Twenty integrations an online shop actually needs.

On 18 August 2026 Google switched off the Content API for Shopping, and from 1 September requests against it began failing progressively. Any shop whose product feed was wired up programmatically before this year is either broken now or breaking shortly, and most of the owners do not know yet — the feed just quietly stops matching the shop. That is what an integration really is: not a feature you add, but a job somebody stops doing by hand, and a dependency you have taken on. This is the list of twenty worth taking on for an online shop, counted down from twenty to one on how certain the payback is. Number twenty helps a narrow set of shops. Number one is the one without which you do not have a shop, you have a catalogue. Every fee below is the vendor’s own published figure, checked in September 2026.

How we ranked them

Four questions, in the same order for every entry. What job does it remove, precisely enough to name? What does that job cost the business now, in hours a week or in mistakes a month? What does the integration cost to build once? And what does it cost to run for ever — because every integration has two meters, the vendor’s and yours.

That last question is the one nobody asks before signing. An API integration is not a purchase, it is a marriage to somebody else’s roadmap. Endpoints get versioned, credentials expire, a field you relied on becomes optional, and a vendor sunsets the whole interface with eighteen months of notice that nobody forwarded to you. The Content API sunset in August is the ordinary case, not a freak event.

One entry breaks the ranking logic on purpose. Payments is not on this list because it pays back; it is on it because without it you are not trading. Compliance work is ranked differently again — it sits low not because it matters less but because whether it applies to you at all depends on where you sell, and a list cannot know that. Everything else earns its place by arithmetic.

One note on the figures. Vendors publish in dollars and euros, and we have converted nothing — a converted price is stale the day you read it, and at any recent rate none of the comparisons change order.

20. Tax and VAT calculation

What it removes: the quarterly spreadsheet in which somebody works out which country each order was taxed in, and at what rate. It is last because for most shops it removes nothing at all — sell only at home and the platform already handles it. It becomes real the moment you sell across borders in volume, and then it stops being useful and becomes unavoidable.

One €10,000 threshold decides which of those two shops you are. Below that in annual cross-border sales to consumers inside the EU you charge your own country’s rate and file nothing extra. Above it you owe the rate of the customer’s country on every sale, across twenty-seven member states whose standard rates run from 17 to 27 per cent and whose reduced rates for things like books and food agree with nobody. The One Stop Shop collapses the filing into one quarterly return in one country, but it calculates nothing — knowing the correct rate at checkout is still your problem, and that is what the integration is for.

Outside the EU it fragments further: imported consignments up to €150 have their own scheme, the United Kingdom has its own registration rules, and in the United States sales tax is a state matter with economic-nexus thresholds commonly around $100,000 of sales into a state you have never visited. If any of that describes your shop, the integration costs less than the first correction. If none of it does, spend the money on number five instead and revisit this when the threshold is in sight.

19. Loyalty and referral

What it removes: nothing, which is why it is nearly last. It adds a mechanism — points, tiers, a referral code — and mechanisms are only worth their integration cost when there is repeat purchasing to accelerate. For a shop selling something people buy every six weeks, a loyalty API pays for itself. For a shop selling mattresses, it is an elaborate way to discount the customer who was going to buy anyway.

Ask one question before building it: what is your current repeat-purchase rate? If nobody knows, that is the project — instrument the shop first, at number five, and let the number decide whether this one is worth the wiring.

18. Address autocomplete and validation

What it removes: failed deliveries caused by addresses people typed wrong, and the support conversation that follows each one. An autocomplete field at checkout also shortens the form, which is worth something on mobile where the abandonment happens.

The reason it is not higher: the courier is frequently the address validator whether you like it or not, and a good share of failed deliveries have nothing to do with typos — the recipient was out, or the phone number was wrong. Fix the phone field before you buy an address API.

17. Reviews and ratings

What it removes: the manual chase for reviews after delivery, and the guesswork about which products are quietly disappointing people. Wire the platform to your order data and the request goes out on its own, keyed to the actual delivery date rather than the order date.

The honest limit: review stars in search results come from structured data on your own pages, not from the review platform’s dashboard, and a badge that renders in a widget after the page loads does nothing for you. If rich results are the reason you are buying this, check that the markup lands server-side before you sign anything.

16. Site search

What it removes: the internal-search experience that a database LIKE query gives you — no typo tolerance, no synonyms, no sensible ranking, and no idea what people searched for and did not find. A hosted search API fixes all four, and the fourth is the one that pays: search terms with no results are a product roadmap somebody is handing you for free.

