Lowe’s gets answers faster with GPT-5: How AI is reshaping home improvement retail

In a short feature created by OpenAI, I discuss how Lowe’s is using GPT-5 to simplify the way customers shop for home improvement, improve associate productivity, and transform core business operations. As Senior Vice President of Data, AI, and Innovation at Lowe’s, I explain why we built our AI strategy around three pillars—how we shop, how we sell, and how we work—and how GPT-5 is proving to be a powerful tool across all three.

In this report-style article I’ll walk you through what we’ve learned so far: the practical capabilities of GPT-5 that matter to retail, how those capabilities translate into customer and associate experiences, the design principles behind our AI approach, the implementation steps we’ve taken, the early operational benefits, and the considerations any large retailer should weigh when adopting a similar strategy. I’ll weave in direct observations and quotes that guided our journey, expand on technical and business implications, and offer concrete examples to make the impact tangible.

📢 Executive summary

I’ll begin with a concise overview for readers who want the main points up front. GPT-5 has helped Lowe’s reduce the number of steps required to answer complex customer and associate questions by holding context longer and applying reasoning across multiple inputs. That decrease in friction means faster, more accurate answers, fewer handoffs, and more confident associates on the sales floor and in support roles. In practical terms, this translates to improved customer satisfaction, reduced time-to-resolution, and more efficient operations.

We aligned our AI efforts to three strategic pillars: how we shop, how we sell, and how we work. GPT-5 intersects all three pillars because it can maintain conversational context across longer interactions, combine different data sources to reason about a problem, and generate actionable, accurate responses that reduce the number of steps required for a human to get to a solution.

"With every generation, the GPT continues to raise the bar and surprise us in a very positive way." — Chandhu Nair, Senior Vice President, Data, AI, and Innovation, Lowe’s

🛒 How we shop: Improving the customer journey

One of the toughest problems in home improvement retail is the complexity of customer projects. Customers often arrive with a mix of unclear goals, partial measurements, conflicting product preferences, and anxiety about making mistakes on projects that can be expensive or time-consuming. As I say, "Simplifying how you shop for home improvement is not an easy task."

GPT-5 helps here in several ways:

Longer contextual understanding: Customers rarely communicate everything they know in a single message. They might share details over the course of multiple interactions—measurements, photos, budget constraints, timeline, and style preferences. GPT-5’s ability to hold context longer means it can remember and reason over these details without repeatedly asking the same clarifying questions.
Multi-input reasoning: Home improvement problems frequently require synthesizing different input types: a photo of a backsplash, a measurement, and product catalogs. GPT-5 can integrate disparate inputs and provide a cohesive, actionable answer.
Reduced steps to resolution: By combining contextual memory and reasoning, GPT-5 can lower the number of back-and-forth steps required to reach a recommendation or solution. That leads to faster, smoother customer journeys.

Concrete customer scenarios

To make this concrete, consider a customer who wants to renovate a small kitchen. The customer begins with a request: "I want to redo my backsplash and countertops but I’m not sure what materials to choose." Over a conversation that could span chat, image uploads, and voice, GPT-5 can:

Remember the kitchen size and budget shared earlier in the chat.
Analyze an uploaded photo to identify existing colors, lighting, and layout constraints.
Pull product compatibility information (for example, which sealants match a chosen countertop material).
Offer a short list of product pairings and an estimated material quantity with a shopping list.

This reduces the customer's need to consult multiple pages, call support, or visit the store multiple times. The associate or digital assistant provides a coherent plan quickly, and the customer gains confidence to proceed.

Personalization at scale

Personalization in home improvement is different from e-commerce personalization in fashion or media. It requires technical accuracy combined with stylistic sensitivity. GPT-5’s reasoning capabilities help generate recommendations that not only fit the customer’s technical constraints but also align with style preferences. For example, GPT-5 can weigh lighting conditions (from a photo), suggest color families, and then recommend products that fit those color families and installation constraints.

