an approved India-region system (CRM or the insurer portals). MMDEP sees counts and rates only (DR-009).
Wired pool: a small gigabit switch, fixed addresses (DHCP reservations), Wake-on-LAN so idle machines sleep and wake on schedule; LiteLLM health checks.
The new phone: the business WhatsApp number for renewals, plus the alert destination for monitoring.
Entry-level developer: a T1 trust level (DR-008 §7). Same issue and routing flow as the agents, harness gates on every PR, the MkDocs site plus skills as onboarding, pairing with agents on small items first.

5. Questions for Atul (answers shape S2)¶

Website analytics: do we have Google Analytics and Search Console on www.moneymarche.com, and who owns them? Current monthly impressions, visitors, average engagement time and leads, for the baseline.
What counts as a lead: form submit, call click, WhatsApp click, calculator email? And where do leads land today (SuiteCRM, email, sheet)?
ULIP renewals:
where do the due dates live (insurer portals, MCRM, Excel)?
how many ULIP policies are in force per policy year?
which insurers?
Channels:
is there a WhatsApp Business account or provider?
is an SMS sender ID / DLT registration in place?
callers: in-house or agency, and what do they log in today?
Consent: do existing clients have recorded consent for WhatsApp/SMS reminders?
New machine: RAM and GPU? (A GPU changes which models we can run locally.)

6. How investment grows¶

At each sprint review: - compare the value line with the sprint's spend (APIs, server, people); - if value rises for two sprints, raise the next sprint's capacity (machines, API budget, then the developer); - if not, fix measurement or direction before adding spend.