CoCounsel LegalProduct · Legal AI
An AI assistant for legal research and contract review, with Knowledge Search to find sources and Tabular Analysis to compare terms.
Thomson Reuters
Making legal AI ready for enterprise use.
Thomson Reuters builds software and information services for legal, tax and business teams. From 2024 to 2026, I led four PMs within a 27-person team and managed a US$12M annual product and delivery budget. My scope covered search, shared integrations, answer quality and enterprise adoption across CoCounsel and the wider corporate portfolio, turning client research into release priorities and funded rollout commitments.
Product Leader2024–2026
Products and portfolio in my scope
ONESOURCE+Portfolio · Corporate compliance
A corporate portfolio linking tax, trade, legal and risk products. My scope covered shared foundations and delivery for legal clients.
01What was the friction?
A checked answer took 21 minutes across separate systems.
Legal reviewers needed to find an agreement, check its terms and verify the supporting clause before advising a colleague. Documents and guidance sat in separate systems, while product teams repeated the work to connect them. We timed the complete task and reviewed answer failures alongside the four product plans. That diagnosis established three requirements: shared access, verifiable answers and a release scope the team could support.
Time the task through to a checked source.
A supplier review crossed agreements, internal guidance and legal reference material. Across 46 teams in eight organisations, we timed research from the first question to the reviewer checking a supporting source.
- 01
Find the agreement
iManage · Client records
- 02
Check the guidance
Internal & legal sources
- 03
Verify the clause
Supporting source text
21 minMedian for the complete taskStarting point
- 01
Search together
Connected repositories
- 02
Open the source
Clause and document link
- 03
Confirm access
Existing user permissions
≤13 minTarget for the same complete taskRelease goal
Find the work repeated across product plans.
Search and document analysis needed the same connections and permission checks. I mapped these dependencies with the four PMs to identify the shared access work that would support both workflows.
| Shared work | Search | Document analysis |
|---|---|---|
| Document connections | Required | Required |
| Permission checks | Required | Required |
One integration owner for the access both workflows needed
Translate answer failures into release criteria.
Specialists checked terms, exceptions and sources across 1,250 fixed review cases. They accepted 963 answers, then classified the remaining failures so the PM could prioritise corrections before release.
963 / 1,250Answers accepted in the first review
287Cases classified for correction
Correct terms
Relevant exceptions
Supporting source
02What approach and priorities did I choose?
Funding shared access before expanding the workflows.
The discovery work showed that faster research depended on shared document access and reliable source checks. I funded those dependencies first, deferred additional workflows and agreed a 13-minute research target alongside 95% expert acceptance. The four PMs then sequenced their work around those requirements, while product and commercial leads defined the client ownership, training and staffing needed to turn each pilot into a paid deployment.
01 / FirstShared dependencies
Build the common access once
Repeated connection work made shared access the first priority. I deferred dependent workflows until the common foundation was ready.
02 / RequiredRelease confidence
Make a checked answer the standard
Source checks, repository permissions and expert review defined release readiness, keeping each answer tied to its legal context.
03 / ExpandClient adoption
Agree a rollout the client can use
Each pilot needed an operating plan. I aligned ownership, training and rollout staffing with commercial leads and the agreed paid scope.
The complete client task determined the sequence; benefit, effort and existing commitments determined the capacity.
- 01Baseline
Define the complete task
Time the research task and classify answer failures to establish a starting point for the search and quality PMs.
46 teams · 1,250 review cases - 02Shared access
Resolve common dependencies
Agree one shared access plan across products and check repository permissions before adding dependent workflows.
Named integration owner - 03Release review
Retest the supported tasks
Check sources, terms and exceptions against the fixed review set, then agree which tasks can reach more users.
Target: ≥95% acceptance - 04Client rollout
Commit funded capacity
Agree a client owner, training and delivery responsibilities before committing features and staffing to paid use.
Target: ≥5 paid deployments
The 2026 Thomson model evaluation used the same release discipline, with a separate set of paired comparisons.
03How did I lead the team?
By giving each PM clear ownership within a shared plan.
Delivering this sequence required four PMs to work from the same capacity and client commitments. Each owned a product area and its measure; I coached their decisions and settled changes affecting another team, the budget or a promised release. For two plans needing the same connection, I mapped shared dependencies with the PMs and agreed the owner and release order. Weekly evidence reviews and monthly staffing decisions kept that plan current.
Felipe Löwe
Product Leader
Portfolio priorities, the US$12M budget and client delivery commitments sat with me. In one-to-one reviews, PMs tested their recommendation against the evidence, client benefit and consequences for other teams.
Search PM
Owned the complete research journey, using task timings and client feedback to prioritise search improvements.
Integrations PM
Coordinated document connections and permission checks across the product plans that depended on shared access.
Answer quality PM
Turned expert feedback into corrections and release evidence, keeping unresolved cases with a named owner.
Client adoption PM
Agreed the client owner, training and paid scope with commercial leads, then coordinated the implementation plan.
Weekly evidence reviews
Research time, expert reviews and client readiness guided the next decision. PMs brought evidence, a recommendation and any unresolved dependency.
Monthly capacity reviews
I reviewed staffing and spending against the US$12M allocation, settling competing requests with functional and commercial leads before committing.
How the 22 specialists supported delivery
Product engineering8
Applied AI3
Product design2
Data2
Legal & tax expertise4
Delivery & commercial3
27 people in total: my role, four PMs and 22 shared specialists. Staffing was agreed with functional leads; my sponsor was the VP of Product, Corporates.
A shared dependency
Two plans needed the same connection.