That personalization leads to higher conversion rates and fewer returns. More importantly, it increases customer satisfaction because recommendations are tailored and feasible for the customer’s specific project.

💬 How we sell: Empowering associates and omnichannel experiences

Our associates are at the heart of the Lowe’s experience. A home improvement project is often complex enough that customers want expert guidance. Historically, associates have had to pull data from disparate sources—product databases, inventory systems, installation guides, and corporate knowledge bases—to answer a single question. I’ve observed firsthand that GPT-5 "has been able to hold the context longer, apply reasoning to multiple sets of inputs, and really get to that accuracy."

Here’s how GPT-5 supports sales and associate workflows:

Faster, accurate answers: Associates can get to the right answer in fewer steps, which speeds up service and reduces cognitive load.
Unified knowledge access: GPT-5 can ingest and reason across multiple knowledge sources, enabling one-stop answers instead of manual lookups.
On-the-floor support: Real-time conversational tools can assist associates during customer interactions, suggesting next steps, compatible products, and installation tips.

Reducing friction in associate workflows

Before integrating GPT-5, an associate might have had to open three or four different systems to confirm compatibility, stock availability, and installation requirements. That process could involve toggling between desktop systems, calling a manager, or directing the customer to online resources. With GPT-5, those steps are reduced because the model can reason across inputs and respond directly.

For example, an associate helping a customer choose a replacement door may need to confirm door dimensions, hinge types, and paint compatibility. GPT-5 can combine that information from a quick chat with the customer, an uploaded image, and internal spec sheets to propose a short list of candidate doors and next steps for measurement and installation.

Omnichannel synergy

One of the strengths of GPT-5 is how well it enables omnichannel experiences. A customer might start a project on their phone, continue the conversation in-store with an associate, and later receive follow-up guidance via email. GPT-5’s context retention ensures that the conversation remains coherent across channels. That continuity increases trust and decreases the chance of miscommunication.

⚙️ How we work: Operational improvements and innovation

The third pillar—how we work—focuses internally on productivity, data-driven decision making, and innovation. I framed our AI strategy around "how we shop, how we sell, how we work," and GPT-5 plays a role in streamlining internal processes and enabling smarter operations.

Operational improvements enabled by GPT-5 include:

Knowledge management: GPT-5 helps surface relevant internal documents, best practices, and playbooks for employees, reducing ramp-up time and improving access to tribal knowledge.
Data synthesis: The model can summarize long reports, extract actionable insights, and prepare concise briefings for leadership.
Cross-functional collaboration: GPT-5 can translate domain-specific jargon across teams, making it easier for product, supply chain, and store operations teams to align.

Example: from data to decision

Imagine a supply chain manager needing a quick synthesis of regional inventory trends and the impact on upcoming promotions. The traditional approach might require pulling several spreadsheets, running analyses, and drafting a summary. GPT-5 can accelerate the cycle by extracting key trends from the data, suggesting operational actions (like re-routing inventory or adjusting promotion timing), and drafting a concise executive summary for review.

That speed reduces time-to-decision and enables the business to be more responsive. When leadership needs a quick read on whether a promotion is likely to create stockouts, GPT-5-augmented workflows can provide the evidence and recommendations much faster.

🔍 Why GPT-5 matters: Context, reasoning, and fewer steps

The changes we’re seeing with GPT-5 are not just incremental. They represent qualitative differences in how models support real-world workflows:

Longer context retention: The model remembers more of the conversation and can use that information reliably across subsequent turns.
Multi-input reasoning: GPT-5 can integrate text, images, and other structured inputs to produce a single, coherent answer.
Reduced handoffs: Because the model can synthesize diverse inputs, there are fewer times when a question must be escalated or bounced between systems or experts.

In short, the number of steps required to get to the right answer has come down. That’s a key metric for us, because each step in a customer or associate journey is an opportunity for friction.

