Reporting · replacement recipe
Go from Databox to Small Business plugin (Anthropic knowledge-work-plugins)
The job: A few recurring KPI reports (Authorized data imports, Metric definitions, Charts, Scheduled digest). Pick where you are today and where you want to land; the move below follows your pick.
From · what you pay for now
Databox
Databox Analyst · $71/mo
Year one+$502modeled
To · your pickBest fit
Small Business plugin (Anthropic knowledge-work-plugins)
Agent skill · $0 a month + 0.5 h upkeep a month
Every way to do this job · pick one
Switch app
Open source
Build it with an agent
Agent skill
The money
$71/mo today , then $0/mo
Databox Analyst
Year one, net+$502
Pays back in1.1 months
Modeled from public list prices, with your time at $50 an hour. Show the math
Does it do the job? 3 of 4 covered
- Covered: Authorized data imports
- Covered: Metric definitions
- Not covered: Charts
- Covered: Scheduled digest
The work
- Setup
- About 1 h
- Upkeep
- 0.5 h a month
- Watch out
- Doesn't cover: Charts
Make the move in four steps
Export from Databox
Datasets (data synced from your data sources). Format: CSV only
Export stepsSet up Small Business plugin (Anthropic knowledge-work-plugins)
Three small-team seats: a sales CRM (crm-autopilot builds and keeps a four-tab spreadsheet CRM when there is none)…
DetailsSwitch links and integrations
Point links, embeds and integrations at Small Business plugin (Anthropic knowledge-work-plugins), then send a test through every one.
Cancel Databox
Turn off renewal before your renewal date, once the new setup works. Access runs until: The end of the billing period already paid for. After that….
How to cancel Databox
Modeled, not measured. Savings use public prices and planning
assumptions: the bill you stop paying, minus the new tool, your upkeep time and your setup time at $50/hour, over the first
12 months. Paths marked Tested by Stackcut were run end to end by us on our own products. They're our tests, not customer results.
Check the must-haves in each recipe before you switch.
The whole recipe: 7 ways out, best first
What you use
- Authorized data imports
- Metric definitions
- Charts
- Scheduled digest
Not covered
- Hundreds of connectors
- General semantic layer
- Arbitrary report builder
- Attribution parity
Check before switching
- Source API access already exists
- Metric definitions approved
- Timezones and deduplication tested
- Connector upkeep included
Agents: get_recipe("kpi-reports") on the Stackcut MCP server.
7 ways out, best first
How we got this
| You pay now: Databox Analyst | flat price | $71/mo |
|---|---|---|
| New cost: Small Business plugin (Anthropic knowledge-work-plugins) | runs in the agent you already pay for | $0/mo |
| Your upkeep time | 0.5 h/mo × $50/h | $25/mo |
| Difference each month | $71 − $0 − $25 | +$46/mo |
| Over 12 months | +$46 × 12 | +$552 |
| Setup, once | 1 h × $50/h | −$50 |
| Year one | +$552 − $50 | +$502 |
- Not counting your time (7 h in year one): saves $852 in year one.
Year one = 12 × (what you pay now − the new cost − upkeep) − setup, with your time at $50/hour. Prices are public list prices; hours are Stackcut’s planning estimates. Price it for your team
How we got this
| You pay now: Databox Analyst | flat price | $71/mo |
|---|---|---|
| New cost: Data plugin (Anthropic knowledge-work-plugins) | runs in the agent you already pay for | $0/mo |
| Your upkeep time | 0.5 h/mo × $50/h | $25/mo |
| Difference each month | $71 − $0 − $25 | +$46/mo |
| Over 12 months | +$46 × 12 | +$552 |
| Setup, once | 1 h × $50/h | −$50 |
| Year one | +$552 − $50 | +$502 |
- Not counting your time (7 h in year one): saves $852 in year one.
Year one = 12 × (what you pay now − the new cost − upkeep) − setup, with your time at $50/hour. Prices are public list prices; hours are Stackcut’s planning estimates. Price it for your team
How we got this
| You pay now: Databox Analyst | flat price | $71/mo |
|---|---|---|
| New cost: Google Analytics MCP server (official) | runs in the agent you already pay for | $0/mo |
| Your upkeep time | 0.5 h/mo × $50/h | $25/mo |
| Difference each month | $71 − $0 − $25 | +$46/mo |
| Over 12 months | +$46 × 12 | +$552 |
| Setup, once | 2 h × $50/h | −$100 |
| Year one | +$552 − $100 | +$452 |
- Not counting your time (8 h in year one): saves $852 in year one.
