Spellbook legal AI: what it does, what it costs, where it stops.

Spellbook is a product, calibrated against the largest customer in the category, with a buying model that pays for itself for operators whose workflow matches the calibration target. The strongest use cases are the horizontal tasks the product was built around: research, drafting, review, lookup, summarization. For those tasks, on data the product was trained against, the output is competitive with bespoke work at a fraction of the up-front engineering cost.

For an operator whose workflow is well-aligned with that calibration target, Spellbook is the right buy. The pricing is predictable. The on-ramp is fast. The roadmap is funded. The category is moving and the product will move with it.

The misfit shows up when the operator's workflow is not the horizontal task the product was built around. For law firms that workflow is some specific combination of intake to matter routing, conflict checks, document automation, matter-to-template matching, timesheet reconciliation. The product, calibrated against the average customer, will get thirty to forty percent of the way to that workflow before the operator-specific gap opens up: a matter taxonomy (the way a firm categorises its cases) the product does not know, a part library the product cannot represent, a carrier pool the product cannot reason about, a dispatch logic the product cannot follow.

The commissioned build closes that gap by being built on the operator's actual data, inside the operator's actual stack (iManage, NetDocuments, Clio Manage, Litify where relevant), with the operator's specific workflow as the calibration target. The trade-off is up-front cost (one fixed fee from $10,000) versus ongoing SaaS (software you rent by subscription) subscription. For operators with a known constraint and a five-to-ten-year horizon, the math favors the commission.

Vertical fit. Spellbook is calibrated for the average customer in the category, which for most product companies is the largest end of the market. ColabContent commissions are calibrated for the specific operator. Mid-market operators are not the average customer.

Custom versus product. Spellbook is a product with configuration knobs. ColabContent commissions are custom code, custom prompts, custom data pipelines. Configuration cannot represent what custom code can represent.

Ownership. Spellbook retains the code, the models, and the data pipeline. ColabContent transfers all three to the operator at handoff. The operator owns the build, can modify it, can run it indefinitely without a vendor relationship.

Pricing model. Spellbook prices by the number of team members on the license, for as long as the firm subscribes. ColabContent charges a fixed fee in two installments, one at production-build start and one at handoff. Total cost of ownership over five years usually favors the commission for law firms.

Time to working system. Spellbook is fast to provision but the operator-specific workflow build sits outside the product timeline. ColabContent ships a working prototype on the operator's real data in seven to ten days and a production system in four to seven weeks.

Reference depth. Spellbook has the larger published reference set, weighted toward larger customers in the category. ColabContent's references are smaller in number but matched to mid-market law firms and named with numbers.

Pick Spellbook if the operator's workflow is the horizontal task the product was built around, the seat count is small enough that per-seat pricing pencils, the operator is comfortable not owning the code, and the operator does not need integration with a specific stack that the product does not natively support.

Commission custom if the operator has a specific workflow that the product calibrates against, the budget exists for a custom build from $10,000, ownership of the code matters, and integration with the existing stack matters more than vendor brand.

Many operators end up with a hybrid posture: Spellbook for the horizontal tasks where it dominates, a commissioned build for the operator-specific workflow where it does not. We have shipped commissions that explicitly call Spellbook as one of their downstream components.

Operators who already have Spellbook in production and are considering supplementing it with a commissioned build face three migration questions: which workflows stay on Spellbook, which move to the commissioned build, and what the integration boundary looks like between them. The right answer is rarely "rip and replace." The right answer is usually "keep Spellbook where it wins, build custom where it loses, integrate cleanly at the boundary."

The audit call works the same way for hybrid postures. We will tell the operator honestly which workflows are right to leave on Spellbook and which are right to commission. The report is yours to keep regardless of the outcome.

Most trials fail as experiments because nothing was defined as a pass. If the firm takes the 60-day Spellbook recommendation above, run it as a measured pilot rather than a license giveaway. Pick five live matters that represent the contract work the firm actually bills, not the cleanest ones. Name one attorney as the owner of the evaluation, and have every participating drafter log two numbers per document: minutes to a reviewable first draft, and how many of the suggested redlines survived partner review. Sixty days produces enough documents to see the pattern.