Who should skip it: shops under a few hundred SKUs where browsing works fine. Search matters in proportion to catalogue size, and below that threshold the money is better spent on the category pages people are actually using.

15. Returns and RMA

What it removes: the email thread. A returns portal that talks to your order system and the courier turns a five-message exchange into a form and a label, and every message you remove is a support hour you keep.

It ranks here rather than higher because the volume has to justify it. Under roughly a return a day, a well-written page explaining the process removes most of the same cost for none of the build. In the EU you owe the fourteen-day withdrawal right regardless, so this is about the mechanics, never about the obligation.

14. SMS and WhatsApp Business

What it removes: the “where is my order” contact, which for most shops is the single most common support message. A delivery notification that arrives before the customer wonders is the cheapest support you will ever buy.

Watch the meter, because it changed. Since 1 July 2025 WhatsApp bills per message rather than per conversation: marketing templates are charged on every delivery, while utility and authentication templates are free inside an open 24-hour customer service window. Marketing templates also cost roughly five times more in Germany than in the United States — about $0.124 per message against $0.025 — so a campaign that is cheap in one market is a budget item in another. Your provider then adds its own markup on top of Meta’s rate.

The practical read: use it for utility messages, which is where it is nearly free and genuinely useful, and be very careful about treating it as a marketing channel. Plain SMS remains the fallback for the customers who do not use WhatsApp, and depending on the market that is a larger group than the WhatsApp figures suggest.

13. Fraud scoring and 3-D Secure

What it removes: manual review of orders that look wrong, and the chargebacks you did not catch. Most payment processors include a risk layer with rules you can tune, which means this is frequently a configuration job rather than an integration — check what you already pay for before buying a second system.

Strong customer authentication is already mandatory in the EU, so the interesting question is not whether you authenticate but how much friction you add to good customers. Every extra challenge costs conversions, and a rule set copied from a blog post will decline real orders. Tune it against your own declines, monthly, for the first quarter.

12. Consent Mode v2 and a consent platform

What it removes: the gap between what your analytics says and what actually happened, in the specific case of European traffic. Without consent signals wired into Google’s tags, advertising and measurement data for EEA users is not merely incomplete — it is not collected at all, and no amount of dashboard work recovers it later.

This is not a growth project, it is the foundation under numbers five and six. A consent banner that blocks tags but never tells Google what the visitor chose gives you the worst of both: the friction of the banner and the blindness of no data. If your shop advertises to EEA customers, verify the signal is actually being sent before you trust a single conversion figure.

11. CRM sync

What it removes: the spreadsheet. Every shop that also sells by phone, quote or invoice ends up maintaining a second, worse copy of its customer list somewhere, and the sync is what stops the two versions diverging. Order history landing on the contact record is what makes a follow-up call informed rather than annoying.

We build three, each its own package: HubSpot Integration at €499, Pipedrive Integration at €499 and Salesforce Integration at €499, described on the HubSpot, Pipedrive and Salesforce pages. The choice between them is mostly about who else in the business has to use it, and the honest answer for a small shop is often Pipedrive because people actually open it.

Who should skip it: a pure self-serve shop with no sales conversations. If nobody ever picks up the phone to a customer, a CRM is a database you pay a monthly fee to admire. The email platform at number three is the tool you wanted.

10. Google Search Console

What it removes: guessing. Which of your product pages are actually indexed, which queries bring people who buy, and which of your own pages are competing for the same search — none of that is visible from inside the shop, and all of it is available through Google Search Console.

The API earns its place over the interface once the catalogue is large enough that page-by-page checking stops being feasible. Bulk URL inspection tells you what Google thinks of a thousand product pages in one pass, which is how you find the four hundred that were never indexed at all.

The honest limit: it is a diagnostic, not a lever. Nothing you do here changes a ranking; it tells you where to look. Shops that treat the coverage report as a to-do list get further than shops that treat it as a scoreboard.

9. Meta Ads Conversions API

What it removes: the purchases your ad platform never learned about. Browser pixels are blocked, cleared and consented away, and every purchase that fails to report is a purchase the bidding algorithm cannot learn from. Meta Ads accepts the same events server-side from your shop, where nothing blocks them.

It ranks below its Google equivalent for one reason: deduplication. Send the same purchase from both the pixel and the server without a shared event ID and you will double-count revenue, then optimise against a number that flatters you. Get the IDs right or do not do it at all.

