Olympusby Old Bones
Get started

Pantheon at work

Real questions, asked in Claude with Olympus connected. Ask in plain words and your AI reads the store's own numbers before it answers. Ask for the Council and Hermes seats the specialists a question touches: each answers in their own lane, and it ends with one decision, who owns it and when to look again.

These are real answers from a real store's data. The store is renamed Northwind Outfitters and its products renamed, and its totals are scaled; every rate, ratio and percentage is as it was.

In Claude, with Olympus connectedNorthwind Outfitters

Our cost per new customer went up this month: is it the ads, the site or the products?

Used Olympus · 13 tools
  1. A metric's values
  2. Breaking down the customer acquisition cost calculation.
  3. List the metrics
  4. Pulling monthly ad spend, customers, sessions and conversion figures.
  5. A metric's values
  6. Comparing monthly customer counts from spend and revenue.
  7. A metric's values
  8. Now the site side: traffic and conversion, September vs October.
  9. Landing page funnels
  10. Tracing where the checkout drop comes from
  11. Campaigns over a window
  12. Checking how Meta cost-per-acquisition changed across campaigns
  13. Product funnels
  14. Tallying new-visitor drops across the top products and daily views
  15. A metric's values
  16. A metric's values
  17. A metric's values
  18. Checking whether ad frequency is easing.
  19. A metric's values
  20. Comparing Google spend cuts with Meta's effect on brand search
  21. The store's timeline
  22. Reviewing price changes on the shell products.
  23. A price change, judged
  24. Weighing how much of the cost rise comes from traffic versus conversion

Short version: mostly ads, then products. The site is fine.

nCAC has been climbing for two months: $41 in August, $47 in September, $51 so far in October (target $42, critical $56). In October spend per day fell 21%, but new customers per day fell 27%, so each one cost more.

Ads: the main cause, and it started first. September's jump happened before any price changes, and it's creative fatigue on Meta. Spend on fresh creative fell from 45% in early September to 2 to 6% every week since September 28. Your lead ad's lifetime frequency has crept from 4.3 to 4.9, critical since mid-September. Meta's 7-day-click CPA went from $72 to $87 to $99 through September, then $102 and $117 the first two weeks of October. Every Meta campaign's CPA is up, not just one.

Google is smaller and mostly your own doing. You cut it about 41%, mainly by halving non-branded Shopping. Branded search revenue is down about 29%, meaning fewer people are coming in looking for you. That's what you'd expect once the top of the funnel goes stale. Two caveats: Google credits conversions late, so October's Google CPAs will improve some, and the Brand Reach $240/day test is underwater on ROAS but is meant to be judged on brand searches on Nov 6.

Products: the October piece. On Oct 3 and 4 you raised prices $10 on the Rain Shell, Vents, Hoods, Expedition, and the Trail Boots, about 17 to 25%. As a group the price rise held up on margin: CM1 per day was +10% against unchanged products, within noise. But orders fell 14% against those products. Fewer first orders per ad dollar is exactly what raises nCAC, even when margin holds.

Product by product:

  • Expedition took it worst: orders down 26% against control, new-visitor buyers per view down 22%.
  • Vents actually paid on margin, though new visitors bought less there too.
  • The Rain Shell went back to $40 on Oct 23, and units are up about 20% against control so far. That's only 6 days of data, so it's too early to call.

Site: not the problem. Cart-to-order is about 52% in both months, and revenue per visit is flat ($2.60 vs $2.62). Conversion slipped from 3.30% to 3.02%, but the drop is at add-to-cart, on the products that got pricier, and across every channel including organic and email. That points to price, not the theme. Checkout completion looks worse (72% to 62%), but cart-to-checkout jumped from 72% to 84% at the same time. That looks like a change in what counts as reaching checkout, not lost buyers.

What I'd do:

  1. Get new Meta creative live. This is the biggest lever, and the cause that predates everything else.
  2. Decide what the price rises are for. If you're optimizing margin, keep them and judge nCAC against the higher first order, not the old $42 target. If new-customer volume matters more right now, Expedition is the first one I'd revisit.
  3. Leave the site alone. Nothing broke there.

In Claude, with Olympus connectedNorthwind Outfitters

Should we run a Black Friday sale this year, and how deep? Ask the Council.

Used Olympus · 9 tools · Convene the Council: Plutus, Ares, Aphrodite, Athena, Hephaestus, Demeter
  1. Choosing which agents to include.
  2. Convening the Council
  3. Break-even and product costs
  4. Calculating the discount multiplier across several rates.
  5. A price change, judged
  6. Checking last year's Black Friday timeline.
  7. The store's timeline
  8. Comparing Black Friday week with the prior year's.
  9. Data over a date range
  10. Checking that last year's tier totals reconcile with the sale-discount figure
  11. Data over a date range
  12. Data over a date range
  13. Checking how discounts changed the holiday profit
  14. Data over a date range
  15. Comparing Meta and Google spend for the holiday window.
  16. A metric's values
  17. Deciding whether to repeat the tiered offer

The Council on Black Friday

Scoreboard: CM2 $48,721 over the last 4 weeks, down from $57,694. nCAC $51.29 on 829 new customers (target $42, was 1,151). LTV:nCAC 2.1 against a 2.4 target. Margin per order is up and volume is down, so both guardrails are already in the red going into the season.

