As Manager of Field Support Analytics at Asurion, I lead a team focused on turning data into action for our expert repair network. We analyze workflows, surface insights, and build tools that help our field and contact center teams deliver fast, high-quality repairs.

Right now, we are driving new initiatives to curate governed, context-rich data for use by everyone including generative AI. My team is helping design the process from the ground up from data rules and validation to stakeholder feedback loops and issue tracing. Our goal is to make sure AI doesn’t just generate answers, but the right ones.

With over four years at Asurion and a background in quantitative analytics, I focus on data visualization, process improvement, and building systems that scale. I’m passionate about using data to solve real problems, support real people, and drive meaningful change in the way field service operates.

Today I was honored to be named Overall Winner in our Enterprise AI Prompt Challenge at Asurion.

I also want to recognize the functional winners who were highlighted alongside me. I did my best to find them on LinkedIn and I may not have been successful but I did at least try. ;) 

Engineering – Christopher Tuncap  
Product – Brandie Martin  
Finance – Andy Patel
Customer Solutions – Teri Bellamy  
Human Resources – Karime Balelo
Security – Ian Kirk
Legal – Lauren Travis
Enterprise Technology – Karwan Hasan
Supply Chain – Pavan Akith Nakka
Marketing – Evan Press
Revenue – Dara Nunn

This wasn’t an individual moment. It reflects how broadly AI experimentation is happening across the company.

The part I’m most proud of isn’t the award. It’s what the prompt actually does.

I built a structured method for using AI to assist with Business Requirements Documentation.

Most teams use AI to generate content.
I wanted to use it to pressure-test thinking.

The prompt turns the LLM into a requirements interrogator.

Instead of “write a BRD,” it does this:

– Forces structured intake (problem statement, stakeholders, constraints, success metrics)
– Separates functional and non-functional requirements
– Surfaces assumptions explicitly
– Identifies risks, dependencies, and open questions
– Requires human confirmation before anything is finalized

Nothing is accepted unless the human agrees to it.

It’s essentially a QA loop between the analyst and the model.

The result:

• Faster draft cycles
• Fewer downstream revisions
• Better stakeholder alignment
• Higher signal in the final document

Every function gathers requirements. That’s why this scales horizontally.

The larger takeaway for me: AI is most powerful when it’s structured as a thinking partner, not just a content generator.

Grateful to be part of a team pushing practical enterprise AI forward.

#AI
#BusinessAnalysis
#Asurion
#AsurionAI

Ryan Simmons

I am a highly skilled Manager of Quantitative Analytics with expertise across various industries such as finance, retail, and insurance. Throughout my career, I have successfully developed and implemented sophisticated quantitative models, dashboards, and reporting systems that have substantially improved the financial performance of multiple organizations. My exceptional communication skills enable me to effectively present complex findings to senior management and collaborate with cross-functional teams to drive successful outcomes. As I continue to progress in my career, my goal is to leverage my extensive experience and expertise to drive innovation and optimize business processes in the dynamic field of quantitative analytics.

LinkedIn

Senior Software Engineer at Asurion, Full Stack Developer

Sr. Manager Sales Compliance at Asurion | People Leader | Compliance Operations

Leadership Development Leader | Instructional Design Expert | Driving Scalable L&D Programs with Data & AI

Human Resource Professional

Sr. Manager, DevOps Engineering at Asurion: Enterprise Data Platform

Director, Client Program Management at Asurion

I help working analysts ship stakeholder-ready insights faster using practical analytics workflows and AI.

After the recent AI prompt contest, my VP stopped by my desk to share his thoughts. He mentioned that he liked one of my submissions that didn’t win.

Curious, I asked which one out of the three he preferred. It turned out to be the prompt I created years ago when DALLE first launched. It wasn’t flashy or new, but it was designed around our brand guidelines.

Asurion maintains a strict visual identity with specific colors, layout rules, and tone constraints. I encoded these elements into a reusable prompt, enabling teams to generate on-brand visuals at any time and for any purpose.

