Integration Framework — Case Study · Vamshi Prakash Kondapalli

01 — The Problem

Two integration failures compounding into a growth ceiling

When I joined Workllama, integrations were being built in the worst possible way — and the damage was showing up in enterprise sales cycles, customer satisfaction scores, and engineering sprint capacity.

Problem 1 — Every integration was a custom one-off build: There was no shared framework, no reusable authentication layer, no standard data mapping approach. Every new integration request triggered a bespoke engineering project. A VMS integration took 6–8 weeks. A job board took 3–4 weeks. The queue of requested integrations had grown to 14 items, and engineering couldn't get through it.

Problem 2 — No standard way to connect the tools clients depended on: Enterprise customers came to Workllama already using tools like Daxtra for CV parsing, Broadbean for job distribution, and various VMS platforms for managing contingent workers. Without native integrations, they faced manual data entry between systems — and some chose competitors with better connectivity over Workllama's superior core features.

14

Requested integrations backlogged, none standardised

6–8 wks

Engineering time per custom VMS integration

3 deals

Enterprise deals lost citing integration gaps in 6 months

The opportunity was architectural: if we could build a reusable integration layer — with standardised auth, data mapping, and sync logic — we could compress integration build time from weeks to days, clear the backlog, and turn our integration ecosystem into a genuine competitive moat.

02 — Discovery & Framing

Understanding what a framework actually needed to solve

Before defining the architecture, I spent three weeks in structured discovery — with engineering, sales, and active customers — to understand where the real friction was and what "good" looked like.

1

Engineering deep-dive — the cost of custom builds

Where all the time was actually going

Sat with two engineers who had built the most recent integrations to map every step of their process. Found that 60% of integration build time was spent on authentication flows, token management, and error handling — work that was identical across every integration. Only 40% was actual data mapping logic unique to each partner. The standardisation opportunity was enormous.

2

Sales pipeline analysis — what integrations were blocking deals

Turning lost deal data into a prioritisation framework

Worked with sales to tag every deal in the pipeline by integration requirement. Found a clear tier: VMS connectivity (Beeline, Fieldglass, SAP Fieldglass) was blocking 6 enterprise opportunities. Job board distribution (Broadbean, Idibu) was needed by 9 accounts. CV parsing (Daxtra, TextKernel) was in every enterprise RFP. These three categories became the first wave.

3

Partner API audit

What we were actually connecting to

Reviewed the API documentation and authentication models for the top 10 requested integrations. Found that 8 of 10 used OAuth 2.0 or API key auth — both handleable by a single credential management layer. Only 2 required custom auth flows. Data formats were more varied, but a configurable field mapping layer could handle 90% of cases without custom code.

4

Framework architecture decision

Build vs buy vs hybrid

Evaluated three options: build the entire framework in-house, use an iPaaS layer like MuleSoft or Workato, or build core infrastructure ourselves and use lightweight connectors for standard integrations. Chose the third — iPaaS was overkill for our scale and would have introduced vendor dependency in the platform's core. Building everything custom was too slow. Hybrid gave speed without lock-in.

03 — Architecture

Four-layer framework built for reuse, not just connection

The Integration Framework was designed as a four-layer stack — each layer handling a distinct concern, each reusable across every integration we'd ever build. The goal was that adding a new integration should require only configuring the top two layers, not rebuilding from scratch each time.

The four-layer integration stack

API & Auth layer

Standardised credential storage, OAuth 2.0 and API key management, token refresh logic, and rate limit handling. Built once, used by every integration. Eliminated 60% of per-integration engineering work.

Data mapping layer

Configurable field mapping between Workllama's data model and each partner's schema. No-code configuration for standard mappings; developer hooks for custom transformations. Reduced mapping time from days to hours.

Sync & webhook layer

Real-time webhook handling and scheduled polling for partners without webhooks. Conflict resolution logic for bi-directional syncs. Event queue for reliable delivery during partner downtime.

Monitoring & health layer

Per-integration health dashboard with sync status, error rates, and last-successful-sync timestamp. Customer-facing status page. Proactive alerting to Customer Success when an integration degraded.

04 — Integrations Shipped

20+ partners connected across 5 categories

With the framework in place, integration build time dropped from 6–8 weeks to 1–2 weeks for standard connectors. The backlog cleared in under 4 months. Here are the key integrations across the five partner categories.

