While You Sleep, AI Analyzes All of Your Company's Data — Introducing Laplace MCP

"Where should we cut ad spend this month?" "What should we reorder first before we stock out?" "Should we keep selling this product or drop it?" The answers to these questions are actually already in your data. The problem was that those answers were scattered in pieces across order spreadsheets, ad managers, inventory systems, CS tools and review pages. Starting today, AI puts those pieces together. Laplace MCP is here.

---

1. The real reason commerce data analysis has been so hard

Every seller "looks at" data: SmartStore revenue, Coupang settlements, ad ROAS, inventory counts. But the questions that really matter always sit between one dataset and another.

• Which ad creative led to sales of which product, and which option?

• Once you subtract COGS, fees, ad spend and shipping from this SKU's revenue, how much is actually left?

• Are the products with lots of returns and CS inquiries eating into your margins as well?

• When a category climbs in the Olive Young rankings, does demand for your products rise with it?

To answer questions like these, you have to stitch together at least 3–4 datasets accurately at the product level. Done by hand in Excel, it takes a full day, and by next week the analysis is already stale. Most sellers make decisions on "gut feeling" not because they're lazy, but because the cost of connecting the data has been far too high.

---

2. What Laplace has already laid down — the Ontology, a "relationship map" of your data

Laplace Analytics was built from day one to solve exactly this problem. At its core is the Ontology: a technology that doesn't just gather scattered data but defines the relationships between data points to form a single knowledge graph.

Laplace comes with the following data assets built in.

Two things are decisive here.

① Linking ad creatives ↔ product data. Instead of stopping at "Creative A: 1,000 clicks" from the ad platform report, it drills all the way down to "the revenue and Contribution Margin of Product X, Option Y" generated by Creative A. Ad spend is automatically allocated at the product level.

② Connecting base products ↔ sold product options. Product and option names that differ from channel to channel ("Black/L", "Blk Large", "BK-L") are grouped under a single base product. So you register a cost only once, and it automatically flows into every sale on every channel.

Each Asset is refined through a pipeline of raw data → data processing (COGS/revenue allocation) → custom columns → final output, and you can freely add custom columns such as a target revenue lookup.

Laplace customers are already running dashboards like these on top of it.

• Executive dashboard: payment amount → product sales → shipping fees → refunds → COGS → sales channel fees → ad spend → Contribution Margin and Contribution Margin Rate, by day/week/month and by channel. Combined with targets from Google Sheets, it goes all the way to target attainment, conversion ROAS and projected revenue.

• Inventory dashboard: inbound/outbound trends, opening/closing/average inventory, inventory turnover, days of supply (DOS), average dwell time, days of inventory cushion, and inventory value.

• Product dashboard: revenue ranking by SKU, GMV, coupons/discounts/shipping fees, number of payments and buyers, AOV, ARPPU, COGS and margin, plus marketplace-funded discounts and fees.

---

3. "Our data probably can't be integrated, right?" — RPA brings in all of it

Laplace's data collection doesn't stop at the official integration list.

The standard integrations alone are already extensive: SmartStore, Cafe24, imweb, Godomall 5, MakeShop, Coupang (Wing), Kakao Shopping (Talk Store, Gift), Toss Shopping, Qoo10, Amazon / integrated management solutions such as EZAdmin and Sabangnet / Google Ads, Naver Search Ads, GFA, Facebook, TikTok, Kakao Moment, Kakao Keyword Ads, Criteo / GA4 / MySQL, PostgreSQL, MongoDB, Oracle, MSSQL / spreadsheets / reviews (vreview, CREMA) / CS (Channel Talk).

On top of that, RPA (robotic process automation) brings almost any data you can see on a screen into Laplace.

• Inventory data from fulfillment and order management tools such as EZAdmin and Sabangnet — automatically collects inventory by warehouse and by SKU

• Sales channel data from channels whose APIs are closed or limited, such as Musinsa, 29CM, Gmarket, 11st and Olive Young

• Influencer marketing costs — turns sponsorship and group-buy costs that never show up in ad platform reports into data

• Market and trend signals such as Olive Young real-time rankings, search keyword rankings and traffic data

• Virtually any other data that people used to check by eye in admin screens

In other words, there is virtually nothing in your company's data flow that can't make it into Laplace. And the moment it arrives, the Ontology links it at the product level to your existing order, product and ad data.

---

4. And today, MCP — AI now works with all of it "directly"

MCP (Model Context Protocol) is a standard protocol that lets AI securely access external data. Issue a Laplace MCP key and register it with an AI such as Claude, and the AI can now:

1. Directly query all the connected data stored in Laplace,

2. Combine data by following the relationships defined by the Ontology,

3. Run arithmetic and statistical calculations on the fly on top of it,

4. And deliver answers and recommendations backed by evidence.

You don't need to know anything about Laplace's chart configurations, table structures or measure names. Your "question" becomes the analysis blueprint, and the AI does everything else.