That prompt has only grown stronger as the image generators have improved over time.

What’s funny is I almost didn’t enter the contest at all because I figured what I was doing was easy but at the last minute I entered three prompts and I entered ones I found useful in my daily work. 

#ArtificialIntelligence #OperationsLeadership #EnterpriseAI

A week ago a Senior Analyst excitedly told me they built a bot to explain their reporting to stakeholders.

This week a Director said their team is building the same thing.

I also know my Senior Manager wants to deploy a similar bot.

That’s the problem.

Smart people. Same company. Same class of stakeholder questions.

So many bots.

Multiply that across functions and you dont get innovation. You get fragmentation.

Different definitions.
Different explanations.
Different cost footprints.
Different maintenance cycles.

All solving the same pattern level problem.

This is how AI spend compounds. Not through one big bet. Through parallel, well intentioned builds that never meet in the middle.

We already lived this with semantic layers. Multiple versions of the same KPI, each technically defensible, none fully aligned.

Now we’re building the AI equivalent.

Before your team builds the next assistant, ask one question:

Is this actually a new problem, or are we recreating the same solution inside a different silo?

Wishing a happy Lunar New Year to everyone who celebrates.

New years are useful because they create a natural pause to reflect, not just on outcomes, but on systems. The last few weeks have been a reminder of how quickly structures can change. Layers flattened. Roles shifted. Some people exited. Others are currently re-evaluating their position and making plans to exit. Still others that will likely stay and while also re-learn where they sit in the organization and how to move their part of the org forward.

What’s easy to miss in moments like this is that discomfort is part of life. I have been uncomfortable before. I have sat in Kuwait waiting for a United States led invasion of Iraq to start. Nothing I have faced in a corporation will ever come close to the uncertainty I felt in the moments before a war started. That gives me both perspective and clarity in this moment. However I recognize that others who are more inexperienced may be caught in this moment alone. In these past weeks this may be the most uncomfortable they have ever felt. I respect that.

At the same time I will just share this though, this kind of reset forces clarity. Mature organizations don’t just remove layers, they reorganize how work, decisions, and accountability actually flow. When that’s done well, it creates space for better ownership. Less translation. Fewer handoffs. Clearer models underneath the work, whether that’s data, decisions, or execution.

A new year doesn’t fix any of this on its own. But it’s a useful moment to focus on what we can influence. Tightening our thinking. Being explicit about definitions and ownership. Supporting the people around us who are still finding their footing.

The good news that I can share in the new lunar year is I feel the structure is finally settling. Decisions have been made. People are using the guidance leadership shared. The new lunar year seems brighter and we are all moving forward as one team.

#lunarnewyear2026 
#persective
#leadership

Most performance problems are airflow problems.

Not literal airflow. Operational airflow.

This weekend I replaced the ceiling fan in the home gym I’m building. The old one worked. It spun. It lit the room.

But it moved almost no air.

The new one cost half as much and on speed one it was almost too much. Six settings. Real output.

Same with the baseboards. I’ve been using a Dremel to flush cut around every wall so the flooring slides cleanly underneath. Slower. More tedious. But it sets up the next step correctly.

In analytics and ops, we tolerate a lot of “it works” systems. Dashboards that technically load. Processes that technically run. Metrics that technically reconcile.

Then we wonder why performance feels stagnant.

Sometimes the upgrade isn’t strategic. It’s mechanical. Better airflow. Cleaner cuts. Infrastructure that actually supports the load.

Where in your org are you still running the old fan because it spins?

#OperationalExcellence
#ContinuousImprovement
#Leadership

Last Thursday I was asked to share a story where I lost a sense of responsibility over a problem. I shared it with a senior leader. I’ll share it with you this week.

Years ago I was assigned to build a labor forecast using a vendor system. I was trained on it. I read the documentation. I selected the algorithm that promised the most stable output based on transaction volume and time per transaction.

