Intelligent Systems & Product Architecture
A B2B decision product, drafted with distinct architecture and built through
to a working MVP.
Problem
[PLACEHOLDER: avatar]
“We spend too much of our precious time and human resources researching prospects and inconsistently scoring leads that may not be a potential match. We need a consistent, research-backed way to evaluate fit so our reps can focus their time and effort on the right opportunities.”
Mid-market B2B sales team · Lead qualification & prospect research
A mid-market B2B sales team of five loses up to ₹35L a year in payroll to non-selling work. There is an unnoticed need for an organised system that provides, at a minimum, a researched and automated initial point of evaluation, so that not all proposals are pushed into human review, wasting time.
| Failure | Why it happens | Cost |
|---|---|---|
| Unstructured research | Analysts pull from disconnected sources with no shared evaluation framework. Two people scoring the same prospect reach different conclusions. | 45–50% of analyst time |
| Subjective fit | No shared way to check a prospect’s scale, regulation, and strategy against what the seller can actually deliver. | Not auditable |
| Generic outreach | Sequences are not grounded in the prospect’s verified operating pressures. | Low reply rates |
What is missing is a consistent decision a sales team can review, override, and reuse.
Product
One organisation name in. A CRM entry, a compatibility verdict, and a personalised outreach draft out. Runtime is about ten minutes. No manual research steps.
| Outcome | What the user gets |
|---|---|
| Shared profiles | Seller capabilities are distilled from internal documents into a consistent eight-field profile, while prospects are built from live web and news signals using a standard six-field profile. This creates a consistent foundation for every evaluation. |
| A verdict with a reason | Prospects are evaluated across six business dimensions, with a review pass adding commercial context and preserving the reasoning behind every final verdict. |
| Human-approved outreach | The system identifies relevant decision-maker roles and drafts personalized outreach grounded in the prospect evaluation. Drafts are held for rep approval before anything is sent. |
Architecture
The system was drawn before it was built. Open the full diagram first, then any component card for that step’s detail.
Company knowledge sits in a retrieval store and is queried on each run, so the profile updates when documents change. Target search and news feed one prospect profile. Governance sits inside the linear flow; outreach runs only after a proceed classification.
What the system returns
One run produces a reviewable verdict, then — when classification is proceed — role-specific outreach held as drafts.
Decisions
| Decision | Why it holds |
|---|---|
| Score first, then review | The weighted score is the objective seed. The review applies commercial judgment on top of it and records the adjustment. The output can be explained, not only accepted. |
| Retrieve company knowledge | Seller context is queried from stored documents on each run, and it updates when the knowledge base changes. |
| Hold outreach for approval | Outreach is marked as a draft and waits for a person. The system recommends. The rep approves. The CRM record, the verdict, and the audit trail stay attached to that decision. |
| Build qualification first | Every later step (research, outreach, CRM) inherits the qualification decision. Without a shared yardstick, effort goes into leads that were never a fit, so the scoring and review layer came before anything downstream. |
| Prove one loop end to end | One organisation name in, a CRM entry and a draft out. A narrow, complete loop shows whether the verdict can be trusted before adding volume. |
Impact and how I'd measure it
Modeled scenario: for a five-person sales team processing 30 leads a week, the model estimates up to ₹32L a year in analyst capacity freed, net of about ₹7L in operating cost.
The key hypothesis is that better-qualified leads and grounded outreach improve conversion. I'd test this with a pilot of 50–100 real leads against an analyst-only baseline, measuring research time per lead, reply rate, and qualified-to-meeting rate.
Evidence
What exists today — open the artifacts below.
Next
What holds
Separating the score from the review is what makes the verdict usable in a sales org. A reviewer can see why the number moved, and a person still approves anything that would reach a prospect.
What to test
The model assumes better-qualified leads and grounded outreach lift conversion. The next step is a small pilot against actual close rates.