Iron Goo
Woman assesses six colored wooden crates on an office shelf, with dusty upper crates and a notebook in hand

The Data Stack for Small Business Without a Data Team

Atamyrat Hangeldiyev
Atamyrat Hangeldiyev
Systems Architect
February 27, 2026
On this page
Analytics & Data

The data stack for small business is usually the reports already in your CRM, payment processor and GA4, plus one spreadsheet. For a company of 10 to 200 people, add a warehouse when a recurring question needs three or more systems joined row by row and someone owns the result every month. One detail changes the free option: the BigQuery sandbox expires tables after 60 days, so a free GA4 export there is not an archive.

Six data stack tiers compared

TierMonthly costSetup hoursSkills neededQuestions answeredLock-in and portabilityWhen to graduate
1. Spreadsheet, GA4 and CRM reports$0 with free tools; paid CRM varies4 to 10Spreadsheet formulas, report setupLeads, revenue and site behavior by sourceSheets and CSVs are portable; reports stay in their toolsReconciliation repeats monthly, or older exploration data is needed
2. GA4 export, BigQuery and Looker Studio$0 within free limits; usage beyond them is billed6 to 16GA4 configuration and SQLEvent history and custom event queriesRaw events are exportable; queries use BigQuery SQLA recurring question needs multiple systems joined
3. CRM reportingHubSpot free for up to 2 users; Starter $10 per seat on monthly billing during a limited-time 50% discount ($20 regular)8 to 20CRM fields, pipelines and dashboardsLeads, deals and customer activityRecords can be exported; report definitions stay behindCRM data alone cannot answer a recurring question
4. Payment analyticsStripe Billing is 0.7% of billing volume; Sigma about $10 to $15 for up to 250 US chargesUnder 1 for Billing; 2 to 8 for SigmaSubscription definitions; SQL for SigmaMRR, churn and payment historyCSV exports are available; Sigma queries use Stripe's schemaRevenue must be joined to other systems regularly
5. ELT and a warehouseFivetran Free $0; Airbyte Standard from $20; Stitch Standard from $100, plus any warehouse charges30 to 80; 2 to 6 monthly upkeepSQL, joins, connector maintenanceRecurring customer-level questions across systemsConnectors can be replaced; SQL dialects differDefined tables need to send data back to operational tools
6. Reverse ETL or an AI agentDepends on the selected tools and usage20 to 60Data definitions, workflow design and ownershipActions based on trusted joined dataMappings and workflows need rebuilding when tools changeReassess when the workflow or its owner changes

Setup and upkeep hours in the table are our estimates for a company with one technically comfortable person and no data engineer. Vendor pages were read on 29 October 2026, and vendors change prices. The tiers describe capabilities, not a shopping sequence: you can configure CRM reporting and payment analytics before you export a single GA4 event.

Four steps rise from spreadsheet and CRM plus payments to optional warehouse and AI agent layers

The two solid lower steps supply many small companies' answers. The upper steps become useful when a recurring question or operational need calls for them.

Tier 1: Spreadsheet, GA4 and CRM reports

When existing reports answer separate questions

Start with a short list of numbers and a source of record for each one. A measurement plan gives you a place to write those definitions before dashboard choices turn into an argument about tools. For example, record what counts as a lead, which date assigns it to a month, and who checks the count.

Use the payment processor or accounting system for money, the CRM for leads and customers, and GA4 for on-site behavior. The CRM is the count of record for leads: do not add conversion totals from ad platforms and call the sum your lead count. The small business analytics guide explains how those sources fit together.

A spreadsheet can hold a monthly snapshot with one row per month and a column for each agreed metric. Keep the source and the date range beside the numbers, so a future reader can reproduce them. A useful first check is whether the CRM's lead count can be matched to the definition your sales team actually uses.

When the spreadsheet stops answering the question

GA4 standard properties let you choose two or 14 months of data retention for explorations and funnel reports. That setting does not shorten standard aggregated reports, according to Google's data retention documentation. If you need an event-level exploration older than 14 months, a monthly spreadsheet total will not recreate the underlying events.

