Anyone can now make a dashboard with AI.

That’s wonderful.

The person closest to the question can answer it themselves. No more waiting weeks in a queue for someone else to build it.

That’s terrible.

Now there are twenty dashboards and analyses for every question, many on stale data, and nobody knows which one is right.

Keep the wonderful. With Seneca, anyone can pull a trusted dashboard or analysis into their own AI tool and build on it.

Lose the terrible. Every one shows how far you can trust it: where its data came from, how fresh the numbers are, and who reviewed it, if anyone did.

Talk with Jake

Repeat Purchase Dashboard

Customers who bought again, by channel

Repeat rate31.4%

Customers7,912

Search 30.9%
Social 24.0%
Referral 41.2%

Made with Claude · Jun · Data as of 3 Mar

Figure 1

mean repeat_rate = 0.278
Codex · Omar · plot_repeat.py

repeat_rate_v3_FINAL.xlsx

	A	B	C
1	Channel	New cust.	Repeat %
2	Search	3,950	30.2%
3	Social	2,880	20.1%
4	Referral	1,140	37.0%
5	Total	7,970	29.1%

Exported 12 Jan · Noor · Sheet1 (2)

Repeat purchase rate

Refreshed 47 days ago

33%

+4.1% from last period

Cursor · Sam · repeat-v2 (don’t edit)

file:///C:/Users/dan/Downloads/dashboard (2).html

Repeat Purchase

Overall repeat rate: 29.8%

Search	31.0%
Social	22.4%
Referral	37.5%

Download CSV generated 2 Feb

repeat_analysis_old.ipynb

In [14]:df.groupby("channel")["repeat"].mean()

Out[14]:

	repeat
search	0.274
social	0.198
referral	0.351
overall	0.265

Kofi · last run 9 Apr

Hana 11:42 AM

here’s the repeat rate, about 30% overall. pulled it last quarter so might be a bit off

image.png

14 replies · Last reply 6 weeks ago

Q3 Repeat Purchase

Repeat rate up to 32.6%
Social still lagging
Referral best?? (confirm)

DRAFT – Priya – numbers from the Q2 deck12

Terminal

$ python repeat_v2_FINAL.py
reading orders_export_jan.csv
channel     repeat
search      0.291
social      0.214
referral    0.366
TOTAL       0.284
(cached 2026-04-11)

Repeat Purchase Dashboard (copy)

Same as Jun’s, with the dates fixed?

Repeat rate30.7%

Customers8,104

Made with Claude · Ines · Data as of 18 Feb

DO NOT USE – old definition

Repeat rate, partners counted as Referral

35.2%

AI-generated · Ava · repeat_dashboard_v2_FINAL

repeat_purchase_final_final.pdf3 / 9

3. Repeat purchase

Repeat rate fell to 27.1% in Q2, driven by Social. Figures as of 30 June.

Channel retention notes

Tomas · Last edited 9 weeks ago

Repeat rate is roughly 31% (check with Finance before sharing)

Referral strongest, see the v3 sheet
Social numbers look off?
TODO: refresh after Q3 close

FiltersRegion: AllWindow: 60 daysChannel: All

localhost:5173

Repeat Rate Explorer v2_FINAL

29.6%

Codex · Ben · cache from 6 Jun

Twenty versions of one question.
No two agree.

With Seneca, start from what’s already known. Find it. Build on it.

Try it: press “Reviewed · see why” for this dashboard’s proof.

Dashboard · seneca / marketing / repeat-purchase-by-channel · v2

Repeat Purchase by Channel

Share of new customers who bought again within 90 days, by acquisition channel

Region Last 12 months

New customers

—

Repeat rate (90 days)

—

Repeat revenue

—

Best channel

—

Repeat rate by month

90-day repeat rate for customers acquired in each month

Repeat rate by channel

Whole period, selected region

Channel	New customers	Repeat rate	Repeat revenue	vs prior 12 mo

Built with Claude · Rhea Osei, Marketing

Why you can trust this

Repeat purchase by acquisition channel. Rhea Osei, Marketing, with Claude. 7 October 2026.

Source

Orders warehouse, certified by the data team on 2 September 2026.

Data as of

Live, today 06:00. The figures are not stored in the page. Each time it opens, its queries run under the reader's own data access.

Checks

Passing Channel totals add up to the customer count.

Passing No customer is counted in two channels.

A passing check proves only what it declares. It is not a review.

Definitions

Revenueestablished is the company's shared definition.

Repeat-purchase windownew here was created for this question and stays specific to it until someone approves it for everyone.

Open to

Marketing-leads group. Details on the Data tab.

History

3 versions · 2 related analyses

Review, by a named person

Reviewed
The channel assignment and the 90-day window are appropriate for comparing channels. A later correction moved partner sign-ups out of Referral. This analysis does not show why Social customers return less often.L. Marsh, Finance · 9 October 2026

Each space sets its own floor. A sandbox accepts anything and says so in the record; the board space requires an accountable reviewer before anything is shown there.