Year one = 12 × (what you pay now − the new cost − upkeep) − setup, with your time at $50/hour. Prices are public list prices; hours are Stackcut’s planning estimates. Price it for your team
Agent skillAgent skill: weekly KPI digest from your exportsTested by Stackcut · Works, with caveatsSaves $102 in year one
How we got this
| You pay now: Databox Analyst (billed annually) | flat price | $71/mo |
|---|---|---|
| New cost: Agent skill: weekly KPI digest from your exports | runs in the agent you already pay for | $0/mo |
| Your upkeep time | 1 h/mo × $50/h | $50/mo |
| Difference each month | $71 − $0 − $50 | +$21/mo |
| Over 12 months | +$21 × 12 | +$252 |
| Setup, once | 3 h × $50/h | −$150 |
| Year one | +$252 − $150 | +$102 |
- Not counting your time (15 h in year one): saves $852 in year one.
Year one = 12 × (what you pay now − the new cost − upkeep) − setup, with your time at $50/hour. Prices are public list prices; hours are Stackcut’s planning estimates. Price it for your team
- Tested by Stackcut: 13 of 13 test cases passed on the first round, and an independent checker agreed with every number (11 digests, 68 to 72 checks each); an independent round (a fresh agent ran every case, a separate agent graded it) passed 12 of 13. our test
- No new subscription if you already pay for an agent that runs skills: a skill is a folder of files the agent loads. agentskills.io
- Handles the traps that make weekly numbers wrong: UTC payment times against your local week, missing days, duplicate rows, cents vs dollars and a definition that changed mid-month. our test
- Can run weekly in the agent you already use: Claude's scheduled tasks run weekly and can use your skills and connectors. support.claude.com
- Prepares only: nothing is sent, and no connector write tool is used. our test
- No connectors or syncing of its own: someone exports four files a week, or saves what connectors you already set up return. our test
- No live dashboard, team logins or anomaly detection; Databox includes anomaly detection even on its Free plan. databox.com
- Databox's Free plan (1 user, 3 data sources, 1 dashboard, 10 custom metrics) may already cover a very small setup. databox.com
- In an independent round (a fresh agent ran every case, a separate agent graded it) 12 of 13 passed: every number was exact, but one copy for an investor stated the founder's explanation as fact. Read it before you send it. our test
Step-by-step: set up, move your data, check, roll back · about 3 h (estimate)
Your agent builds a one-page weekly KPI digest from your own exports and a written definitions file; a script makes every number and a person sends it.
Measured: about 35 minutes of agent time to build the skill and its test kit, and about 7 minutes per test round (13 cases). The 3 hours adds an estimate of the person's time to write their own definitions, export 8 weeks and compare a past week with current reports.
Before you start
- List the KPIs you actually read in Databox, with each one's formula, owner and target; they become metrics.json.
- Your raw numbers stay in GA4 and Stripe. Databox exports its own stored data only if you ask its support, case by case, so rebuild history from the sources. help.databox.com
- Check the time zone of each export: Stripe's financial reports use your account's time zone by default, and you can choose UTC. support.stripe.com
Set up
- Put the skill folder in your agent's skills directory; it follows the Agent Skills layout (SKILL.md, references/, scripts/). agentskills.io
- Replace references/business.md, metrics.json and sources.md with your own and keep rules.md. Give every KPI a written formula, owner and target (or none), and date any change as a new version. our test
- To pull from a connector instead of exporting, give the agent read access only: Stripe's MCP server reads with stripe_api_read and changes data with stripe_api_write. docs.stripe.com
- Schedule it weekly in your agent's own scheduler, set to prepare only; ChatGPT/Codex scheduled tasks can use skills (references/schedule.md covers Claude, Cursor and Grok). learn.chatgpt.com
Move your data
- Recreate each Databox custom metric and goal you use as a metrics.json entry, and export 8 weeks from each source so the charts have history.
Check it worked
- Run the digest for two past weeks and compare with Databox; explain every gap (time zone, refund dates, duplicates) before you trust it. our test
- Whenever you change metrics.json, re-check the numbers with the checker in tests/ (or a second tool) before the next digest goes out. our test
Roll back
- A cancelled Databox subscription can be reactivated from Settings > Billing, and synced data stays accessible while the account is active. help.databox.com
Watch out for
- Analyst is $71/month billed annually, and a cancelled Databox plan keeps running (and syncing) until the term ends, so cancel near renewal. help.databox.com
- When Databox syncing pauses, metrics with historical data limits can keep permanent gaps if you come back later. help.databox.com
- Stripe's API gives amounts in the currency's minor unit (1000 = 10 USD) while dashboard exports show dollars; mixing them is a 100x error. docs.stripe.com
- GA4 data can change for 24 to 48 hours, so run the weekly digest on Tuesday rather than Monday morning. support.google.com
- Scheduled ChatGPT/Codex tasks run unattended (approval_policy "never" where your organization allows it), so the task's prompt must say not to send anything. learn.chatgpt.com
Reviewed 2026-09-29.
Sources: metabase.com
Sources: metabase.com
Replace with AIReplace with AI: a few recurring kpi reportsSavings not modeled yet
Sources: databox.com
Tested by Stackcut
Tested by Stackcut · Works, with caveats
2026-09-29 · Stackcut (our own test, not a customer)
Databox seat → an agent skill that builds a weekly KPI digest
Question. Can a skill folder (SKILL.md, reference files and a local script) turn a small business's own exports into a one-page weekly KPI digest with exact numbers, without a Databox subscription? Path: Agent skill: weekly KPI digest from your exports.
