Sales Tax Compliance Software: Custom Build Versus Avalara and Vertex

What Avalara, Vertex and Anrok actually do well

It is the tenth of the month. The billing export, the trial balance and the returns workbook do not tie, a state has sent a notice about a period that closed fourteen months ago, and returns are due in eleven days across dozens of jurisdictions. Start from what the market already handles.

Avalara and TaxJar calculate rates and file returns competently, and their content teams track rate and rule changes across thousands of jurisdictions so you do not have to. Vertex and Sovos serve larger enterprises with deeper content and stronger enterprise resource planning (ERP) integration. Anrok is built for software sellers and understands digital taxability properly. Thomson Reuters ONESOURCE fits multinationals with indirect tax obligations beyond the United States. Stripe Tax is enough for a lot of early stage sellers.

Buy and stop there if most of this is true. One storefront or one billing platform with a stable schema. One order per document. A customer with an address and a product with a code. Physical goods or straightforward digital products. No usage based pricing rated at period end. No bundles mixing taxable and exempt elements. Under those conditions the connector works as designed and a build would be money spent on a problem you do not have.

Say the strong version of this. Nobody should build rate content. Maintaining rates and rules yourself produces an expensive product that is permanently out of date, and any developer proposing it has not filed a return.

Where they stop: your order source, your bundles, your certificates

Every product above assumes a clean transaction source. Three things break that assumption.

Your revenue does not arrive as one order, one document. Metered usage rated at period end, bundles where a single item contains taxable and exempt elements, credits modelled as negative lines on a future invoice, partial refunds, bad debt write offs, intercompany transactions between entities. The connector becomes a mapping exercise that somebody maintains in a spreadsheet, and that spreadsheet quietly becomes the compliance function.

Taxability of what you actually sell is a determination, not a lookup. Software delivered electronically, software as a service, a data service, a digital good, professional services bundled with a licence, training, support. Treatment varies by state and can hinge on whether the customer takes possession and whether the service is separately stated on the invoice. Engines hold good content per category. The gap is mapping your catalogue into those categories, and when the right answer is to unbundle on the invoice, that is a billing change before it is a tax change. Store the determination with its reasoning and effective date, because an auditor examining 2023 asks about the decision, not today's setting.

Exemption certificates expire quietly. A resale certificate photographed by a salesperson in 2021, emailed, filed in a shared drive, covering one state, since lapsed. This is the single most common source of assessment, because the auditor samples exempt sales and asks for the certificate, and every sale you cannot support becomes tax plus interest plus penalty. Checking a certificate at invoicing rather than at audit is the whole game.

The arithmetic: per transaction pricing versus cost to build

Engines are priced per document processed, with a platform fee and a separate charge per return filed. The figures we see modelled land between roughly $0.05 and $0.20 per invoice document, plus $25 to $50 per return, plus an annual platform fee.

Run five years. At $0.12 a document, one million documents a year costs $120,000 and $600,000 over five years. A build replacing determination entirely and landing at $325,000, plus $60,000 to normalise and migrate transaction history, plus about $55,000 a year to run it, totals near $605,000. So the crossover for full replacement sits near one million documents a year.

Almost nobody should act on that number, and here is why. Crossing it means taking on rate content maintenance forever, which is the one part of this problem where a vendor's scale genuinely beats yours. The hybrid is the honest answer: keep the engine, and build only the layer your business actually breaks on. That layer prices against a much smaller number, and it is the part you will still own when you change engine, billing platform or filing provider.

What the custom layer actually costs

A focused first release covering transaction normalisation from your billing platform, taxability mapping with stored determinations, nexus threshold monitoring with alerting, and reconciliation from filed liability back to invoice line runs $70,000 to $150,000 and ships in 12 to 18 weeks. A full platform adding certificate lifecycle management, return preparation and filing workflow, registration and notice management, and value added tax or goods and services tax handling for other regimes runs $200,000 to $450,000 phased across 6 to 12 months.