"The number of steps have definitely come down, which means it makes it much easier for the associate to get to the answer for our customers." — Chandhu Nair

🧭 Implementation approach: How we deployed GPT-5 at scale

Deploying a powerful AI model across a national retail footprint requires careful planning. I’ll outline the high-level phases we followed and the design principles that guided our work. These steps are practical and can serve as a blueprint for other retailers considering a similar path.

Phase 1: Pilot and hypothesis validation

We started with narrow pilots focused on high-impact use cases where we could measure clear outcomes. Examples included:

Associate support for in-store customer questions.
Digital planner assistance for small remodel projects.
Internal knowledge summarization for store managers.

For each pilot we set success metrics—time-to-answer, number of steps reduced, customer satisfaction scores, and associate satisfaction. This allowed us to evaluate whether GPT-5 materially improved the experience rather than just adding another point solution.

Phase 2: Integration and orchestration

Once pilots validated the potential, we moved to integrate GPT-5 with our operational systems—inventory, product catalogs, scheduling, and CRM. This phase required building APIs and orchestration layers so the model could access the right data in real time without exposing sensitive systems directly.

Key concerns here included:

Ensuring the model’s responses were grounded in authoritative internal data.
Maintaining low-latency responses suitable for in-store interactions.
Implementing fallbacks when the model’s confidence was low, routing to human experts as needed.

Phase 3: Scale and governance

At scale, governance becomes critical. We established guardrails for safety, privacy, and accuracy. This included:

Policies for data access and retention.
Monitoring for hallucinations or incorrect factual statements.
User experience patterns that clearly indicate when suggestions are AI-generated and when human judgment is recommended.

We also built a monitoring stack to track key metrics: model accuracy, customer satisfaction, time saved, and instances where human escalation was required. Those metrics drove iterative improvements to prompts, grounding strategies, and integration points.

🪜 Design principles that guided us

Several core principles shaped how we implemented GPT-5 at Lowe’s. These principles helped us keep the focus on usable outcomes rather than chasing technological novelty.

Human-centric design: The AI should augment human expertise, not replace it. Our goal was to make associates more effective and customers more empowered.
Grounded answers: Responses must be traceable to authoritative sources—product specs, installation guides, and inventory systems.
Minimal interaction steps: Reduce friction points and the number of steps needed to reach a solution.
Safe defaults and fallbacks: When the model is uncertain, it should ask clarifying questions or defer to a human.

📈 Early results and qualitative impact

While I won’t share specific proprietary numbers here, I can describe trends we observed during early deployment:

Faster resolutions: Associates reached answers more quickly, improving customer throughput and satisfaction.
Fewer escalations: Because the model could synthesize information across systems, associates were able to resolve more questions without manager intervention.
Higher confidence: Both customers and associates reported more confidence in recommendations that were generated from combined contextual inputs.

Those qualitative improvements reduced the cognitive overhead for associates and made customers more likely to proceed with their projects.

🔐 Safety, privacy, and accuracy considerations

Deploying GPT-5 in a retail environment poses specific safety and privacy considerations. We treated these as non-negotiable from day one:

Data minimization: Only necessary data fields are passed to the model. We avoid sending personally identifiable information unless absolutely required and with explicit consent.
Model grounding: Responses that affect purchasing or installation decisions are grounded in an authoritative database or include citations to product documentation.
Human-in-the-loop: For high-risk decisions—electrical, structural, or safety-critical work—the model recommends involving certified professionals and provides references to contractors or in-store resources.
Monitoring and feedback loops: We monitor outputs for hallucinations and maintain a feedback mechanism so associates can flag incorrect model responses for rapid retraining or prompt adjustments.

When the model should defer

AI is powerful, but it’s not a substitute for professional judgment in safety-critical scenarios. For example, if a customer’s project involves electrical rewiring or structural changes, the model’s role is to provide initial guidance and recommend qualified professionals. In our system, GPT-5 explicitly flags high-risk situations and suggests next steps that include human expertise.