Manual reconciliation also has a cost. If someone repeatedly compares GA4 leads, CRM records and payments, record the time spent and the matching rule used. A mismatch can come from definitions or missing identifiers, so buying a warehouse before resolving those issues can preserve the disagreement with greater precision.

Tier 2: GA4 export to BigQuery and Looker Studio

When raw event history or custom SQL wins

The GA4 BigQuery export gives you raw events to query. Standard properties have a limit of 1 million events per day for Daily export; Streaming export has no event limit. Export to the sandbox is free within its limits, while BigQuery charges for storage and query processing outside the sandbox.

Looker Studio is free and connects to both GA4 and BigQuery. Use it to present a query result alongside a GA4 report, but keep the underlying definitions visible to the people using the dashboard. A paid Looker Studio Pro also exists.

The following query counts events named generate_lead by event date from October 1 through October 28, 2026. The dataset name is a placeholder; replace it with your own BigQuery project and GA4 dataset. The result is an event count, so compare its definition with the CRM lead definition before treating the two as equivalent.

SELECT event_date, COUNT(*) AS leads
FROM `your-project.analytics_123456789.events_*`
WHERE event_name = 'generate_lead'
  AND _TABLE_SUFFIX BETWEEN '20261001' AND '20261028'
GROUP BY event_date
ORDER BY event_date;

When sandbox limits or missing identifiers get in the way

The BigQuery sandbox includes a lifetime limit of 10 GiB of storage, which is not refunded when data is deleted, and 1 TiB of query processing per month. Tables, views and partitions expire after 60 days; streaming data, DML statements and the BigQuery Data Transfer Service are unsupported. Our reading is that a GA4 export in the sandbox holds at most about 60 days of daily tables, and only Daily export is available there because the sandbox does not support streaming. Google's GA4 export help warns that the export can start failing once the project exceeds the sandbox limit, so it can stop entirely once the lifetime 10 GiB is used.

The sandbox is not a GA4 archive

Enable billing before you need history beyond 60 days, or before the sandbox's lifetime 10 GiB is used up, and check the expiration settings on the tables you intend to retain. A free export is useful for testing queries, but expired daily tables cannot serve as your historical record.

With billing enabled, BigQuery's on-demand pricing includes the first 1 TiB of queries per month free, then charges $6.25 per TiB. Active logical storage includes the first 10 GiB free each month. Google's example prices 1 TiB for a full month at $23.552 in us-central1, which is about $0.023 per GiB-month by our arithmetic.

Say a hypothetical dataset uses 50 GiB of active logical storage for a full month. With billing enabled, after the free 10 GiB a month, approximately 40 GiB is billable: 40 × $0.023 is about $0.92 for storage. Query processing is a separate calculation, and the first free TiB applies to that processing.

The export does not automatically match a browser to a paying customer. You need a usable identifier and a documented matching rule before you can join an event to a CRM record or payment. An accurate event count is useful on its own; it is not proof that the matching problem has been solved.

Tier 3: CRM reporting built into HubSpot, Pipedrive or Zoho

When pipeline and customer questions stay in the CRM

A CRM dashboard can answer questions such as how many qualified leads entered a pipeline, how many deals moved stages, and which records need follow-up. Before building it, decide which fields are required and what each stage means. Our estimate is 8 to 20 setup hours, mostly for those field and pipeline decisions.

On HubSpot's pricing pages, the free CRM was listed for up to two users. HubSpot listed 10 dashboards on Free, 30 on Starter and 75 on Professional, each holding up to 50 reports; those are Marketing Hub figures, with Free and Starter also shown on the CRM Starter page. On its Sales Hub pricing page, read on 29 October 2026, HubSpot listed Starter at $10 per seat per month on monthly billing under a limited-time 50% discount, making the regular rate $20 by our arithmetic, or $7 per seat per month billed annually at 65% off. Professional was $100 per seat per month on monthly billing, or $90 per seat per month billed annually. The custom report builder starts at Professional, with up to 100 custom reports.

