Overview
Summit Outdoor, an omnichannel outdoor gear retailer with 45 stores and a growing e-commerce business, was losing revenue to a personalization system that had no memory. By deploying Persistent Memory (Episodic Memory) and Dynamic Decision Router (Blackboard), Summit transformed stateless product recommendations into context-rich, relationship-aware interactions — increasing recommendation click-through by 52% and repeat purchase rate by 23%.
The Challenge
Summit Outdoor sells technical outdoor gear — trail running shoes, backcountry ski equipment, climbing hardware, and camping systems. Their customers are not casual browsers. They are athletes and enthusiasts with specific goals, body types, terrain preferences, and seasonal training plans. A customer preparing for a first ultramarathon has fundamentally different needs than a customer replacing a worn-out pair of daily trail runners, even if both search for "trail running shoes."
Summit's existing recommendation engine did not know the difference. Built on collaborative filtering, it treated every session as a blank slate. A customer who had spent 20 minutes in a live chat discussing half-marathon nutrition strategies with a support agent would, 10 minutes later, see a homepage carousel of best-selling tents. The system had no memory of the conversation. It had no memory of anything.
The problem extended beyond recommendations. Summit operated three disconnected AI-adjacent tools from three different vendors: a chatbot for customer support, a product recommendation engine, and an email personalization platform. Each maintained its own silo of customer data. The chatbot knew a customer had asked about waterproof ratings for a specific jacket. The recommendation engine did not. The email platform might send that same customer a promotion for a jacket they had already purchased — or worse, one they had returned.
Customer support suffered a parallel failure. Summit's support queue used keyword-based routing, and 28% of tickets landed in the wrong department. A customer writing "my order hasn't arrived and I need it for my race this weekend" would get routed to general shipping instead of the priority fulfillment team. The urgency — "race this weekend" — was invisible to a keyword matcher looking for shipping-related terms.
The business impact was measurable. Recommendation click-through rate sat at 8.3%, well below the 12-15% industry benchmark for specialty retail. Repeat purchase rate had plateaued at 31% despite a loyalty program redesign. Customer satisfaction scores for support interactions averaged 3.4 out of 5. Summit's Chief Digital Officer, Priya Kapoor, summarized the problem in a quarterly review: "We have 380,000 customers and we treat every single one of them like a stranger."
The Solution
Persistent Memory (Episodic Memory)
Persistent Memory gives the AI system what Summit's previous tools lacked: the ability to remember. Not just purchase history — any system can store transactions — but the full arc of a customer's relationship with the brand.
Summit's implementation builds an episodic memory profile for each customer. Every meaningful interaction becomes a stored episode: the live chat where a customer mentioned training for a half marathon in October, the browse session where they spent 14 minutes comparing two hydration vests, the support ticket about sizing on a previous shoe purchase, the in-store consultation where a sales associate noted they overpronate. Each episode carries structured metadata — the topic, the customer's expressed intent, any products discussed, and the emotional tone of the interaction.
When a customer returns — whether to the website, the app, or a physical store — the system retrieves relevant episodes and constructs a contextual profile. The customer training for a half marathon does not see a generic homepage. They see a curated view: nutrition products relevant to their training timeline, the hydration vest they compared but did not buy, and shoes that accommodate overpronation. The system remembers because it was architecturally designed to.
Memory also enables what Summit calls "relationship continuity." If a customer contacts support about a shoe that developed a tear after 200 miles, the system recalls the original purchase, the customer's running volume (inferred from previous conversations), and their stated terrain preference. The support agent sees all of this before typing a single response. The customer does not have to re-explain their history.
Dynamic Decision Router (Blackboard)
Persistent Memory provides the what — the accumulated knowledge about each customer. Dynamic Decision Router determines the how — which system, channel, or action should respond to a given customer signal at a given moment.
The Blackboard architecture maintains a shared state object for each active customer interaction. Multiple AI components — the recommendation engine, the support classifier, the email personalization system, and a new proactive outreach module — all read from and write to this shared state. Instead of three disconnected vendors producing three conflicting customer experiences, a single shared context drives every touchpoint.
When a customer who has been browsing trail shoes adds a pair to their cart and then opens a support chat, the Dynamic Decision Router reads the shared state and makes several simultaneous decisions: route the chat to a footwear specialist (not general support), pre-load the agent's view with the customer's foot measurement history and past shoe purchases, and suppress the abandoned-cart email sequence that would otherwise fire in 30 minutes (because the customer is actively engaged, not abandoning).
This routing intelligence replaced Summit's keyword-based support classification. The 28% misrouting rate existed because keywords cannot capture intent. The Blackboard architecture captures the full interaction context — what the customer is doing right now, what they have done recently, and what they have told the brand about their goals — and routes accordingly.
The two architectures compose naturally. Persistent Memory accumulates long-term customer knowledge. Dynamic Decision Router applies that knowledge in real time across every channel. Memory without routing would produce good recommendations but still misroute support tickets. Routing without memory would classify tickets accurately but still recommend tents to marathon runners. Together, they produce an experience that feels coherent.
The Results
Summit Outdoor deployed both architectures in a phased rollout: Persistent Memory first for the e-commerce site and support system, followed by Dynamic Decision Router four weeks later to unify the cross-channel experience. Results stabilized after the second month:
- Recommendation click-through rate increased 52%, from 8.3% to 12.6%, placing Summit above the specialty retail benchmark for the first time.
- Repeat purchase rate rose 23%, from 31% to 38.1%, with the strongest gains among customers who had at least three prior interactions stored in episodic memory.
- First-contact resolution for support improved 41%, as agents received full customer context before responding, eliminating the "can you tell me more about your issue" back-and-forth.
- Customer satisfaction for support interactions increased 29%, from 3.4 to 4.4 out of 5.
- Support misrouting dropped from 28% to 5%, as the Dynamic Decision Router replaced keyword-based ticket classification.
- Time to measurable ROI: 9 weeks from initial deployment to the first month showing statistically significant improvement across all four primary metrics.
"We got an email from a customer that said, 'It's like you know me.' That used to be a privacy complaint. Now it's a compliment. The system remembered she was training for a half marathon, and when she came back three weeks later, everything she saw was relevant to that goal. She bought shoes, a vest, and a nutrition kit in one session. That is what memory does — it turns browsing into buying." — Priya Kapoor, Chief Digital Officer, Summit Outdoor
Key Takeaways
- Stateless personalization is a contradiction. Recommendations without memory are just popularity contests. Summit's previous system surfaced best-sellers, not best-fits. Adding episodic memory turned a generic engine into one that understood individual customers.
- Channel unification requires shared state, not shared vendors. Summit tried to solve fragmentation by consolidating vendors. It did not work because the problem was architectural, not contractual. Dynamic Decision Router's shared Blackboard gives every system access to the same customer context regardless of which vendor built it.
- Memory compounds over time. The 52% click-through improvement was strongest among customers with three or more stored episodes. Each interaction makes the next one more relevant — a flywheel that deepens the customer relationship.
- Misrouting is a personalization failure, not just a support failure. Sending a customer to the wrong support queue is the same structural problem as showing them the wrong product recommendation. Both stem from not knowing who the customer is and what they need right now.
Ready to Explore AI Memory for Your Retail Business?
If your customers feel like strangers every time they return, the problem is not your product catalog or your support team — it is your system's inability to remember. Summit Outdoor's experience shows that Persistent Memory and Dynamic Decision Router can transform fragmented interactions into coherent relationships. Talk to our team about bringing AI memory and intelligent routing to your retail operations.