First determine the use case and business decision
Starting from customer needs, product messaging and sales progression, clarify who needs what information at what moment and what decisions they make based on it, then define the agent's tasks.
ANDY Ren Xu
AI Strategy Consultant
Enterprise AI and agent application practitioner
First clarify the work you are doing, then decide which part the Agent should take on.
You don't need to start by talking about models or writing requirements. In the first conversation, look at who does this work, where the materials are, who you need to hand off to, and which confirmation points must not go wrong; only after sorting this out can you tell whether an Agent is suited to prepare, remind, coordinate or record.
Why work with me on this?
I have over twenty years of enterprise services experience, having worked in pre-sales, marketing, training, sales management and business unit operations, and have also taken on integrated management of sales, pre-sales, delivery and development teams. When discussing a piece of work, I look at business goals, engineering conditions and team collaboration together, sort out materials, owners, handover points and confirmation points, and form a work map and Agent application judgment.
CREDIBILITY / The perspective I bring
Look at different roles within the same piece of business work.
Complex B2B work rarely belongs to just one role. An opportunity passes through customer understanding, solution preparation, product messaging, sales progression, team collaboration and delivery handover. What I do is preserve the context at these handover points, so the next step has evidence to build on.
Connect knowledge, tools, and checks to the process
Organize task inputs, product knowledge and applicable rules, clarify tool permissions, output checks and exception handling, and use prototypes and trials to test whether the agent can enter daily work.
Arrange responsibilities, collaboration and continuous improvement
Drawing on my experience running teams, I define what people and agents each handle, where handoffs occur and who reviews the results, then adjust knowledge and processes based on user feedback.
EXPERIENCE
Over twenty years, from enterprise services to agent applications.
Bachelor of Laws, China University of Geosciences (Beijing); member of Mensa. Long engaged in enterprise services, business development and team operations management; in recent years involved in enterprise AI products, agent applications and organizational AI transformation.
April 2014 to present
UINO
Experience spans sales management, solutions, delivery, and development team operations, with responsibility for integrated business unit management; in recent years, involved in AI product and agent practice.
July 2013 – March 2014
TÜV NORD Greater China
Senior Marketing Supervisor, working on marketing and enterprise services in management system certification.
June 2012 – June 2013
GDS
Inside Sales Director, built the team from scratch and took part in enterprise services business development and team management.
April 2010 – June 2012
British Standards Institution (BSI)
Business development manager, turned ISO 27001 auditor-day rules into an Excel calculator that business staff can use.
January 2005—April 2010
Epson China
Worked in pre-sales support, team management and training, and developed a repair dispatch lookup tool.
METHOD / From work to Agent
First turn the business problem into an inspectable work map.
I usually first ask about the task's object, materials, rules, participants, handoff points and risks, then decide what the Agent can prepare, what it needs to invoke, and where it must stop and wait for human confirmation.
Look at real tasks
Who does what and when? Which materials, experience and manual handoffs does it rely on today?
Organize objects and evidence
Map out customers, opportunities, tasks, deliverables and information sources, and clarify the applicable conditions and confirmation responsibilities for each piece of information.
Define the human-Agent division of labor
Agents can retrieve, organize, remind and draft; professional conclusions, commitments and irreversible actions are confirmed by humans.
Review in the flow of work
Run a trial with real materials, record gaps and exceptions, then update materials, rules, and processes.
PRACTICE / Application in practice
Put knowledge and methods into the tools the team uses every day.
Bring scattered product knowledge into sales and pre-sales work
Personally built a product knowledge Q&A assistant on Dify, organizing product knowledge, use cases and usage boundaries, and setting up a separate quality check step. The assistant is now used for product Q&A by sales and pre-sales, with a focus on answer review, business rule updates and knowledge sharing.
Connect retrospectives, opportunity judgment and next-step actions
Initiated and built a sales agent to help sales and pre-sales teams add and verify business facts, review deals, assess opportunities and plan next steps. I continue to refine its context, collaboration methods and operating boundaries as the team uses it.
Determine the solution from the user's decision scenario
Led requirements analysis, solution design and business development in data intelligence projects, coordinating business needs, product capabilities and delivery conditions. In 2019, participated in defining a digital twin product, first clarifying what users needed to understand and judge, then determining the models and features.
Turn complex rules into usable tools
Built a lookup tool for repair dispatch information and a calculator for audit person-day rules. I also led teams participating in nationwide rollout projects and the development of operations visualization standards. This work involved coordinating requirements, aligning solutions and handing over implementation.
NEXT / Start with one real piece of work
Is there a piece of work that always needs repeated preparation, follow-up, or handoff?
You can start by describing what happens on the ground: who does it, where the materials are, which step takes the most time, and what results must not go wrong. You don't need to write requirements first, or decide which model to use first.
After adding WeChat, you can say:
"Hi, I came from the website and want to talk about a piece of work."
First assess together whether the problem is suitable for building an Agent, and which judgments must remain with humans.