Hours later the forecast was done. I walked my director through the inputs, the logic, and the numbers.

Then the questions came.

Why did Market A increase 6 percent?
Why was Tuesday suppressed?
What changed versus last month?

I could explain the inputs.
I could explain the algorithm I selected.
I could not explain the movement.

The system was a black box. Numbers went in. Numbers came out. What happened in between was hidden.

And in that moment, I realized something uncomfortable.

I did not own the forecast.

Part 2 tomorrow.

#Responsibility

I used to think an analysis failed when the numbers were wrong.

Now I see another failure more often. The numbers are right, and nothing happens.

You finish the work. The review goes fine. A few good questions. Maybe even a “this is great.”

Then the cadence slips. Stakeholders stop opening the dashboard. The refresh job keeps running. View counts trend to zero. One day DevOps asks if they can shut it down because no one is using it.

I’ve had that happen.

The metric was clean. Definitions aligned. Edge cases handled. But no one had to do anything differently because it existed.

There was no trigger tied to it. No threshold that forced a staffing move, a cost decision, or an escalation. It was insight without obligation.

Lately I hear a version of this internally… the business should own it now. They have GenAI. Why should we have Analysts acting as interpreters?

I understand the intent. Decision ownership belongs with the business.

But handing over a dashboard and a chatbot doesn’t create ownership. It removes friction.

GenAI can summarize the trend. It can describe the variance. It can restate what happened.

It cannot choose the tradeoff. It cannot accept the risk. It cannot decide to shift headcount or absorb margin pressure.

That still requires a name next to the decision.

The lesson wasn’t about data quality. It was about decision ownership.

If no one is required to act when the number moves, the work will drift into the background.

So someone needs to be the responsible adult. Someone has to act or all the tools in the world will not be able to save the business.

#dataanalytics #decisionmaking #stakeholders

Part 2 of the story where I lost a sense of responsibility over a problem. I was using a vendor system to devise a labor forecast. I was successful in following my training. The forecast was right but the Director asked hard questions and I didn’t have enough detail to answer them.

The system was not broken. It was designed that way.

The vendor was protecting their IP. If the value is the algorithm, you don’t expose the gears. From a commercial standpoint, that makes sense.

But inside an operating environment, that design creates a different risk.

A labor forecast is not an academic exercise. It drives hiring, scheduling, cost, and service levels. Leaders are not just asking if the math is sound. They are asking what moved and why.

When a number swings 6 percent, they want to know which assumption did it. Was it transaction growth? Seasonality? Productivity drift? A smoothing parameter?

Instead, we had inputs and outputs. And a paragraph in a PDF.

That gap matters.

Accuracy without explainability erodes trust. And when trust erodes, the model stops being a decision tool and becomes a suggestion.

Part 3 tomorrow.

Part 1: https://lnkd.in/ewaWQkej

#Responsibility

Part 3 of the story where I lost a sense of responsibility over a problem. I was on a project to create a labor forecast, the forecast was accurate but lacked detail, I checked with the vendor and they couldn’t walk me through the math because their system was proprietary.

There is a difference between technical accuracy and executive trust.

A model can be statistically valid and still fail in the room.

Accuracy answers one question: is the math defensible?

Trust answers a different one: am I willing to bet headcount and cost on this?

In that meeting, I leaned on the fact that the algorithm was designed to smooth volatility. That it had been validated. I pointed to documentation which supported the choice and the requirements that did not want volatility.

What I did not have was a clear operational story.

If transactions spike on Tuesday afternoons, what happens to required hours?
If productivity drifts 3 percent, where does it show up?
If one market behaves differently, can we isolate why?

Without that, the forecast felt abstract. Clean. Distant from the field.

And when numbers feel abstract, leaders hesitate.

That hesitation is not about intelligence. It is about risk.

Part 1: https://lnkd.in/ewaWQkej
Part 2: https://lnkd.in/et8Pqabe

Part 4 tomorrow.

