Steve verifies that customers actually did the things your loyalty program is rewarding them for. Photos, receipts, before-and-afters. AI scores the proof. Open Loyalty awards the points. No humans in the middle, unless you want them there.
A photo, a receipt, a before and an after. Through your app or ours. Metadata travels with it.
Your workflow runs. AI extracts the fields you defined. Each one gets a confidence score.
Above your threshold, it auto-approves. Below it, a human reviews it in seconds with Steve's reasoning attached.
The approved submission fires as a custom event in Open Loyalty. Your campaign awards the points.
Loyalty programs are great at rewarding what people buy.
They're terrible at rewarding what people do.
For thirty years, loyalty has been brilliant at rewarding transactions. Scan, buy, get points. The mechanics are mature.
Rewarding the actions around the transaction is a different story. Cleaning a roof. Keeping a brand display up. Changing tires before they fail. Scanning a receipt from a partner. These behaviors move the business, but verifying them at scale has stayed expensive and slow.
Steve closes that gap. The same vision models that recognize brands and read receipts now do verification at submission speed. The job left for product is turning imperfect AI scores into reward logic users trust.
Same machine, different proof. Pick the one that maps to a behavior your program already wants to reward.
Retailers and distributors photograph the shelf. Steve checks whether your brand is present, where it sits, how much space it owns, and how much of the shelf belongs to competitors. Self-submission replaces field reps. Scales infinitely.
"Brand is present and sitting at eye level. Twelve facings across the planogram. Share of shelf reads 0.35 — below the contract minimum of 0.40."— Steve, on a real shelf submission
The customer submits two photos. Steve compares them, confirms the action happened, and produces a structured analysis with damage detection. Insurance loyalty has been weak because there is nothing offline to reward. Steve unlocks an entire category of preventive-behavior mechanics.
"No visible structural damage. Shingles intact. Drain clear. Photo metadata consistent with submission timestamp."— Steve, on a real roof submission
The customer uploads one to three documents. Steve extracts structured fields with confidence scores. Required fields are configurable. Manual receipt processing is the most common reason offline-earning programs fail or stay small. Steve makes them operationally cheap.
"Invoice DE-449182 parsed cleanly. Total 14.21 EUR. No duplicate hash in the last ninety days. Issue date falls inside the promotion window."— Steve, on a real receipt submission
Field user takes a photo of a retail shelf. An on-device model and a local product catalog identify which products sit where, scoring shelf compliance. Works in stores with no signal.
"Seventeen products detected. Six brand A facings. Two unknown SKUs flagged for human review. Pipeline ran fully on-device in under half a second."— Steve, on a real in-store frame
Steve's name is a clue. Steve sounds like sieve. Every photo is a claim about something that happened in the real world. Steve's job is to filter the real from the gamed.
You set the auto-approve bar per workflow. High-confidence submissions fire instantly. Anything below the line lands in the review queue with Steve's reasoning attached, so a human decides in seconds.
Capture timestamp, GPS, device, whether the photo was modified after capture. Every signal travels with the submission and shows up in the inspector. Stale photos, mismatched coordinates, and edited files are flagged before AI even runs.
Perceptual hashing catches the same photo (or a near-identical crop) being submitted twice for two rewards. The flag carries the original submission ID so reviewers can see the match and the original campaign.
The review queue inherits everything Steve already extracted: the structured fields, the confidence scores, the flagged signals, the reasoning. Reviewers approve or reject. The queue is a tool for confidence, not a bottleneck.
Every approved submission lands in Open Loyalty as a custom event with the structured fields attached. Build campaigns on top. Award points by score, not by hope. Scale the reward with the confidence, or set tiered thresholds that fire different rules.
Steve is not a separate product you have to wire up. The integration is the product.
Steve is built for any program that wants to reward offline behavior at scale. The mechanic is the same. The proof is the variable.
AI Labs is our co-development program for enterprise loyalty teams. You bring the challenge. We bring loyalty data, domain experts, and AI engineering. If it works, it becomes a portfolio product, and you keep founding-client status with roadmap influence and direct access to the team.
Tell us what you're trying to solve. We'll tell you whether Steve can do it, or what would need to change so he can.