Set the exit criteria before the trial starts. If first-draft time falls meaningfully on standard paper and the team keeps the add-in open without being reminded, buy the seats and stop; that is a good outcome, not a consolation prize. If the numbers barely move because the expensive hours were never in drafting, the trial has still paid for itself: it has located the constraint somewhere the product cannot reach, which is exactly what the Billable-Hour Recovery Diagnostic is built to price. Run the diagnostic inside the same sixty days, and the comparison on this page turns into a decision specific to the firm.

Spellbook does not publish a price. As of September 24, 2026, its pricing page quotes every team individually: the price is set by the number of team members on the license, and the figure comes from a demo call. What it does publish is a 7-day free trial, longer trials for larger organizations, and free access for law schools. The page lists two packages. The law firm package covers the Word add-in and Associate, the multi-document agent, with group training and a self-serve trial. The in-house package adds a Playbook Build Service and starts with a demo. Teams over 10 get dedicated support on both. Because there is no list price, ask the demo for three numbers in writing: the per-seat annual price, the price at your five-year headcount, and the renewal terms. The figure that matters is the per-seat price times the lawyers who will actually draft in it, not firm headcount.

It lives in the Word ribbon on the document the lawyer already has open, drafting clauses, suggesting redlines, flagging missing or one-sided terms, and benchmarking language against common market positions. Nothing asks the lawyer to leave the document or upload the deal somewhere else, which is the whole reason it gets used.

Not really, and that is a design choice rather than a defect. It is a transactional drafting tool. Litigation workflow, matter intake, discovery review, and anything that has to know your firm's matter taxonomy sit outside what a Word add-in can see.

For contract drafting inside Word, several assistants compete on the same surface, and Microsoft's own Copilot covers part of the job for firms already paying for it. For a wider legal AI need the comparison set moves to platform tools. For firm-specific workflow the alternative is not a product at all.

For a transactional team whose bottleneck genuinely is first-draft speed, yes, and buying it beats commissioning anything. Firms should be honest about that. Most firms that outgrew it did not outgrow the drafting; they discovered drafting was never the expensive part of the month.

When the constraint is the firm's own operation rather than the document. One anonymized law firm runs matters, invoices, and IOLTA (the trust account a law firm must keep client funds in, separate from its own operating funds) trust on a commissioned platform: 13,296 matters, 4,396 clients, 5,684 invoices, trust reconciled byte-identical. No drafting add-in gets near that work. The engagement is a fixed fee and the firm owns the code at handoff.

Vendors in this category publish data handling terms and those terms change, so read the current data processing addendum rather than a summary, and have whoever signs the malpractice policy read it too. The question to ask in writing: is any firm content used to improve a model that other customers can reach.

A firm whose contract volume is real, whose lawyers live in Word, and whose operations are otherwise running fine should buy seats and stop. Commissioning a build to solve a drafting problem a product already solves is how firms waste six figures. A build is for the problem no vendor is shipping.

Yes, and it is often the bigger money. Jim Glaser Law runs 3,787 AI-handled calls and 5,514 minutes across five channel-specific agents covering PPC, organic, TV, Meta, and LSA, which gives the firm per-channel attribution on calls that were actually answered. That is an intake and marketing problem, and no drafting add-in touches it.

Four things, in writing. Where the data goes, and whether it trains a model other customers can reach. Whether the audit trail shows which documents an answer used, and for which user. Whether the tool respects existing permissions and ethical walls. And what the price is at your five-year headcount rather than today's. Most buying mistakes come from skipping the fourth.

The audit fee is refunded if you feel you got no value from it, no questions asked. For a build, the working prototype is proven on your own documents before any build fee is due, so the firm is deciding on a system already shown to work on its own files.

For the audit: the firm's current matter-management workflow, document management system, and highest-volume document types. For a build: access to the relevant systems and one named person who owns the workflow being automated; the working prototype itself ships in 7 to 10 days.

No. Every system ColabContent commissions is scoped to absorb a specific bottleneck past the document edge, such as intake, triage or billing, not to replace attorney judgment. Staffing decisions stay the firm's call.