Pipedrive's entry plan is called Lite. An archived copy of its pricing page from 23 October 2026 listed Lite at CA$19 per seat per month billed annually; check Pipedrive's live page for pricing in your currency. Zoho CRM lists a free edition for up to three users and unlimited reports on Professional; the page served to us showed Indian rupee prices, so check your local price.

When a CRM report cannot join the records

A CRM report is only as dependable as the fields people maintain. A required stage, a consistent close date and a named owner for exceptions can improve a report without adding another reporting product. Check for empty fields and stale deals before making the dashboard more elaborate.

Exporting CRM records helps preserve the data when you switch products. Our reading is that report definitions, pipelines and dashboards do not migrate with those records, so budget time to rebuild and check them. If the question requires payment and advertising records joined to each customer, the CRM report alone may not have the necessary rows or identifiers.

Tier 4: Payment processor analytics and Stripe Sigma

When subscription and payment reports answer the question

If you already use Stripe Billing, open Billing overview before buying a separate subscription dashboard. Our estimate is under one hour to get started with its existing analytics. Stripe Billing's pay-as-you-go price is 0.7% of billing volume, which is a cost of billing itself, not a separate analytics subscription.

Stripe's Billing analytics definitions are specific enough to change how you read the charts. MRR is the monthly-normalized value of active and past_due subscriptions, excluding taxes, free plans and usage-based products. Subscriber churn divides subscribers churned in the last 30 days by active subscribers 30 days ago plus new subscribers in the last 30 days; Stripe's example is 100 divided by 1,100, or 9.1%.

Stripe estimates subscriber lifetime value as ARPU divided by subscriber churn rate. Its example uses $500 divided by 9% to produce $5,555. Our reading is that this is a revenue estimate, not profit, because the formula does not subtract costs.

You can export a metric's summary table or underlying rows as CSV. Prebuilt reports include MRR per subscriber per month, subscription metrics per month, and customer MRR changes. Our reading is that Billing analytics covers subscriptions billed through Stripe Billing, so check other payment or accounting sources separately.

When dates differ or a custom payment query is needed

Stripe's prebuilt reports use UTC dates, while charts use the account time zone. Monthly totals can therefore differ around a boundary, even when both reports are working as designed. Metric configuration changes take 24 to 48 hours to show, so record the change and allow that window before investigating a discrepancy.

Stripe Sigma supports SQL over Stripe data when the prebuilt reports do not answer a payment question. Stripe Sigma pricing is about $10 to $15 a month at up to 250 charges in the US; Stripe prices by region. Our estimate is 2 to 8 setup hours for the first SQL reports, and those queries depend on Stripe's schema.

Tier 5: No-code ELT into a warehouse

When a recurring question needs customer-level joins

ELT moves records from source systems into a warehouse, where you can join and query them. A warehouse earns its upkeep when a recurring question needs three or more systems at row level, such as ad spend, CRM deals and Stripe revenue per customer. The person responsible for the report also needs a matching rule for customers and a monthly check for broken or incomplete data.

Ads, CRM and payment records join into one customer row for a recurring warehouse question

Notice the shared customer row: the question needs records from ads, the CRM and payments together. A recurring need for that join is a practical trigger for a warehouse.

BigQuery, Postgres and Snowflake are possible destinations. Postgres is open source; you pay for the server or managed instance that runs it. Snowflake offers a free trial and uses credit-based pricing, so check its consumption table for your expected workload. SQL concepts travel between warehouses, although dialect differences mean queries can need changes.

Fivetran's Free plan includes up to 500,000 monthly active rows, or MAR, for connections. MAR counts inserts and updates, including deletes; unchanged rows in re-syncs and the initial bulk load are excluded. A $5 base charge applies to standard connections with 1 to 1 million MAR, while the Free plan is exempt.

Airbyte Core is free to self-host, with your team responsible for running, upgrading and supporting it. Airbyte Standard cloud starts at $20 a month with five credits, about 1 GB of syncs, and hourly syncs; credit use varies by source type. Stitch Standard starts at $100 a month for 5 million rows, with US-only pricing shown on its page. Compare the units before comparing the prices: a MAR, a synced gigabyte and a row allowance are different measures.