#Responsibility

Kickoffs bring excitement, and as someone in the data field, I always appreciate the value of information. This year, the Asurion kickoff theme was "Accelerate, Innovate, Grow," which inspired me to create an AI corporate song that incorporated many ideas shared by our senior leaders.

As I mentioned earlier this week, I began with a song idea, shared it with a language model, and collaborated with various bots to refine the output. It's important to recognize that there isn't just one perfect prompt to achieve the desired results... it often requires a back-and-forth process. This iterative approach is something I integrate into the ChatGPT Custom GPTs that we offer in our own GPT store for the Enterprise.

The theme of the event and song resonates with lessons I've learned through trial and error. I often tell my kids that if they don't listen to me, life will teach them in its own way. You either learn, or life will compel you to learn. With three degrees, and probably a fourth if you count the school of hard knocks, I embrace this philosophy.

I work a little each day while being fascinated by the world around me. I've been fortunate to have a wonderful partner, great kids, supportive colleagues, and leaders who uplift me. I also understand the challenges of making decisions in a fast-moving marketplace, and my life experience as not only an analyst but a person who fully understands workforce management helps me appreciate the numbers that drive demand, labor, and its many permutations and fluctuations.

We truly live in a fascinating time. 

#asurion
#leadership
#accelerate
#innovate
#grow

Part 4 of the story where I lost a sense of responsibility over a problem. I was on a project to produce a labor forecast, I was trained on a vendor system designed to accept inputs and output a forecast, but it lacked detail to answer any questions. I pointed to the vendor system and it’s documentation.

This is where I lost responsibility.

I had operated the system correctly. I followed the documentation. I selected a defensible algorithm. I produced the forecast on time.

But I treated the model as the owner of the answer.

When it was challenged, I defended the tool instead of interrogating it. I explained how it worked instead of whether it made sense.

That is a subtle shift.

If you cannot trace a number back to an operational driver, you are borrowing credibility from the model. You are not standing on your own reasoning.

The vendor owned the code.

But I owned the outcome.

I just had not accepted that yet.

Conclusion Tomorrow. 

Part 1: https://lnkd.in/ewaWQkej
Part 2: https://lnkd.in/et8Pqabe
Part 3: https://lnkd.in/eMMKNh-t

#Responsibility

Part 5 of the story where I lost responsibility over a problem. I had been trained on a vendor system designed to produce a forecast based on inputs. I had followed my training but when questioned I cited the documentation and when I looked for detail on how the forecast was derived it wasn’t there. I lost responsibility over the problem when I couldn’t justify changes the model made over time. There was only one solution. 

The team and I had to rebuild the forecast outside the black box.

Development was slower. We made the model simpler on purpose because explainability was critical. I worked closely with the vendor to understand how to feed their system a complete forecast input while retaining transparency on our side. Now we had something that was end to end something we could explain. 

In the end, we broke it into core components. Volume by market. Time per transaction. Daypart patterns. Seasonality. We pressure tested assumptions against what was actually happening in the field.

It was not as polished as the vendor output. But it was explainable from the front line to the executive level. It aggregated cleanly, and more importantly, it gave leadership control over the assumptions driving cost. That control ended up being more important than I realized at the time. 

If a market moved 6 percent, we could point to the driver. If transactions slipped, we could show the sensitivity.

The model stopped being something we ran and became something we understood.

In the end tools are useful. Vendors protect IP. Algorithms can be statistically sound.

But if a number drives hiring, cost, or service, you are responsible for it whether you built the system or not.

If you cannot explain it in operational terms, you do not own it.

That lesson has stuck with me. I build data stories now and make sure I can walk operations leaders through them. If I don’t understand them myself I rework everything until I do. I anticipate questions better and I smile when leaders ask the same questions I asked myself during the build. I smile because I am ready for them this time.

This is the end of my story but tomorrow we will talk about how I tackled the new black box, GAI.