Let us show you, through real-world scenarios, why this is such a big deal.

---

5. What Ontology + AI actually delivers — 12 scenarios

Scenario 1. AI knows in advance "when" inventory will blow up

Combined data: inventory history (EZAdmin/Sabangnet via RPA) × orders × base products

People look at inventory "quantities." AI calculates inventory probabilities.

• It computes the mean and standard deviation (σ) of daily shipments for each SKU,

• Draws a safety line that accounts for demand variability with safety stock = Z × σ (std. dev. of shipments) × √lead time,

• And automatically calculates the reorder point (ROP) = average daily shipments × lead time + safety stock for every SKU.

"At a 95% service level, pull every SKU that will stock out unless we reorder this week. Include recommended order quantities and the reasoning."

With hundreds of SKUs, no person can do this. The AI calculates every SKU at once and tells you to hold thicker safety stock for products with volatile shipments (high σ). Conversely, for products with abnormally long days of supply (DOS), it translates that into "this is how much cash is tied up here."

Scenario 2. The uncomfortable truth: "your No. 1 bestseller is actually losing money"

Combined data: base product orders (COGS) × ads × sales channel fees

Laplace already calculates Contribution Margin by SKU, deducting refunds, COGS, channel fees, allocated ad spend and seller-paid shipping from the payment amount. The AI takes it one step further.

"Among our top 20 products by revenue, find the ones with negative Contribution Margin or below 5%, and break down whether the cause is ad spend, fees or COGS."

In setups like this, you'll find products like "No. 3 in GMV, but -480% Contribution Margin after ad spend allocation." It's a fact you could never discover from a revenue ranking alone. The AI breaks down the cause, too: "This product's cost ratio is normal; the problem is a specific channel's fees plus excessive ad allocation. Cut the ads, and organic sales alone can restore Contribution Margin to 12%."

Scenario 3. Slice thousands of SKUs with ABC analysis — all the way to a discontinuation list

Combined data: base products (categories) × orders × inventory × CS

If you have too many SKUs, register categories such as device type, model, color and size for each product. Then:

• ABC analysis: automatically grades products A (core 20%)/B/C by cumulative contribution to Contribution Margin

• Automatic detection of discontinuation candidates: the intersection of grade C + bottom-tier inventory turnover + top-tier return/CS rates

• Star product detection: products ranking high on sales growth, Contribution Margin Rate and Repurchase Rate all at once

"Which colors are selling best? Tell me which SKUs have size inventory allocations that don't match their sales mix."

Problems like "black keeps selling out while beige has been sitting in the warehouse for 6 months" are uncovered across your entire catalog at once along the color and size category axes, complete with calculated order ratios for next season.

Scenario 4. CS contribution rate — the "hidden cost" revealed by customer inquiries

Combined data: Channel Talk CS × orders × base products × reviews

Attach CS inquiries at the product level and a new metric emerges: CS contribution rate (what % of all inquiries this product generates).

"Find products with an abnormally high number of CS inquiries relative to sales volume, and split them by inquiry type (shipping/quality/size). Also show how much the real margin shrinks if we assume CS handling labor costs KRW 3,000 per inquiry."

If a product that generates 5% of revenue is generating 30% of CS, its "true margin" is different from what the books say. Products with concentrated size inquiries call for a better size chart on the product page; those with concentrated quality inquiries call for stricter inspection or discontinuation. The inquiry type itself becomes the action.

Scenario 5. Catch the moment reviews start eating into revenue

Combined data: reviews (CREMA/vreview) × orders × returns × repurchases

"Find products whose average rating over the last 30 days dropped by 0.3 points or more compared with the previous 90 days, and show the changes in their return rate and Repurchase Rate alongside."

More low-rated reviews → higher return rates → lower repurchase rates translates into falling revenue with a lag of several weeks. When the Ontology ties reviews, orders and returns together, AI catches this chain reaction before revenue turns down. It even extracts recurring keywords from review text ("pilling," "runs small") to point out what to improve.

Scenario 6. Influencer marketing, measured by Contribution Margin rather than "vibes"

Combined data: influencer costs (RPA) × orders × base product orders (COGS)

Sponsorship and group-buy costs don't show up in ad platform reports, so the conversation always ended with "so, did we actually make money?" Once influencer costs flow in as data via RPA:

"For each influencer campaign last quarter, calculate how much we really made: the incremental revenue of the featured products during the campaign minus COGS, fees and sponsorship costs. Rank influencers by Contribution Margin, not follower count."

Because it calculates incremental revenue over the baseline (normal sales) and deducts COGS as well, it separates campaigns that "got great views but actually lost money" from influencers who "quietly made us money."