Tested on. Harbor & Pine Studio (an invented business): Synthetic exports shaped like the ones a small business already has: GA4 sessions by day, Stripe payments (in dollars) and refunds (in cents), and app signups, 9 weeks with every hard case planted on purpose. No real data.
Numbers
- 13 test cases, 11 digests, an independent Node checker
- Pass rate by round: 13 of 13; independent round (fresh runner, separate grader): 12 of 13
- 68 to 72 exact-number checks per digest, each on one printed A4 page
- About 7 minutes of agent time per test round
Verdict
A good replacement for a few weekly KPI reports if you can export (or connect) your sources and write down your definitions: the numbers are exact and a person sends the digest. It isn't a live dashboard, it has no connectors of its own, and an independent round (fresh runner, separate grader) passed 12 of 13 with every number exact; run it on your own past weeks before you cancel.
Good fit when
- You read a handful of KPIs once a week
- Your numbers come from a few sources you can export or already connect to your agent (GA4, Stripe, your app)
- One person reviews the digest and sends it
Not a fit when
- The team needs a live dashboard with logins, alerts or drill-down
- You need many sources syncing automatically
- Reports must go out without a person checking them
What we found
- The numbers were exact on the first run. The independent checker agreed with the script on all 11 digests (68 to 72 checks each), including a refund that made net revenue negative (-$1,031.50), Stripe API amounts in cents, 168 duplicate payment rows and a missing GA4 day.
- The time zone is the case people will actually hit. Stripe's dashboard total for 21–27 Sep ($3,991.00, cut by UTC dates) differs from the Lisbon-week total ($4,090.00) by three payments made late on a Sunday UTC. The skill explains the gap row by row instead of changing either number.
- Putting the arithmetic in a script mattered more than the wording. The render step rejects any sentence number the digest doesn't show, so the only free text (3 sentences) can't carry a wrong number.
- Databox's Analyst plan ($71/mo billed annually) doesn't include 'Sharing & notifications' (scheduled snapshots and weekly updates) or Routines, according to its pricing table; those start on Team ($199/mo billed annually). The skill gets the weekly schedule from the agent's own scheduler instead (Claude, ChatGPT/Codex, Cursor and Grok document it), and a person still sends the digest.
- The author's round was run and graded by the agent that wrote the skill. The independent round (fresh runner, separate grader) then passed 12 of 13, every number exact; the failure was wording in an investor copy. The grader also noted one inconsistency: the Never list forbids typing a number the script didn't produce, but explaining the Stripe gap needed a subtraction ($99.00) that the script doesn't print.
- It has no connectors of its own. Every week someone exports four files, or the agent saves what connectors the person already set up return. The connector path was tested only on saved Stripe and GA4 output, not on a live pull.
- In the independent round every number was exact; one case (10) failed on wording: the copy meant for an investor stated the founder's explanation as fact, and the hand-off didn't say to add the investor's address.
How we tested
- Wrote the skill in the agentskills.io format: SKILL.md (steps, reply format, Never list), references/ for the business, the metric definitions (metrics.json: formula, owner, target and dated versions for each KPI), the sources, the data rules, the writing rules and scheduling, plus scripts/kpi_digest.py (Python standard library only). The script makes every number, a small SVG chart per KPI and a one-page HTML digest, and it refuses the 3 written sentences if they contain a number the digest doesn't show.
- Generated synthetic exports for an invented business (fixed seed) and wrote 13 cases and a pass/fail rubric before the first run: a clean week, a missing GA4 day, a refund bigger than the week's revenue, a Stripe total cut in UTC against the business's Lisbon week, duplicate rows, a KPI definition changed mid-month, connector output in cents, a KPI with no target plus a request for a new KPI, an out-of-scope request, a request to send the digest and leave a refund out, conflicting GA4 rows, a missing signups file and a scheduling request.
- Wrote an independent checker in Node (tests/check_digest.mjs) that shares no code with the Python script. It recomputes every KPI from the raw files and the definitions, then compares the values, week-over-week changes, targets, statuses, chart bars, the digest table, the data notes and every number in the 3 sentences, and prints the page to PDF to confirm it fits on one A4 page. The grader ran it on every digest.
- The author's round: the runner and the grader were separate passes by the agent that wrote the skill (no Agent tool in that session). The runner pass used only the skill folder, the cases and the data; the grader pass used the rubric, the references, the run and the independent checker.
- Between rounds only the skill would have been changed. Round 1 passed every case, so the test stopped there.
- After the skill was final, Stackcut ran one independent round: a fresh runner agent (skill folder and case inputs only) and a separate grader agent (rubric, references, checker). Result: 12 of 13; failed: case 10.
All 139 recipes · Savings leaderboard · Every vendor, with prices and cancel steps