How enterprise transformation has evolved from Lean Six Sigma to the agentic era

• The process excellence era (early 2000s): When process transformation meant reducing variation, defects, and waste, lean tools like value stream mapping and the define, measure, analyze, improve, and control (DMAIC) framework were the playbook for structured improvement

• The digital automation era (2010–2015): As robotic process automation (RPA) and workflow tools gained traction, it became clear that automation doesn't fix broken processes. LSS helped to identify the well-understood, clean, rule-based tasks that could be automated to deliver business value

• The predictive analytics era (2015–2020): As more processes went digital, they created the data to power AI models that could predict failures before they happened. LSS helped deliver reliable data inputs and structured output interpretation to boost model trust and adoption

• The AI and agentic AI era (mid 2020–present): Agentic systems don't just predict or recommend – they act. But for this kind of autonomy to work, you can't just plug an agent into a messy process. LSS can help clean up workflows, define measurable outcomes, and deliver continuous improvement

There are four key phases of the agentic AI lifecycle, and LSS can play an important role in each.

1. Identifying the right use cases

Choosing the right starting point matters. Many AI programs fail not because of the model, but because they're solving the wrong problem. LSS can help solve two key challenges:

• Define: Voice of the customer (VOC), suppliers, inputs, process, outputs, and customers (SIPOC), and cost of poor quality (COPQ) tools can help understand customer pain points, desired outcomes, and business priorities

• Measure: Time studies, rework rates, and throughput analysis can establish the baselines to measure impact against

A global consumer goods company was dealing with thousands of customer deduction claims. Using LSS, it identified a lot of effort being wasted chasing disputes of no real value. LSS helped it identify where to deploy agentic solutions in the part of the process with the highest manual effort and impact potential.

2. Building agentic systems

Unlike traditional software, AI agents need ongoing adjustments as they evolve and learn in live environments. This is why the DMAIC framework is a good match with the agent development lifecycle.

• Define: What should the agent do, and what counts as failure? For example, a vendor helpdesk agent could be expected to resolve queries related to payment status, purchase order mismatches, or deductions. A failure could be a wrong answer, an unnecessary escalation, or simply a slow reply

• Measure: Track how the agent is performing, both technically and in terms of its business impact. Metrics such as first-time-right response, average resolution time, escalation rate, and agent confidence scores are vital

• Analyze: When things go wrong, LSS can help find out why. In one deployment of Genpact's vendor query agent, sudden spikes in escalations were traced to missing payment terms in the underlying data feeds – a data quality issue rather than a model problem. Fishbone analysis helped to quickly isolate the problem

• Improve and control: Guardrails such as confidence thresholds, fallback workflows, and alerts for anomalies can help monitor agent behavior and update edge cases

3. Deploying agentic systems in real-world operations

Here's where most agentic deployments get stuck. You've trained the model, but now it needs to work inside live processes with real business consequences. Deployment isn't just hitting the go-live button. It involves several coordinated steps:

• Process readiness checks: Are exceptions defined? Are the upstream and downstream systems stable?

• Role redefinition: Who steps in when the agent fails? What new skills are required?

• Pilot phasing: What percentage of work does the agent handle to start with?

• Operational monitoring: Are you capturing the right performance signals?

LSS can also help stabilize the upstream data pipelines that power agentic systems. AI agents are only as effective as the data they are trained on or operate with. By applying LSS tools such as root cause analysis, process mapping, and measurement system analysis (MSA), teams can identify and resolve inconsistencies in master data, delays in real-time feeds, or structural issues in how data is captured and transformed.

Tools like value stream mapping, the 5S methodology, and error-proofing can be used to clean up the deployment environment. Then, digital control charts and dashboards can monitor the rollout.

While deploying Genpact's AP Suite for a global manufacturer, the team followed a phased go-live plan:

• Week 1–2: The agent handled about 10% of invoices for North America

• Week 3–5: This expanded to EMEA and roughly 40% of invoices

• Week 6–8: Full-scale rollout including vendor helpdesk and payment scheduling agents

During the rollout, LSS helped standardize the exception handling process across regions, segment invoice types, and clean up the vendor master data that fed into the agents. By month three, the agents processed at least 75% of transactions autonomously.

4. Realizing sustained value

Too many companies declare victory once an agent is live. But real value is only realized when business outcomes change for good.

This is where Genpact's proprietary Smart Enterprise Processes (SEP) framework is useful. Using value trees, SEP traces operational metrics back to financial impact, such as:

• Agent reducing invoice-on-hold cases → reduces aging → improves cash flow
• Faster query resolution → increases supplier satisfaction → boosts on-time delivery
• Better classification logic → fewer routing errors → reduces labor rework cost