8. Automation glue

What it removes: the copy-paste between two systems that will never have a native integration. New order in the shop, row in the sheet the warehouse actually reads, message in the channel the packer watches, invoice draft in the accounting tool. None of that deserves a bespoke build, and all of it deserves to stop being manual.

We wire all three: Zapier Automation at €299, Make Automation at €299 and n8n Automation at €399, described on the Zapier, Make and n8n pages. The price difference reflects the meters: Make bills per module step, so a ten-step scenario costs ten operations per run, while n8n bills per workflow execution regardless of steps and can be self-hosted with no licence fee at all. Which is cheaper depends entirely on the shape of your flows, and we work that out before quoting. The long version is in Zapier vs Make vs n8n.

The honest limit, and it is the one people learn expensively: automation glue is where undocumented business logic goes to hide. Six scenarios in, nobody remembers what runs at 3am or why. Anything load-bearing eventually deserves real code and a test; the glue is for the long tail.

7. Accounting and ERP sync

What it removes: re-typing every order into the accounting system, and the month-end reconciliation between what the shop says it sold and what the books say it invoiced. For a shop doing more than a few dozen orders a day, this is the largest single block of manual work left in the business, and it is entirely mechanical.

It ranks this high because it is usually the same project as invoicing rather than a neighbour of it: the invoice has to exist correctly in your own system, with a number sequence nobody can break, before anything downstream can report it or reconcile it. Shops that treat the two as separate jobs end up with two numbering schemes and an auditor asking about both.

What to check before you build: whether your accounting software has an API at all, or only an import format. Several of the packages small shops run on are import-first, and a nightly file drop is a perfectly respectable integration if that is what the other side supports. The goal is that nobody types an order twice, not that the transport is fashionable.

6. Google Ads conversion import

What it removes: bidding on the wrong outcome. Left alone, Google Ads optimises towards whatever it can see, which for many shops is a form submission or an add-to-cart rather than a paid order. Importing the actual purchase — with its actual value — changes what the algorithm chases, and it is the single highest-leverage change available to most shops spending on search.

Enhanced conversions and offline import matter most where the sale completes away from the browser: bank transfer, cash on delivery, or an order confirmed by phone two days later. Cash on delivery is still a large share of orders across much of Central and Eastern Europe, which means a shop reporting only card conversions is training its bidding on a biased sample of its own customers.

The honest limit: better data does not fix a bad offer, and it will not rescue a campaign structure that is fighting itself. It does mean that when you fix those, you will be able to tell.

5. Google Analytics 4

What it removes: arguments. Which channel produced which revenue, where the checkout leaks, whether the redesign helped — all of it is unanswerable from order data alone, and all of it is answerable once purchase events are landing correctly with their items and values attached.

We do this as Google Analytics 4 Setup at €199, which is the cheapest thing on this list and the one that makes the four entries above it measurable. The Google Analytics 4 page has the detail. Server-side purchase events are the part worth insisting on: a browser-only measurement plan loses exactly the events you care about most, to ad blockers and to consent.

The honest limit: a dashboard nobody opens is an expense. Pick three numbers you will actually act on — revenue by channel, checkout completion, and the conversion rate of your top ten products — and ignore the other four hundred until one of those three moves.

4. Marketplace listings

What it removes: running a second shop by hand. Selling on a marketplace without an integration means maintaining prices, stock and orders in two places, and the failure is not gradual — you oversell an item that went out of stock on your own site an hour ago, and the marketplace penalises you for it.

The one we build is eMAG Marketplace at €899, the most expensive single integration we sell, and the price reflects what it involves: category mapping, mandatory attributes per category, offer and stock sync, order pull, and invoice handling. The eMAG Marketplace page has the mechanics; the API is at version 4.5.1 as of March 2026 and does not stand still. Amazon, eBay and the regional marketplaces are the same shape of work with different field names.

The arithmetic to do first, whichever marketplace it is: commission runs from about 5 per cent in electronics to 20 per cent in fashion depending on category, on top of your own margin. For some catalogues that is the cheapest customer acquisition available anywhere. For others it is selling at a loss with extra steps, and the integration cost is the smallest number in the decision. Work out the per-category margin before you build anything.

3. Lifecycle email

What it removes: the revenue you are currently leaving in abandoned carts. This is the one integration on the list whose return is easy to calculate in advance: count the carts abandoned last month, take the recovery rate a competent flow achieves, multiply by average order value. It is almost always larger than the integration and the subscription combined.