🔧 Real-world examples of GPT-5 in action at Lowe’s

To ground the previous sections, I’ll walk through a few real-world vignettes that illustrate how GPT-5 improves customer and associate experiences.

Case 1: In-store associate assistance

An associate named Maria is helping a customer who wants to replace windows. The customer provides some measurements and a photo of the current window. In the past, Maria would cross-reference a printed guide, check inventory, and possibly call a specialist. With GPT-5, Maria enters the measurements and uploads the photo into a store tool. The model:

Identifies the window type and suggests compatible replacement models based on the measurements.
Checks regional inventory and flags nearby stores if the exact model is out of stock.
Provides installation requirements and recommends whether in-home measurement by a certified installer is necessary.

Result: Maria resolves the question in a single customer visit and schedules a follow-up measurement service if needed.

Case 2: Digital planner for DIY homeowners

A customer begins a kitchen remodel project on Lowe’s digital planning tool. They upload photos, input a rough budget, and describe their style preference. The GPT-5-backed planner:

Analyzes the images for lighting and space constraints.
Suggests a range of materials and finishes that fit the budget and style.
Generates a shopping list with SKU-level recommendations and an estimate for material quantities.
Offers installation options, including estimates and next steps to book a contractor or a store appointment.

Result: The customer leaves with an actionable plan and a clear next step to buy or book services.

Case 3: Internal decision support

A regional operations leader needs a quick synthesis of trends in bathroom fixture returns after a recent marketing campaign. The leader uses an internal GPT-5 tool to summarize returns data, identify common failure modes, and draft a memo recommending changes to product packaging and installation guidance.

Result: The leader receives an actionable summary in hours rather than days, enabling a faster operational response.

📚 Training, prompts, and grounding: How we keep answers reliable

Reliability in a retail context depends on how the model is prompted and what knowledge it can access. We invested heavily in prompt engineering and data grounding so the model’s outputs are useful and trustworthy.

Prompt templates: We developed role-specific prompts for associates, digital planners, and back-office users. Each template sets expectations for the model’s behavior, including when to ask clarifying questions and when to cite sources.
Grounding sources: The model pulls from product catalogs, installation manuals, and inventory systems. Whenever a recommendation involves quantities, compatibility, or safety, the model cites the source data in its response.
Feedback loops: Associates can rate answers and flag incorrect outputs, which feeds back into prompt tuning and data adjustments.

These steps collectively reduce hallucination risk and improve overall quality.

📦 Integrations: Connecting GPT-5 with retail systems

Technical integration is a core enabler of value. GPT-5 is powerful, but without connection to real-time systems (inventory, pricing, scheduling) the model’s value is limited. We built an orchestration layer that mediates access to internal systems and enforces policies on what data the model can use.

Key integration points included:

Real-time inventory and regional availability.
SKU-level product specifications and compatibility matrices.
Service catalogs and scheduling APIs for installation and measurement services.
Customer CRM records (with consent) to personalize interactions.

By combining GPT-5’s reasoning with these real-time systems, we achieved answers that were not just plausible but operationally actionable.

🧑‍🏫 Training and change management for associates

Technology adoption is as much about people as it is about models. To ensure effective use of GPT-5 tools, we invested in training and change management:

Role-based training: Associates learned when to trust the model, when to verify, and how to interpret confidence signals from the system.
Playbooks: We provided playbooks showing example interactions and common edge cases, especially for safety-critical scenarios.
Champion network: Local champions helped other associates adopt the tool, providing real-world tips and collecting feedback for continuous improvements.

These human-centered efforts helped embed GPT-5 tools into daily workflows and increased overall adoption and trust.