Plutus (CFO): Margin on product sales is 86.5%, so a discount d needs this many times the orders to break even: 15% needs 1.21x, 20% needs 1.30x, 25% needs 1.41x, 30% needs 1.53x. Last year's BFCM window (Nov 21 to Dec 2) got orders up 21% over the 12 days before. That covers a 15% sitewide and nothing deeper. What actually worked last year was the ladder (15% at $100, 20% at $150, 25% at $200): only 180 of 703 orders hit a tier, it cost $5,854, and gross profit rose 35% ($38.3k to $51.9k). CM2 went from about $8.9k a week to $16.2k in Black Friday week and $11.1k in Cyber week. The loser was BOGO-Classic-Trails: 47 orders, $40 average order, 50% off. At this margin, 50% off needs 2.4x the orders. Kill it.

Ares (Paid): Last year's sale didn't buy new customers. New customers went 416 to 431 (flat), while the dedicated BFCM 2025 Prospecting campaign paid $24 CPMs and a $56 CPA, against $12 and $42 on evergreen prospecting. With nCAC already $51 against a $42 target, I don't want a separate BF prospecting push. The one BFCM campaign that earned its keep was Retention, at a $34 CPA. BF creative should go there.

Aphrodite (Email): The list is tuning out. Email sales are down to $13,105 from $19,000, and clicks fell to 0.3%. A week of daily blasts to everyone will burn it further. Send to the engaged segment, and lead with one simple, loud offer. I'd rather have a flat sitewide number than a ladder people have to work out.

Athena (Customers & Product): Black Friday here is a returning-customer event. Last year returning orders rose 68% (161 to 271) while new customers stayed flat. Already 56% of orders use a discount, so depth trains people to wait. The October price rises were judged "no clear change" together (+10% against control, inside the ±26% band), and the Alpine Shell with Vents alone "pays." A 20% tier on Vents at $60 lands at $48, under its old $50. So the ladder already hands back the whole rise to anyone who reaches the tier. Refunded orders also doubled during last year's sale (12 to 23), so watch returns. The $150 free gift is still running, and it would stack on the 20% tier.

Hephaestus (Site): Last year's sale moved conversion, not traffic. Sessions were flat (16.4k to 16.2k), while add-to-cart went from 5.9% to 9.0% and orders per visit from 2.5% to 3.3%. Checkout completion dipped from 70.1% to 68.6%. Today add-to-cart is 5.7% (from 6.3%) after the October price rises. The ladder only works if shoppers can see how far they are from the next tier, so the cart needs to show that.

Demeter (Search): Google search sales are $25.6k (from $28.0k), and the average position is steady at 8. Two reads land right around the sale. The Brand Reach Meta test is judged on brand searches through Nov 6, so read it before Black Friday traffic muddies brand demand. The Rain Shell SEO title change gets judged around Nov 20, so no Black Friday title rewrites on those pages until then.

The clash: Aphrodite wants a flat sitewide offer the list will notice. Plutus's rule says a sitewide 25% needs 1.41x the orders, and last year's sale delivered 1.21x, 20 points short. Athena sides with Plutus on depth because of discount habituation. Ares and Plutus agree on ads, and that agreement matters: the guardrails already say don't buy more new customers at today's nCAC, and last year's data says Black Friday wouldn't buy them anyway.

Decision: Yes, run it, but rerun last year's ladder rather than going sitewide: 15% at $100, 20% at $150, 25% at $200. Nothing sitewide, no BOGO, nothing deeper than 25%. Pause the $150 free gift while the ladder runs so that tier isn't paid twice. Don't launch a separate BF prospecting campaign: hold the prospecting budget and run BF creative to retention audiences. Email goes to the engaged segment only, with one clear message: "the more you get, the more you save."

Who owns it: You set the offer and dates. Whoever runs the ads keeps prospecting flat and builds the retention campaign. Whoever sends email builds the engaged-only BF sends.

Look again: Monday Dec 7. It worked if CM2 for Nov 23 to Dec 6 beats last year's same two weeks ($27,331) while nCAC stays under the $68.70 first-order line.

One gap: I didn't pull whether last year's BFCM buyers came back afterward. If you want the Council to weigh in on that, it's a cohort read worth doing before Dec 7.

In Claude, with Olympus connectedNorthwind Outfitters

Convene the whole Council on the business: what should we do differently this week?