When connector maintenance costs more than the question

Our estimate for the first trustworthy joined report is 30 to 80 setup hours, followed by 2 to 6 hours a month for upkeep. That monthly work includes checking connector failures, source schema changes and whether the join still produces credible rows. A free connector plan does not supply that owner.

Consider a hypothetical 12-person company with four HubSpot Sales Hub Starter seats, monthly billing and no data engineer. It uses GA4 and Looker Studio, stays inside BigQuery's free tier, has up to 250 US charges for Sigma, and keeps its Fivetran usage within the Free plan. Assume an internal cost of $60 per hour for the person doing the warehouse work.

Hypothetical monthly tool bill and warehouse labor

CRM at HubSpot Starter's regular $20 rate after the current limited-time discount ends = 4 seats × $20 = $80/month GA4 + Looker Studio = $0/month BigQuery within free tier = $0/month Stripe Sigma = about $15/month Fivetran Free = $0/month Tool total = about $95/month

Tier 5 setup = 30 to 80 hours × $60 = $1,800 to $4,800 once Tier 5 upkeep = 2 to 6 hours × $60 = $120 to $360/month

The hours are our estimates, and the hourly rate is an assumption for this example. Stripe Billing's 0.7% charge is excluded because it is the cost of billing itself. The tool bill is small beside the setup and upkeep time.

Server-side tagging is not a starting tier. Google's Cloud Run setup guide recommends at least two instances and estimates approximately $45 per server per month, or about $90 for two by our arithmetic. A server container cannot recover a purchase the browser never sent; the tracking plan guide addresses what to check before adding that infrastructure.

Tier 6: Reverse ETL or an AI agent layer

When defined warehouse data needs to trigger work

Reverse ETL sends selected warehouse data back to operational tools. It can be useful when a trusted customer table needs to update a CRM field on a recurring schedule and someone owns both the definition and the destination. Our setup estimate for this tier is 20 to 60 hours.

Before sending data back or giving an agent access, use the checks for getting data ready for AI: write down what each table and field means, who owns it, and how errors are corrected. Once those definitions exist, operations automation can act on CRM and payment data as part of a defined workflow. Keep the action and its source visible to the person responsible for checking it.

When an automated action has no owner

An agent cannot settle a disputed definition by reading more rows. If “active customer” means one thing to finance and another to sales, write the rule for the intended action before using it in a workflow. Check that the destination field has an owner and that the owner can identify and correct a bad update.

A warehouse table with no written definition is also a weak input for reverse ETL. Sending it into the CRM can spread an error to the people who use that field every day. Keep the process manual until the source, matching rule and review responsibility are clear.

Graduation rules for the next recurring question

These are our decision rules for choosing the next step. Each trigger is something you can observe, and each assumes someone is responsible for the resulting report or workflow.

  1. Leave the basic spreadsheet workflow when the same GA4, CRM and Stripe reconciliation is done by hand more than once a month, or when a question needs exploration data older than 14 months.
  2. Turn on the GA4 BigQuery export when you need event-level history or joins the GA4 interface cannot do. Enable billing before you need history beyond 60 days, or before the sandbox's lifetime 10 GiB is used up.
  3. Add ELT and a warehouse when a recurring question needs three or more systems joined at row level and a named person owns the result monthly. Write the customer matching rule before treating the joined total as authoritative.
  4. Add reverse ETL or an AI agent when the warehouse tables have written definitions and an owner, and a recurring operational action needs those trusted fields.
Free ebook

Put your next steps in order

Build on what you have just read with a free ebook on semantic SEO and AI visibility. Get worked examples, decision rules, checklists and a do-this-week action in every chapter, plus a six-week plan for your site.

By signing up and confirming your address, you consent to receive the free ebook plus one email every two weeks with 3 or 4 new Iron Goo videos, guides and articles, and nothing else; unsubscribe any time with one click.

Related in Analytics & Data

Ready to move?

Tell us where your business is today. A written assessment lands in your inbox within two business days.

Talk to us