We build three, each its own package: Klaviyo Integration at €249, Mailchimp Integration at €249 and ActiveCampaign Integration at €249, described on the Klaviyo, Mailchimp and ActiveCampaign pages. Klaviyo is the e-commerce-native one and the one we recommend by default for a shop, because product and order data arrive as first-class objects rather than as custom fields somebody has to maintain.

Watch how it scales, because it does not scale on sends. Klaviyo bills on active profiles — a change made in February 2025 — and is free to 250 of them, $20 a month at 500 contacts, around $175 at 5,000 and roughly $350 at 10,000. A list you never clean is a subscription that grows on its own, which is why the first flow we build on any shop is the one that removes people who have not opened anything in a year. Above $10,000 of monthly spend the top tier becomes mandatory and adds twenty per cent, which is a cliff worth knowing about long before you reach it.

2. Shipping carriers and AWB

What it removes: the courier’s web portal. Generating waybills by hand for every order — copying the address, picking the service, printing the label, then pasting the tracking number back into the shop and into an email — is fifteen minutes an order that disappears entirely, and it is the integration whose absence is most obvious to the person doing the work.

It also removes a whole class of error. An address typed twice is an address that can differ twice, and the version the courier has is the one that decides whether the parcel arrives.

The honest limit: every carrier is its own project. A carrier like DPD or DHL comes with its own API, its own idea of what a parcel is and its own pickup-point network, and the regional carriers you will also end up wanting are no different. Three carriers is three integrations, not one with a dropdown, and the pickup-point selector at checkout is a separate piece of work again. Start with the one that carries most of your volume and add the second only when the volume justifies it.

1. Payments

The one that is not optional in any sense. Everything else on this list makes a shop better; this one makes it a shop. It goes first not because it is clever but because its arithmetic is the largest of any integration here — a fee measured in percentage points of everything you ever sell.

So read the fees properly, not the headline. In the EEA, Stripe charges 1.5% + €0.25 for standard European cards, 1.9% + €0.25 for premium and commercial European cards, and 3.25% + €0.25 for cards from outside the EEA, plus 1% for currency conversion. Two details matter: the fixed component follows the settlement currency of your account rather than always being €0.25, and commercial cards are not an edge case — if you sell to businesses, a meaningful share of your volume pays the higher rate, and a margin calculated on 1.5% is wrong.

We build it as Stripe Payments at €399, described on the Stripe page. What that price buys is not a checkout button; it is webhook handling that survives retries and duplicates, an order state machine that agrees with the payment state, and refunds that do not require somebody to log into two systems. The button is the easy half.

The local payment method question is separate and it is not optional either. A card-only checkout quietly declines a real share of buyers in any market where cash on delivery, bank transfer or a local wallet is what people expect, and no amount of processor tuning recovers them. Whatever you conclude about the rest of this list, do not conclude that payments are solved because cards work.

What we left out, and why

Electronic invoicing, which is only left out in the sense that it depends entirely on where you invoice. Where a country mandates it, it stops being a ranking question and moves to the front of the list on its own. Romania has required B2B invoicing through the ANAF platform since 1 July 2024 and B2C reporting since 1 January 2025, on a clearance model in which the invoice is not legally valid until the platform validates it, and we build that as e-Factura at €499 — the e-Factura page has the mechanics. Germany is looser and later: every business has had to be able to receive an EN 16931 invoice since 1 January 2025, but issuing only becomes mandatory on 1 January 2027 above €800,000 of turnover and on 1 January 2028 for everyone else, with no central platform in the middle. If you invoice somewhere with a live mandate, ignore its absence here and do it first.

A PIM. Product information management is the correct answer to a catalogue maintained by four people across three channels, and the wrong answer to a catalogue maintained by one person in a spreadsheet. The threshold is real but most shops asking about it are below it.

Personalisation and recommendation engines. They work, and they work in proportion to traffic. Below a few thousand sessions a day the engine has nothing to learn from, and a hand-picked “goes well with” list built by somebody who knows the products beats it outright.

Multi-currency. Usually a setting in the payment processor rather than an integration, and frequently a mistake: showing prices in eight currencies while shipping to two countries adds checkout complexity in exchange for nothing.

A headless CMS, which people ask about in this context but which is an architecture decision rather than an integration. It belongs with WordPress or headless CMS and what is headless commerce, not on this list.

And one of our own, for completeness: ClickMed is an integration we build and sell, and it has no place on an e-commerce list — it wires clinic software, which is a different business with different obligations. Listing it here to pad the count would have been the same dishonesty as the forty-item lists this one is a reaction to.