Used Olympus · 10 tools · Convene the Council: every specialist
  1. Convening the Council
  2. Coordinating specialist agents to pull the data.
  3. Break-even and product costs
  4. A price change, judged
  5. Campaigns over a window
  6. Reviewing Google shopping bid change and ad spend impact.
  7. Ad account changes
  8. Ad account changes
  9. Reviewing non-branded shopping spend and rank changes.
  10. Email results by flow and campaign
  11. Checking Shopify revenue attribution across campaigns.
  12. Landing page funnels
  13. Comparing the two checkout definitions against each other
  14. Every Google search over a window
  15. Checking non-brand search opportunities near page one
  16. Google searches over a window
  17. Settling the Brand Reach test and Shopping budget shift

Scoreboard: CM2 was $48.7k over the last four weeks (Oct 2 to 29), down from $57.7k, a 16% drop. nCAC rose to $51.29 from $46.76 against a $42 target, on 829 new customers (was 1,151). LTV:nCAC is 2.1 against a 2.4 target. With two guardrails red, nothing this week adds spend overall.

Plutus, CFO: Margin is healthy. Margin after product costs is 78.3%, MER is 2.81 against a 1.33 break-even, and nCAC is still under the $68.70 a new customer can cost on the first order. The drop is volume, not margin: spend fell $11.3k and CM2 fell $9k with it. The October price test, read across all seven products together, comes out "no clear change": CM1 a day is +10% against control, inside its ±26% band. One campaign fails my rule outright. Brand Reach on Meta returned 1.11 against its 1.34 break-even, leaving it $823 short after ads.

Ares, Paid: Most of the spend drop is Google, down $8.9k. The biggest piece is Non-Branded Shopping, which went from $9.8k to $5.2k. Its impression share is 33%, and the 67% it loses is lost to rank, not budget. It still returned 1.71 against a 1.32 break-even and kept $1,524 after ads. Its target ROAS dropped from 2.2 to 2.0 on Oct 20, too recently to read, and Google is still crediting conversions for this window, so that ROAS may rise. Meanwhile Brand Reach went from $72 to $240/day on Oct 16 and has Meta's worst platform CPA at $83, against a $72 target.

Aphrodite, Email: Email sales fell 31% to $13.1k, by Shopify's count. Flows made $8.8k of that, $5.3k from Welcome alone. Campaigns go to about 22,000 people roughly twice a week and get 0.2% to 0.5% unique clicks; the 20%-off Trail Day Bundle got 0.21%. That's a list tuning out, a month before Black Friday, when deliverability matters most. My change: weekly campaigns go to Engaged Customers only, and the full list is saved for BFCM. No new codes: at your 86.5% product margin, a 20% discount needs 1.3x the orders just to break even.

Athena, Customers & Product: The price test says the Alpine Shell with Vents at $60 pays. Its CM1 a day went from $401 to $455, +47% against control, just beyond its ±44% band. Keep it there. Hoods, Expedition and both Trail Boots read "no clear change," meaning the same margin dollars from fewer units, so there's no reason to roll them back. The Rain Shell's return to $40 is too early to read after 6 days. Returning customers rose to 35.5% of orders. Refunds rose to 2.4% from 1.9%. I can't see which products drove that, but the two Oct 1 launches are where I'd look first.

Hephaestus, Site: The site isn't the problem. Revenue per visit is $2.58, against $2.61 before. Visits fell 18%, from 1,673 to 1,379 a day, and paid search visits fell 36%. Higher prices and slightly lower conversion net out flat per visit, and checkout completion held at 74.4%. Nothing to change on my side.

Demeter, Search: The Brand Reach test is judged on brand searches, and those are down. Impressions for "Northwind Outfitters" fell from 1,730 to 1,381 in a window that includes 12 days of the $240 budget. Organic sales slipped 9% to $25.6k. There are cheap ranking wins:

  • "packable rain jacket" fell off page one (9.2 to 11.2), and its impressions collapsed from 2,411 to 122.
  • "waterproof hiking pants" sits at position 11.4 with 259 impressions.
  • "merino hiking socks" slipped to 10.5.

The rain jacket page gets looked at first.

The clash: Ares, Demeter and Plutus against finishing the Brand Reach test. You logged it as three weeks ending Nov 6, judged on brand searches. Two weeks in, your main brand term is down 20%, and the campaign is $823 below break-even. Finishing it means another week at $168/day over the old budget, buying customers that push nCAC further from $42. The only argument for running to Nov 6 is that Meta's last seven days are still settling. On prices, everyone agrees: keep them, Vents especially.

Decision: Cut Brand Reach back to $72/day today and end the test a week early. Lower Non-Branded Shopping's target ROAS from 2.0 to 1.8, still above its 1.32 break-even, so it can win back the impressions it's losing to rank. Total spend stays about flat.
Owner: whoever runs the ads (your agency made the last changes).
Look again: Friday, Nov 13. It worked if nCAC for Nov 6 to 12 is below $51.29, new customers beat this window's ~207 a week, and Shopping's impression share is above 33%.

Ask about your own store.

Pantheon adds Hermes and the six specialists to Olympus, in Claude, ChatGPT or any AI, from $249 a month with a 14-day free trial. See pricing, or everything Olympus does.

Get started free