Part 1: https://lnkd.in/ewaWQkej
Part 2: https://lnkd.in/et8Pqabe
Part 3: https://lnkd.in/eMMKNh-t
Part 4: https://lnkd.in/eAC3nGRZ

#Responsibility

AI costs dont explode all at once.

They compound quietly.

This week I told a few senior leaders something that felt small but isnt.

I created markdown files to help AI understand the staging layers of a data model. Clear definitions. Clean transitions. Context for joins.

It worked.

Then another analyst does the same thing.

Then another.

Each one feeds the model more tokens. More context. More repeated structure. More staging documentation copied into prompts.

Token costs compound over time. Not because one person was careless. Because everyone was being thoughtful in isolation.

If ten analysts all paste 2,000 extra tokens into daily workflows, that isnt experimentation anymore. Its operating expense.

The risk isnt bad prompts.

Its ungoverned scale.

We need to think about AI governance the same way we think about data governance. Shared context layers. Standardized model summaries. Central artifacts instead of duplicated markdown sitting in personal folders.

Otherwise we will optimize locally and pay globally.

Unlimited AI feels great right up until the invoice hits.

AI companies love free trials. Generous limits. “Unlimited” access. Most teams never model what happens when that ends.

What is your monthly budget per person for AI tokens?

Not in theory. In dollars.

What’s your daily driver model? Is it the flagship model because it’s impressive, or because the problem requires it?

I’ve seen teams burn premium model spend summarizing meeting notes and drafting status updates. That’s a sports car doing grocery runs.

Can a smaller, cheaper model handle 70 percent of your workload? If so, why is the expensive one your default?

These are the questions I push with my team. Before usage scales. Before finance asks.

Are you actively managing model mix and token budgets, or just enjoying the free ride?

AI tokens surface a management problem most leaders avoid.

We say we want experimentation.
We hand out access.
We celebrate usage.

Then we’re surprised when the bill scales with curiosity.

The real question isn’t “how many tokens per person?”

It’s “for what class of work?”

Are tokens being used to accelerate defined workflows?
Or to explore loosely defined ideas with no decision attached?

Exploration has value. But it needs a lane and a limit.

Otherwise token growth tracks with ambiguity, not impact.

I’d rather see disciplined, high-impact token usage than broad, impressive activity.

Do you tie token access to decision responsibility, or is it just a perk?

High AI token usage is usually a framing problem.

If your token bill is climbing, your decision clarity might not be.

AI doesn’t create ambiguity. It just makes it expensive.

Every analyst gets excited at first. Big context windows. Flagship models. You feel invincible. You spin up background agents that run overnight.

Then it stops.

Monthly token limit hit.

Now what?

Do you slow work down?
Go back to the old way?
Ask finance for more coins?
Or just wander the arcade watching others play?

Most teams never design for this moment. They treat AI like an always-on utility instead of a metered resource.

In operations, we model labor to volume.
In cloud, we model compute to workload.

AI is next.

The real question isn’t “how powerful is the model?”

It’s “what is the sustainable token budget per decision?”

We’re entering a phase where model judgment will be evaluated.

Not just output quality.
Cost per interaction.
Model choice by task.
Token burn per analyst.

If two leaders drive similar outcomes and one spends 3x the tokens, that gap won’t stay invisible.

Are you building model discipline now, or waiting for finance to build it for you?

Since Sunday I’ve been telling a story about losing responsibility to a black box model.

Now let’s talk about the new one.

A few years ago it was obvious we were in another shift. I’ve seen a few of these. The early Internet. The move to Cloud. The rise of SaaS.

AI felt different, but not unfamiliar. A new abstraction layer. A new dependency risk.

So I did what I usually do when the ground moves. I went to books.

When GPT 3.5 was released in  March 2023, O’Reilly published a book on Prompt Engineering by James Phoenix and Mike Taylor around a couple months later. That was my entry point. I trust O’Reilley as I have been reading their tech books for years.