Two lines nobody quotes. Normalising and migrating transaction history runs 10 to 25 percent of the build, and it reaches the top of that range when your billing platform has changed its schema twice and credits were modelled differently before 2023. Year two onward runs 15 to 20 percent of build cost annually, which here mostly buys keeping up with your own product changes rather than tax changes, since the engine handles those.

What pushes it up: the number of source systems, since each billing or order platform is its own normalisation problem. Usage based and bundled pricing, where taxability mapping becomes genuinely difficult rather than clerical. Multi entity structures with intercompany flows. International scope, because value added tax with the One Stop Shop and Import One Stop Shop regimes, and country electronic invoicing mandates such as Italy and India, are separate projects rather than extensions. What holds it down: normalisation and reconciliation first, certificates second, filing workflow last.

The four situations where building wins

Regulatory fit. Since South Dakota v. Wayfair in 2018, economic nexus is a rolling calculation rather than a status, commonly around $100,000 or 200 transactions, with several states dropping the transaction count and others measuring on different bases and periods. Home rule jurisdictions in Colorado, Louisiana and Alabama administer their own local returns. Twenty four states follow the Streamlined Sales and Use Tax Agreement and accept its exemption certificate form. Computing thresholds on your own normalised ledger, with an alert well before crossing, is what turns an exposure into a registration.

Scale economics. Above the crossover in the arithmetic above, though in this category the labour case usually arrives first. Count the days a month your tax manager spends reconciling before you count documents.

A workflow that is your competitive advantage. The tie out. A single normalised transaction ledger between source systems and everything downstream, with both the return and the general ledger posting produced from that one place, so reconciliation is a report rather than an investigation. Build the drilldown from a filed figure back to an invoice line before you build anything else. It is what an auditor asks for and what your own team needs at eleven at night on the tenth.

Integration sprawl across three or more systems. Billing, the general ledger, the tax engine, the customer record holding certificates, and the filing provider. Physical presence triggers make it worse, because a remote hire in a new state or inventory placed in a third party warehouse are human resources (HR) and logistics events that no transaction feed knows about.

How to decide in a week

Two tests. Both are unpleasant and both are decisive.

Monday and Tuesday: take last month's filed liability for your three largest states and drill each one back to invoice lines. Not to a summary report, to lines. Record how long it takes and how many adjustments could not be explained without asking a person.

Wednesday: pull 25 exempt sales at random from the last two years and try to produce a valid, unexpired certificate for the right jurisdiction for each. This is exactly what an auditor does. Most teams find between three and eight they cannot support, and that number converts directly into tax plus interest plus penalty.

Thursday: list every state where you crossed 60 percent of the economic nexus threshold in the last twelve months but are not registered. If nobody can produce that list from a system, that is the finding.

Friday: decide. Clean tie out, certificates all supportable, nexus list produced in an hour means buy the engine and connect it properly. Any other result means the layer is worth building, and you now have its scope written for you.

Then pay for discovery rather than accepting a free estimate. At Digital Heroes that produces a signed product requirements document before any code is written: the normalised transaction model, the taxability determination store with reasoning and effective dates, the nexus calculation per state, the certificate lifecycle and the reconciliation contract, plus acceptance criteria. You own that document whether you build with us or take it to your existing provider and require them to meet it.

Who we are wrong for: a single storefront selling physical goods, and anyone hoping to replace rate content, which we would advise against building. We fit sellers with homegrown or heavily customised billing, usage based or bundled pricing, and multiple legal entities. More than fifty specialists, over 2,000 projects delivered, and a named team you meet before signing. Our India LLP, US LLC and UK LTD entities mean the intellectual property assigns under your own law, which matters because tax positions have long tails. Checkable on Clutch, Trustpilot, Fiverr Vetted Pro and D-U-N-S, and our own products ShopScore and HeroCheckout came out of commerce work like this.

Book a 30-minute call with Digital Heroes and get a written plan and a fixed quote within 48 hours.