Build, buy, or keep outsourcing records review

Every firm doing records work at volume has three options, and most firms have never compared them on the same page.

Keep outsourcing. Buy a product. Build something.

They are priced in different units, which is most of why the comparison does not happen: one is dollars per page, one is dollars per seat per month, and one is a project. Units that do not match are units that do not get compared.

Here is the arithmetic, and an honest answer about who should pick each.

What each one actually costs

Outsourcing. Reported figures put a 1,200-page review at around $4,500. Retrieval is separate and runs roughly $0.25 to over $1.00 per page depending on the state, before the records custodian work even starts. About 47% of firms outsource at least some records work, which tells you it is a reasonable answer rather than a failure.

The cost scales with volume, which is the point in both directions: it costs nothing in a slow quarter and it costs a great deal in a busy one.

Buying. Legal AI products for records work are typically priced per seat or per matter. The number is knowable and the sales process will tell you. What the sales process will not tell you is the cost of the mismatch — the parts of your process that have to change because the product expects them to work another way.

That cost is real and it does not appear on the invoice. Sometimes it is small. Sometimes it is the reason the tool is abandoned in month four.

Building. A single well-scoped workflow runs $60,000 to $90,000, with $75,000 the common middle, plus 15 to 25 percent of that annually to keep it working, which is the part quotes leave out. Over three years a $75,000 build lands between roughly $109,000 and $131,000.

In-house, by hand. Not free, just already spent. A routine treating file costs a day and a half of a paralegal’s time building the chronology; a file with several providers and a year of records costs three or four days. Price those at a fully loaded rate, multiply by your annual case count, and you have a number most firms have never written down.

The number everyone leaves out

All four of the above are the cost of doing the work. None is the cost of not being able to do it.

Seventy-five percent of firms report operating at or beyond capacity, and 42% say case volume regularly exceeds what the team can handle. That constraint does not show up as an expense. It shows up as a partner deciding, quietly and reasonably, that the firm cannot take on another treating case this month.

That declined case is worth more than everything else on this page, and it is in nobody’s budget.

Which is why the honest version of this comparison has four lines, not three:

1. What you pay vendors annually for records work.
2. What in-house records time costs annually, at a loaded rate.
3. What a build plus three years of support would cost.
4. What you turned away because the work was queued.

Most firms can fill in the first three in an afternoon, and the calculator does the first three for you. The fourth requires an uncomfortable conversation and is usually the largest.

When to buy

Buy when a product fits your process without you reshaping the firm around it.

This is the right answer more often than anyone selling builds will admit. A product is cheaper, it is supported by a company whose entire business is that one thing, it improves without you paying again, and it is available this afternoon rather than in six weeks.

The test is not whether the product is impressive in a demo. It is whether you would still use it in month six, given how your firm actually works. The way to find out is a trial on your own records — not the vendor’s sample file — with the people who will actually use it.

If the product does the job, buy the product. We say so on the record because it is true and because the alternative is selling somebody a build they will regret.

When to keep outsourcing

Keep outsourcing when your volume is modest or lumpy.

These are fixed costs recovered over cases. A firm running a caseload that does not clear the arithmetic cannot get there through enthusiasm, and a vendor who charges you per page in a slow quarter is doing you a favor you would not get from a build.

Outsourcing also has a real advantage nobody mentions: it is somebody else’s problem to staff. Records work is hard to hire for and harder to retain, and paying a vendor to absorb that is worth something on its own.

When to build

Build when the value is in connecting systems you already run.

That is the specific case, and it is narrower than the market suggests. Your case management system, your document store, your accounting, your intake tool, and the spreadsheet that actually runs the firm. Nobody sells the layer between those, because it is different in every building — which is exactly why it is the one thing a product cannot do for you.

The second reason to build is ownership. At the end you have source code in your repository, running in your cloud, with no seat licenses and nothing to renegotiate. Whether that is worth the premium depends on how long you plan to be doing this.

If your annual volume multiplied by the hours saved per case does not clear a build comfortably, buy something off the shelf. Comfortably means with room to be wrong about your own numbers.

What actually goes wrong

Not the technology. In our experience and in the published failure data, three things in order:

Scope. A project that starts as one workflow and becomes four does not take four times as long, it takes longer than that and often does not finish. One workflow, shipped and adopted, is worth more than three that are nearly done.

Integration. Whether your case management system has an API you can write to, or whether it has a login and a support address. This single fact can move a project by weeks and is knowable in an afternoon. Any estimate produced without checking it is a guess.

Adoption. A system that is accurate and that nobody opens is worth nothing. Software gets routed around when it is built for whoever signed the contract rather than whoever touches it forty times a day, and nobody tells the partner who chose it.

Note that two of those three are decided before any code is written.

Start smaller than you think

Whatever the spreadsheet says, there is a cheaper first move than a full build.

Take one document type, or intake alone, and do that. It costs a fraction, it delivers in weeks, and it answers the question the arithmetic cannot: will the people at this firm actually use the thing you made.

Adoption is the risk that kills these projects, and it is much cheaper to test than to assume.

If the small thing gets used, the arithmetic on the big thing is no longer theoretical. If it does not get used, you have learned that for a fraction of the price, and the answer to build-or-buy turns out to have been neither.