💡 Lessons learned and recommendations for other retailers

From our deployment, several lessons emerged that I believe can help other retailers consider and accelerate their own AI journeys:

Start small, measure impact

Begin with narrow, high-value use cases where success metrics are clear. Prove the model’s value in those contexts before expanding to broader functions. Measure time saved, steps reduced, and satisfaction improvements to build the business case.

Ground the model in authoritative sources

Retail requires operational accuracy. Ensure the model can access and cite product specs, inventory, and service rules. When a suggestion involves cost, safety, or compatibility, it must be traceable back to a trusted source.

Keep humans in the loop

AI should augment expertise. For high-risk decisions, make human oversight mandatory. Use the model to prepare options and evidence, but keep final authority with trained professionals.

Home improvement often requires photos, measurements, and spoken descriptions. Build interfaces that accept these inputs and let the model reason across them. This enables richer problem-solving than text-only approaches.

Iterate on prompts and UI together

Model prompts and user interfaces influence each other. Iteratively refine prompts based on real-world usage and adjust the UI to present answers in the most actionable format for associate or customer workflows.

🔮 The future: Where GPT models can go next in retail

GPT-5 is already showing meaningful value, but the path forward is rich with possibilities:

Deeper multi-modal capabilities: Improved image and video understanding will further reduce the need for explicit measurements and help diagnose site conditions more accurately.
Personalized project timelines: Models could generate projected timelines and logistics plans that coordinate product availability, contractor schedules, and delivery windows.
Smarter supply chains: AI could anticipate localized demand based on project trends and automatically adjust distribution plans.
Virtual design assistants: Real-time AR/VR-guided installations or design walkthroughs combining model reasoning with site scans.

These innovations will continue to blur the lines between digital planning and physical execution, enabling customers to move from inspiration to installation with less friction.

🧾 Final thoughts: AI that helps people do complex things

At its core, our work with GPT-5 is about enabling people—customers and associates—to solve complex, real-world problems more easily. Home improvement projects are inherently personal and often technical. By reducing the number of steps required to find the right answer, by holding context longer, and by applying reasoning across multiple inputs, GPT-5 helps deliver clearer, faster, and more confident outcomes.

"It is really around applying reasoning with speed. And that is what we have been able to see so far with GPT-5." — Chandhu Nair

We framed our AI strategy around how we shop, how we sell, and how we work—and GPT-5 runs across all three pillars. As we continue to iterate and scale, our focus remains the same: build reliable, human-centered tools that make home improvement more accessible and successful for everyone.

📝 Appendix: Practical checklist for retailers starting with GPT-5

To leave you with a practical takeaway, here is a checklist of actions I recommend for retailers considering GPT-5 or similar advanced models:

Identify 2–3 high-impact pilot use cases with clear success metrics.
Ensure access to authoritative grounding data (product specs, inventory, service rules).
Design multi-modal input channels (photo uploads, chat, voice) where relevant.
Build an orchestration layer for controlled access to internal systems.
Establish governance for privacy, data minimization, and monitoring of model outputs.
Train associates with role-based playbooks and maintain a champion network for adoption.
Set up monitoring dashboards to track time-to-answer, escalation rates, and customer/associate satisfaction.
Design fail-safe behaviors: clear deferral to humans in safety-critical contexts.
Iterate on prompts and UI based on real-world usage and feedback.
Plan for incremental scale and a governance model that evolves with deployment.

I’m excited about the opportunities ahead. For Lowe’s, GPT-5 is not an experiment—it’s becoming an integral part of how we help customers realize their home improvement projects, how we enable associates to be more effective, and how we run the business more efficiently. As the technology continues to advance, my expectation is that the bar will keep rising, and the possibilities for improving everyday experiences will expand even further.

📣 Acknowledgment

This article is based on a presentation by OpenAI featuring insights from Chandhu Nair, Senior Vice President of Data, AI, and Innovation at Lowe’s. The examples and lessons above reflect Lowe’s implementation experiences and operational considerations as shared in that feature.