What feeds each panel

1Headline figures repeat_summary on orders, customers
2Repeat rate by month repeat_by_month on orders, customers
3Repeat rate by channel repeat_by_channel on orders, customers, channels
4Channel table, same query as panel 3

Tables

orders, customers, channels. All three certified; none copied into the page.

Filter

Region, passed to every query as a parameter. The readout under the dashboard shows the rows each run returned.

Who can open

Marketing-leads group.

Two people opening the same dashboard can see different rows, because each query runs under the reader's own data access. Sign-in is through your company's identity provider, so these are the groups you already run.

Everyone is sharing the dashboards they’ve made with AI.
But which ones can you trust?

You've been sent another HTML dashboard over Slack. It's beautiful. It's interactive. Is the data up to date? Did anyone check it? The person who built it may know, but the page doesn't say. In Seneca, it does.

	Sent over Slack	In Seneca
The numbers	Copied into the page on the day it was built. They look fresh, but could be days or months old.	Queried live from your warehouse each time it opens, with the data's time in its record.
The source	Not on the page. It may be a certified table or an export from someone's laptop, and the reader can't tell which.	Named in its record, and marked certified or not. The tables and the query behind every panel are listed.
Who checked it	No one knows.	A reviewer signs by name, with what they checked and what they did not. Each space sets its own floor.A sandbox accepts an unreviewed draft and says so. The board space requires an accountable reviewer.
Where it lives	A chat thread, a shared drive, a link that stops working.	One stable address, opened only by named groups, with every query run under the reader's own data access.Sign-in is through your company's identity provider, such as Okta or Entra, so the groups are the ones you already run.
What changed	A new copy, with no record of why or by whom.	Every version kept, with who changed what and why. If new data changes an approved conclusion, the approved version stays up until someone checks the new one.
The next person	Starts from nothing and asks the same question again. Twenty times a quarter, in twenty threads.	Is shown what already answers it before they start, and builds on it. Anything new on the same question is linked to what came before.
The AI tool	Whichever one that person happened to use. It follows the company's definitions only if they connected it to them.	Any of them. Claude Code, Codex, Cursor, and whatever comes next, through one open tool interface. Each is handed the company's definitions first.

Before anyone builds a new dashboard, they are shown what already exists.

A dashboard is a standing view: the same question, answered live every day. An analysis is one investigation that ends in a conclusion, usually in a chat thread. Seneca holds both, and resolves the sprawl by making existing work the starting point, not by turning new work away.

Dana, in Sales operations, asks Claude why EMEA revenue fell last quarter.

Before Claude writes a line of SQL, it asks Seneca what the company already knows. It gets back two analyses and a dashboard that already answer the question, and the definitions it should use. Dana reads them in ten minutes instead of rebuilding them in a week, and her own work starts where theirs ended.

They reach three different conclusions, and Seneca keeps all three side by side rather than letting the newest become the official one. If Dana files a fourth, it is cross-referenced to the others at the door. Nothing is turned away, and nothing is pretended: an unreviewed draft simply says so in its record.

Illustration with made-up names and entries.

Dana's Claude asks Seneca What already answers this: why did EMEA revenue fall in Q3?

Seneca replies

Two analyses and a dashboard already answer this. Start from one of them rather than from the warehouse.

EMEA Q3 revenue: restatement effectAnalysis · Leo Marsh, Finance, with Codex. 7 Oct. Moves two large contracts into Q2.ReviewedP. Nair, Data team
EMEA Q3: competitor pricing in DE and FRAnalysis · Amara Osei, EMEA sales, with Cursor. 8 Oct. Filed beside the above as a second explanation.Different conclusionkept beside Finance's
EMEA campaign spend, July onwardDashboard · Marketing, with Claude. 9 Oct. Shared with the marketing team only.Not reviewed yetsandbox space

Also handed over: the established definitions of revenue and fiscal quarter, and the Data team's note that Q3 is incomplete until the 15th. Not handed over: any row Dana's own access does not allow.

Beside your semantic layer, not instead of it.

A semantic layer is exact by design. What Seneca keeps has room for different views, and that is deliberate.

Don't have a semantic layer? We can help with that too.

The semantic layer answers "what is revenue?"

One definition, one number, the same for everyone. Seneca references those definitions from the systems that own them and marks them on every dashboard and analysis as established. It never redefines them.

Revenue, net of refunds
Active customer
Fiscal quarter

Seneca answers "what did we learn, and where do views differ?"

Corrections, with who made them. Different explanations of the same result, kept side by side. The note that the current period is incomplete. The calculation one analyst invented that may or may not deserve to become shared. This is what the next person and their AI tool are handed first.

Partner sign-ups moved out of Referral, 9 Oct, L. Marsh
EMEA Q3: three different conclusions, all kept
Repeat-purchase window: new to one dashboard, not yet shared

Early, and building with analytics leaders.

Seneca is early, and we are building it in the open with the people who are living this problem.

We are looking for analytics and business leaders who are already watching AI-built dashboards and analyses multiply in their teams and want one place for them to live, and one place to look before building another. If that sounds like you, we would like to hear what you are seeing and to show you Seneca.