Scenario 7. Olive Young rankings and search keyword rankings = leading indicators of demand

Combined data: Olive Young real-time rankings and search keyword rankings (RPA) × your own sales × inventory

The moment market signals sit on the same chart as your sales data, they become demand forecasts.

"Show me whether our sales also rose in weeks when search keyword rankings for our category climbed on Olive Young. And tell me whether current inventory for products tied to today's trending keywords can withstand the expected demand."

How our traffic changed when a competitor entered the rankings, or how many days after a keyword spike our revenue responds (lag analysis): this is analysis that's simply impossible if someone is screenshotting ranking pages every day.

Scenario 8. Traffic is coming in, so why aren't we selling? — funnel diagnosis

Combined data: GA4/traffic data × ads × orders

"Calculate the purchase conversion rate relative to traffic for each channel, and find the channel–product combinations that rank high in traffic but low in conversion. Also estimate how much Contribution Margin would increase if we moved that ad spend to where conversions are strong."

Since traffic data and orders are linked at the product level, you can tell whether "the ads failed to bring people in" or "they brought people in but the product page lost them." The former is a creative problem, the latter a page problem, and the prescriptions are different.

Scenario 9. Products that sell together, and products that drive repeat purchases

Combined data: orders (customer level) × base products

"Compare the 90-day Repurchase Rate by first-purchase product. Which 'gateway product' is best at driving repeat purchases? If we built bundles from the top 10 products frequently bought together, how much would AOV go up?"

With customers, orders and products connected, Cohort Repurchase Rate and basket co-purchase (cross-selling) analysis take a single question. "Gateway products" are worth putting more ad budget behind, and co-purchase combinations are ready-made ideas for themed promotions.

Scenario 10. Are coupons and discounts medicine or poison? — price elasticity

Combined data: orders (coupons/discounts/marketplace-funded) × base product orders (COGS)

Laplace tracks coupons, discounts, marketplace-funded discounts and coupons, and fees separately. So:

"Compare sales volume and Contribution Margin between weeks with and without discounts. List the products whose total Contribution Margin actually fell despite the discount (low-elasticity products)."

It tests the misconception that "discounts always mean more sales" with data. And because it distinguishes discounts funded by the marketplace from those you fund yourself, it even reveals that the same promotion leaves you with different margins on different channels.

Scenario 11. The gap between targets and reality — plus what-if analysis

Combined data: Google Sheets targets × orders × ads

"Show this month's target attainment by channel and calculate the month-end projection based on the remaining business days. How would projected revenue and Contribution Margin change if we cut ad spend by 20%?"

With target revenue (from spreadsheets) linked to actual results, attainment and pacing are a given, and it even runs simulations based on the historical relationship between ad spend and revenue. You get answers you can use for management decisions right away, such as "revenue drops 4%, but Contribution Margin rises 9%."

Scenario 12. What humans can't see — all-around anomaly detection

Combined data: everything

This is the most underrated capability. People only look at the charts they've decided to look at. AI can sweep through every combination of all your data.

"Find everything in last week's data that deviates from the usual patterns."

• A specific channel's fee rate quietly creeping up and draining margin

• Traffic suddenly drying up for just one bestselling option (timed with a competitor entering the rankings)

• Ad spend allocation concentrating on one creative even as that creative's conversions decline

• Returns piling up only after a specific point in the logistics process

These are things you "never even thought to look for" in the first place. Because the Ontology has already connected all the data, AI can ask questions on its own that no human designed.

---

6. Why people + Excel can't do it, but Ontology + AI can

Here's the bottom line. Until now, the bottleneck in data analysis wasn't "analytical skill" but "the labor of connecting data." Laplace eliminated that labor with the Ontology, and MCP puts a tireless analyst on top of it.

---

7. Connecting takes just 3 minutes

1. In your Laplace project, go to Settings → API Integration → Issue MCP Key

2. Choose an expiration policy (30 days / 90 days / 1 year / no expiration — renewing every 90 days is recommended)

3. Register the issued key with your AI client — Claude (web and desktop), Cursor and VS Code are supported

Because authentication is key-based, you control the scope of access and expiration yourself, and keys can be revoked instantly if needed. Your data opens up to AI only under your control.

---

8. Closing thoughts — the size of your questions is the size of your analysis

Until now, you could only use your data "as far as you knew how to analyze it." Now you can use it "as far as your curiosity goes."

• Inventory, not as quantities but as stockout probability and tied-up cash;

• Revenue, not as rankings but as Contribution Margin;

• Marketing, not as view counts but as incremental profit;

• The market, not by gut feeling but through the correlation between ranking and keyword signals and your own sales.

Connect your data to Laplace, issue an MCP key, and ask the AI your first question.

"Find the one thing our company needs to fix first, right now."

Chances are, its answer will be more convincing than this article.