The order to build them in

Nobody builds twenty integrations. The sequence that has served our clients best runs: payments first, because trading is not optional, and electronic invoicing immediately beside it wherever a mandate applies to you; then whichever of shipping or marketplace carries your volume; then measurement, meaning analytics, consent and ad conversions as one piece of work rather than three; then lifecycle email, which is the first one whose return you can put on a slide; then glue for the long tail.

Everything after that should be triggered by a number you are already looking at, not by a list. A rising “where is my order” contact rate justifies notifications. A catalogue that outgrows browsing justifies search. Repeat purchases justify loyalty. If you cannot name the number an integration is supposed to move, you are buying a dependency and calling it a feature.

If you want this sequenced against your own shop rather than in the abstract, that is what our API Integrations work is, and the ones we build ourselves are fixed-price packages you can buy on their own.

Sources

Every fee, rate and deadline above comes from the vendor’s or the authority’s own published pages, checked on 9 September 2026. Where a figure changed recently, the change is noted — several of these moved this year.

• Google — Content API for Shopping release notes ↗

  Content API for Shopping was sunset on 18 August 2026, with requests experiencing progressive errors from 1 September 2026. Migration is to the Merchant API; manual file uploads, scheduled fetches and Google Sheets feeds are unaffected, so only programmatic integrations break.

• Stripe — EEA pricing ↗

  Standard EEA cards at 1.5% + €0.25, premium and commercial EEA cards at 1.9% + €0.25, non-EEA cards at 3.25% + €0.25, plus 1% for currency conversion. The fixed component is charged in the settlement currency of the account — an account settling in RON is charged 1 LEU rather than €0.25.

• European Commission — VAT rules for e-commerce ↗

  Since 1 July 2021 a single EU-wide threshold of €10,000 in annual cross-border sales to consumers replaces the former per-country distance-selling thresholds. Above it, VAT is due at the customer’s rate, reportable through the One Stop Shop as a single quarterly return filed in one member state. Imported consignments up to €150 are handled through the Import One Stop Shop. Standard rates across the twenty-seven member states range from 17% to 27%.

• eMAG Marketplace — API documentation ↗

  Marketplace API at version 4.5.1, valid from 2 March 2026. Covers product and offer upload, category mapping with mandatory attributes per category, stock and price synchronisation, order processing and invoice issuance.

• Meta — WhatsApp Business Platform pricing updates ↗

  Per-message pricing replaced conversation-based pricing on 1 July 2025. Marketing templates are billed on every delivery; utility templates sent inside an open 24-hour customer service window are free; authentication templates are always billed. Rates are per-country — marketing runs about $0.025 in the United States against roughly $0.124 in Germany — and providers add their own markup on top.

• Klaviyo — pricing ↗

  Billed on active profiles since February 2025 rather than on profiles emailed. Free to 250 active profiles; the email plan starts at $20 per month at 500 contacts, around $175 at 5,000 and roughly $350 at 10,000. Features are identical across tiers — the price tracks list size. Above $10,000 monthly spend the top tier becomes mandatory and adds 20%.

• Make — pricing ↗

  Core at $9 per month for 10,000 operations, with a free tier at 1,000. Each module action in a scenario counts as one operation, so cost scales with the number of steps as well as the number of runs.

• n8n — pricing ↗

  Cloud Starter at €20 per month for 2,500 executions, billed per full workflow run rather than per step, with unlimited steps and workflows. The self-hosted community edition carries no licence fee.

• Google — Consent Mode for the EEA ↗

  Advertisers using Google’s advertising and measurement products for EEA traffic must pass consent signals to Google’s tags. Without them, ad and measurement data for those users is not collected, and the loss is not recoverable retrospectively.

• ANAF — RO e-Factura ↗

  Mandatory B2B e-invoicing through the RO e-Factura platform since 1 July 2024 and B2C reporting since 1 January 2025, in CIUS-RO format on UBL 2.1. A clearance model: the invoice is not legally valid until the platform validates it. From 1 January 2026 the submission deadline is five working days from issue, previously five calendar days.

• Bundesministerium der Finanzen — German e-invoicing mandate ↗

  Enacted by the Wachstumschancengesetz. All businesses must be able to receive EN 16931-compliant e-invoices from 1 January 2025; issuing is mandatory from 1 January 2027 for businesses above €800,000 turnover and from 1 January 2028 for all others. Accepted formats include XRechnung, ZUGFeRD 2.1+ and Peppol BIS 3.0. Exchange is decentralised — there is no central clearance platform.