🔍

CV Parsing

Daxtra & TextKernel

AI-powered resume parsing feeding directly into candidate records. Eliminated manual CV data entry. Required by every enterprise RFP.

📋

VMS

Beeline, SAP Fieldglass

Bi-directional sync with Vendor Management Systems — job orders, submissions, placements. Unlocked 6 blocked enterprise deals post-launch.

📡

Job Distribution

Broadbean, Idibu

One-click job posting to 100+ job boards via distribution networks. Replaced manual posting across multiple logins for every open role.

✅

Background Checks

Glider, Symphony

Automated background check initiation and status updates synced back to candidate records. Removed manual tracking from recruiter workflow.

💬

VOIP & SMS

Twilio, RingCentral

Click-to-call and SMS from within candidate profiles. Call logs and message history written back to the ATS automatically.

💰

Payroll

ADP, Paychex

Placement data synced to payroll systems for contingent worker onboarding. Eliminated duplicate data entry between ATS and payroll — a key request from staffing agencies.

05 — Key Decisions

The tradeoffs that shaped the framework

Decision	Options considered	Choice	Rationale
Build vs buy infrastructure	MuleSoft / Workato vs custom build	Custom + lightweight connectors	iPaaS added cost and vendor lock-in at our scale. Custom core gave us speed and control without rebuilding everything.
Sync direction	One-way vs bi-directional sync	Bi-directional where needed	VMS and payroll required bi-directional. Job boards and CV parsers were one-way. Decided per integration based on data flow requirements.
Customer visibility	Internal only vs customer-facing status	Customer-facing health dashboard	Integration failures were a major trust issue. Giving customers real-time visibility reduced support tickets and demonstrated reliability proactively.
Wave 1 prioritisation	By technical ease vs by revenue impact	Revenue impact first	VMS and CV parsing were harder to build but unlocked enterprise deals. Starting with easy wins would have cleared backlog without business impact.

"The framework meant we stopped dreading integration requests. We started treating them like configuration, not construction."

— Engineering Lead, Workllama

06 — Results

20+ integrations, enterprise deals unlocked, engineering freed

The Integration Framework transformed integrations from a recurring bottleneck into a competitive advantage. Within 12 months of launching the framework, Workllama had the deepest integration ecosystem in its competitive tier.

20+

Integrations shipped via the framework

6→1 wk

VMS integration build time before vs after framework

60%

Reduction in per-integration engineering time

6

Enterprise deals unblocked by VMS integrations within 3 months

−45%

Integration-related support tickets post-health dashboard launch

#1

Most cited platform differentiator in enterprise RFP responses

The framework was subsequently cited in Workllama's Ardent Partners Key Provider 2022 recognition and became a core part of the enterprise sales narrative — "we integrate with your existing stack" became a first-meeting selling point rather than a post-sale project.

ⓘ

On measurement: Build time comparisons (6 weeks → 1 week for VMS integrations) are based on engineering sprint logs comparing pre-framework and post-framework integration projects of equivalent complexity. The 60% engineering time reduction is an average across 8 integrations built on the framework in the first 4 months — individual results varied by partner API quality. The 6 enterprise deals "unblocked" reflects deals where integration gaps were explicitly noted as a blocker in CRM notes, and which closed within 90 days of the relevant integration going live.

07 — What I Learned

Lessons from building platform infrastructure as a product

Infrastructure is a product — it needs a PM, not just engineers. The framework didn't get built because someone had a technical vision. It got built because someone (me) translated the business cost of the status quo into a concrete product investment case. Framing 6-week integration builds as "3 lost enterprise deals per quarter" unlocked the engineering investment immediately.

Prioritise by revenue impact, not technical ease. The temptation was to clear easy integrations first to show momentum. Instead, starting with VMS (the hardest, highest-impact category) meant the first wave unblocked real deals. Momentum came from business outcomes, not shipping velocity.

Customer-facing health dashboards reduce support load better than faster fixes. The monitoring layer's customer-facing status page cut integration support tickets by 45% — not because integrations failed less, but because customers knew about issues before calling. Transparency is cheaper than responsiveness.

Reusable infrastructure compounds — the value grows non-linearly. The 20th integration took 3 days to build. The 1st took 6 weeks. That compounding is only possible if you invest in the foundation early. Platform thinking always beats feature thinking at scale.

Integration Framework — turning one-off builds into a scalable partner ecosystem