Translation Agency Software: Custom Build or Buy Plunet and XTRF

What Plunet, XTRF and the computer aided translation tools actually do well

It is 4:40 on a Friday and a client has sent eighteen files: an InDesign package, two XLIFF exports, a regulatory annex as a portable document format and a folder of subtitle files. Your project manager runs the analysis, opens the management system, then opens the spreadsheet where the actual business lives. Before you price a build, be clear about what the products already give you.

Plunet and XTRF are mature translation business management systems. They hold clients, quotes, jobs, purchase orders, invoicing and a vendor database, they connect to accounting, and they cost a fraction of a build. memoQ, RWS Trados Studio, Phrase and XTM produce the weighted analysis your whole pricing model rests on, hold translation memory and termbases, and run mechanical quality checks on tags, numbers, terminology and consistency. Rebuilding any of that would be a poor use of your money and we would talk you out of it.

If you run under about 100 jobs a month, work in fewer than eight language pairs, or serve a single vertical with highly repetitive content, keep buying. At that size the routing intelligence genuinely fits in one competent vendor manager's head and the spreadsheet beside the system is not a crisis. It becomes one later.

Where they stop: the vendor database is a list, not a ranking

Your management system holds language pair, rate and a specialisation dropdown. What your project manager actually decides on is none of those. This linguist is excellent on technical content into German and slow on marketing, has never missed a deadline under 5,000 words but missed two above that, is meaningfully cheaper than the fallback, and the client's in country reviewer in Osaka likes her. Plunet and XTRF let you filter. They do not rank, because ranking needs structured outcome data that nothing currently writes back.

The consequence is quiet and expensive. When your senior vendor manager takes two weeks off, throughput drops and quality complaints rise, and everyone calls it a bad month. It was not a bad month. The routing intelligence walked out of the building.

The second gap is the one that costs real margin every single job. Your computer aided translation tool produces a weighted analysis across repetitions, full matches, high fuzzy bands and no match segments. Your client contract applies one grid. Your linguist agreement applies a different one. Nothing reconciles them per job. When a client negotiates a better fuzzy grid, agencies almost never renegotiate the vendor grid to match, so the discount comes straight out of margin, invisibly, on every job, for years. That is the single most common silent leak in this category and most operators find it only when someone finally computes margin per job.

Third, quality. A reviewer sends back a complaint about a batch and you have no counter evidence, so you credit the invoice or lose the account. Your tools catch mechanics. They do not produce an auditable record tying an error to a linguist, a job and a client standard, classified against your error typology. The distinction between a genuine terminology error and a reviewer's stylistic preference is worth real money, because preferential changes are not vendor errors and should never justify a credit or count against a linguist's score.

Custom versus off the shelf: the arithmetic per job

Price this per job, because per seat pricing hides the thing that matters. In Digital Heroes delivery experience a focused first release runs $60,000 to $130,000 across 12 to 16 weeks. Midpoint $95,000, plus migration and four years of support, puts five years near $179,000, so roughly $36,000 a year.

At 200 jobs a month, that is 2,400 jobs a year and the build costs about $15 a job. Now price one instance of the leak. A 6,000 word job where the vendor rate rose three months ago and the system still holds the old figure, at a gap of three cents a word, costs you $180. You need one of those every twelve jobs for the build to pay for itself, and if you have never reconciled client and vendor grids per job, you have considerably more than that.

So the crossover sits somewhere between 120 and 200 jobs a month, and it arrives earlier when your client grids have been renegotiated and your vendor grids have not. Below 100 jobs a month the arithmetic does not work at any grid, because the leak is smaller than the build and a careful vendor manager catches most of it.

One more figure belongs in the comparison and nobody computes it. Take the hours your project managers spend chasing availability by email on a typical week and cost them properly. Sending nine availability messages and getting six answers by morning is a 26 hour turnaround that a ranked offer sequence turns into 90 minutes at five in the morning with nobody awake.

What a custom build actually costs, including the migration nobody scopes

A focused first release covering the assignment engine, margin per job and a linguist portal, running alongside your existing management system rather than replacing it, runs $60,000 to $130,000 in 12 to 16 weeks. A full platform adding an intake portal, computer aided translation integrations, quality assessment capture, vendor payments and finance synchronisation runs $150,000 to $400,000 phased over 6 to 12 months.

Data migration runs 10 to 25 percent of the build and it is where these projects quietly go over. Pulling eight years of job history cleanly out of Plunet or XTRF is real work, and it cannot be skipped, because that history is exactly what the scoring engine depends on. An assignment engine with no outcome data is a filter with extra steps.

Year two and each year after runs 15 to 20 percent of build cost annually. In this sector that funds integration drift more than anything else: memoQ, Trados, Phrase and XTM each have their own interface personality, Trados behaves differently on premise from its cloud offering, and any of them can change under you.

The line that arrives around month seven is file preparation. Clean InDesign handling and scanned document preparation is genuine engineering rather than a library call, and multi currency vendor payments across dozens of countries with tax handling is a second project hiding inside the first. Scope both explicitly or watch them appear as change orders.

The four situations where building wins

Regulatory and certification fit. If you hold ISO 17100 certification, the workflow has to produce the audit trail as it runs rather than by a person assembling it before a surveillance visit. Data residency is the sharper issue: your linguists are contractors in a dozen jurisdictions, your enterprise clients will audit where content sits and who processed it, and a data processing agreement that does not cover a particular country is a real commercial problem. Generic tools treat this as a checkbox and your first enterprise security review will find that out.

Scale economics. Past roughly 120 to 200 jobs a month, the per job cost of a build drops below the per job cost of the leak.

A workflow that is your competitive advantage. Routing is the value you add. The linguists produce the words. If picking the right one at the right price inside the right hour is what clients pay for, that logic deserves to be an asset you own rather than a relationship one person holds.

Integration sprawl across three or more systems. Two or three computer aided translation tools with different interfaces, an accounting system, a payment provider covering dozens of countries, and a client asking for an interface into their own release pipeline. Localisation buyers who can trigger jobs from their own build process do not leave.

How to decide in a week, then commission a written specification

Take last month's twenty largest jobs and compute actual gross margin on each by hand. Quoted revenue against the client grid, actual vendor cost against the vendor grid applied to the same analysis, plus rush handling and file preparation time. Do not estimate. Open the analysis logs.

Three things fall out and all three are useful. First, the spread, which is usually much wider than anyone expects and often includes at least one job delivered at a loss. Second, the accounts you have been subsidising, because margin by client is the view nobody has and operators routinely find one or two running far below what they believed. Third, the hours the exercise took, which tells you what a month end costs you today.

If the spread is tight and the exercise took an afternoon, keep your management system and negotiate your renewal with the number in hand. If you found an account running at a third of the margin you assumed, that single finding usually exceeds the annual cost of the build.

Then buy a specification rather than a build. At Digital Heroes that means a signed product requirements document covering the job, language task and step model with independent state, the grid reconciliation rules, the integrations by name and version, and acceptance criteria, written before any code exists. It is yours whether or not we build, so take it to three other firms and get quotes priced against the same document.

We are the wrong firm for you if you want a translation memory engine or a machine translation model built. Those exist, they are good, and you should buy them. We are also wrong if nobody will own the error typology, because a quality record nobody maintains is an opinion with a database behind it. What we bring is more than fifty specialists, over 2,000 projects delivered, our own products including ShopScore and HeroCheckout so we carry systems past launch, and India LLP, United States LLC and United Kingdom LTD entities so intellectual property assigns under law your advisers already read. Clutch, Trustpilot, Fiverr Vetted Pro and our D-U-N-S listing are all public.

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