What struck me wasn’t clever phrasing. It was measurement.

Prompt engineering is not about getting a good answer once. It is about designing inputs so you can run them repeatedly and evaluate the outputs.

Run it once and you are impressed.

Run it a thousand times and you start asking harder questions.

How stable is it?
Where does it drift?
What variables change the answer?
What does failure look like?

Many people treat prompt engineering as a gimmick. As an analyst, I saw something else.

It was a framework for interrogating the black box instead of admiring it.

If AI is going to influence cost, staffing, service, or strategy, the responsibility problem is back.

The tool is new.

The ownership question is not.

#Responsibility

Extra shine rarely fixes ambiguity.

I stopped chasing polish when I realized clarity was doing more work. Last week I saw a small misunderstanding ripple through a conversation. It wasnt a data issue. It was framing.

The easy move is to add more context. Another view. Another annotation. But more context can dilute the signal.

Stakeholders dont need every dimension. They need to know what changes when the number changes.

Speed helps. AI can compress the build cycle. But it doesnt automatically surface the human tension behind the data. That still comes from sitting with the stakeholder and understanding what theyre accountable for.

Ship something usable. Let it run into real questions. Then tighten it.

Thats where impact shows up.

Where have you seen polish mistaken for progress?

I think a lot of companies and teams are deep in AI fatigue. New tools every week. Training sessions. ChatGPT Enterprise already in play. A few weeks ago I was still using VS Code with Copilot and didnt see the point when another team started pushing Cursor.

Then our CTO mentioned Cursor at a meeting and our software development team held a live working demo during kickoff. I went ahead and started using the tool.

The difference showed up once immediately. 

I started letting it build its own working context under my supervision.

For example, I had it read PBIP files and generate markdown context files so it understood the reporting model structure. Then I designed prompts that let it explore the project and summarize what mattered before I started asking questions. This gave the bot a level playing field so we could have a constructive conversation about the problems.

Research that used to take hours or days got cut down pretty quickly.

I shared what I learned and now the entire Field Analytics team is shifting to that model. I believe having the entire team working to learn and experiment together will drive further discoveries and make us all better. 

As the team catches up I am moving to experimenting with better orchestration. I’m moving from just having it build documentation to having it try to work a few minor tickets, again under supervision.

The idea is I will have Cursor draft a plan, run that plan past another model for critique, then bring the feedback back to Cursor and force it to accept or reject the changes.

Then watch it work the ticket with the context documentation another bot created.

In testing I’m starting to start to see the reasoning tighten.

The lesson for me is simple. Fail fast, fail often, but don’t be afraid to experiment.

Most people are still using these tools like autocomplete.

They get more interesting when you treat them like a employee that needs a bit of coaching from time to time. 

Curious how many teams are actually doing that versus just using AI for code suggestions.

#Cursor #ArtificialIntelligence #AIOrchestration

taplio

Ryan Simmons's Linkedin Analytics

Ryan Simmons

Performance ScoreRyan Simmons's LinkedIn Performance

What is Ryan talking about?

Ryan Simmons's Best Posts (last 30 days)

Ready to grow your LinkedIn presence?

taplio

.css-6ytdur{background-image:linear-gradient(284deg, rgba(39, 43, 72, 1) 0%, rgba(33, 47, 183, 1) 100%);color:transparent;-webkit-background-clip:text;background-clip:text;}Ryan Simmons's Linkedin Analytics

Ryan Simmons

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Ryan Simmons's LinkedIn Performance

What is Ryan talking about?

Ryan Simmons's Best Posts .css-1emoo84{font-size:var(--chakra-fontSizes-md);font-weight:400;color:var(--chakra-colors-gray-600);}(last 30 days)

Ready to grow your LinkedIn presence?

Ryan Simmons's Linkedin Analytics

Performance ScoreRyan Simmons's LinkedIn Performance

Ryan Simmons's Best Posts (last 30 days)