The PMs brought the client requirements and delivery options into one review. Funding the common access work first allowed both product plans to use the same foundation.
- FirstShared document accessOne integration owner · Connections & permissions
- NextDependent product workflowsSearch and review build on the agreed access
What each person could decide.
- Product PMs
- Recommended the sequence and managed changes within the agreed product scope.
- My decision
- Allocated people and budget, and agreed changes to client commitments with commercial leads.
- Delivery leads
- Confirmed staffing, release checks and ongoing support before the revised plan was promised.
In one-to-one reviews, PMs tested what would change their recommendation, then led the discussion with peers and clients.
04Which bottleneck shaped the delivery plan?
Repeated integration work held up dependent releases.
Search and document analysis needed the same connections and permission checks, drawing repeatedly on shared capacity. I funded one integration plan, moved dependent features behind it and aligned the four PMs around a common sequence for validating the complete client task before wider release.
05How did we deliver and validate the product?
By making every answer easier to inspect and verify.
The requirements shaped our work on two CoCounsel capabilities: Knowledge Search, introduced in 2025, connects relevant sources, and Tabular Analysis compares contract answers with their supporting evidence. The reconstructions below show source inspection and review; separate tabs cover release checks and the 2026 evaluation of Thomson, the company’s own AI model, which informed the supported tasks and capacity for the client rollout.
Workflow examples
Release evidence
Find the source behind a contract term.
Supplier agreements with termination provisions
The executed agreement supplies the clause wording, and the playbook and Practical Law guidance frame the review; Tabular Analysis then compares the notice period and termination right with their evidence, so the reviewer can check each term and any qualification before recording the source review.
Compare supplier contract terms.
Thomson modelTabular Analysis · Four example agreements
| Agreement | Notice period | Termination right |
|---|---|---|
| Master services agreementSource available | Either party | |
| Supply agreementSource available | Customer | |
| Distribution agreementCheck qualification | Either party | |
| Support agreementSource available | Customer |
How reviewers tested the complete task.
Client walkthroughs followed a question through search, source inspection and the final checked answer. The PM used failed steps to refine the result, document link and treatment of exceptions before the next release.
Find the answerRelevant documents are accessible
Inspect the sourceThe cited clause supports the term
Check another userRepository access is respected
Reviewers checked access to each document, the cited clause and relevant exceptions together, using the complete legal task to guide the PM’s next release.
Conditions required before wider use.
- Source evidence
- The answer opens the correct document and cited clause, giving the reviewer a direct way to inspect how the source supports the extracted term.
- Access permissions
- Search results and document links follow the reviewer’s existing repository permissions, so source inspection uses material available to that user.
- Expert acceptance
- Review terms, relevant exceptions and sources on the same fixed review set, using the agreed 95% acceptance target to decide which tasks are ready.
How we assessed the Thomson model.
For the 2026 introduction of Thomson in Tabular Analysis, five client groups completed 1,100 paired comparisons. Each comparison used the same documents and task, so specialists could examine the answer and the effort needed to correct it.
Same documents and questions5 client groups × 220 comparisons
Thomson outputTerms · Exceptions · Sources
Comparison outputTerms · Exceptions · Sources
Expert review + correction effort + task cost
The comparison made each answer’s review burden visible: specialists checked terms, exceptions and sources alongside the effort and cost of completing the client’s task.
What the evidence decided.
The evidence linked model choice to the work clients could complete, helping the PM and me select the first supported tasks.
- Initial scope
- I agreed the supported tasks with the PM, using the findings to align the first client rollout with correction and delivery capacity.
- Further expansion
- The PM carried each question and source into the correction plan, then retested the affected task before extending the client scope.
06What did the OKRs and KPIs show?
Faster research and stronger answers supported paid adoption.
The PMs compared research tasks, repeat answer reviews and signed agreements with the targets set before delivery. Median research time fell from 21 to 12 minutes, and experts accepted 1,200 of the same 1,250 answers after corrections. All six pilot clients signed for paid deployment, with US$18M in three-year expansion value. I used that evidence to allocate rollout capacity while retaining owners for the work and answers still requiring attention.
Less time per research task
Median task time · 46 teams in eight organisations
21 → 12 min43% reduction
Before 21 minAfter 12 min
240
Target ≤13 minutesTarget met
What the result changed
The median fell by nine minutes to 12, beating the 13-minute target. Alongside stronger answer quality, this supported wider rollout of the research workflow.
See the measurement details
Across 46 teams in eight organisations, the PM measured the median time from a research question to a checked supporting source over a six-month rollout cycle. It fell from 21 to 12 minutes: 42.86% less time, rounded to 43%.
More answers accepted by experts
Legal review · The same 1,250 cases in both rounds
77.04% → 96%+18.96 percentage points
Target ≥95% acceptanceTarget met
What the result changed
Experts accepted 237 more answers under the same criteria. The PM retained a source, task and correction owner for each of the 50 remaining cases.
See the measurement details
Legal specialists reviewed the same 1,250 cases for correct terms, relevant exceptions and supporting sources. Acceptance rose from 963 to 1,200 answers, or 77.04% to 96%: a gain of 18.96 percentage points after corrections.
Pilots converting to paid deployment
Six selected CoCounsel Legal client accounts
6 of 6paid deployments
010203040506
6 paid deployments0 remaining in pilot
Target ≥5 accountsTarget met
What the result changed
All six clients signed for paid deployment, one above target. I allocated capacity against each client’s access, training and implementation commitments.
See the measurement details
The twelve-month programme followed six selected CoCounsel Legal accounts from pilot to paid use. Each signed against an agreed scope and rollout plan. The programme began with no converted accounts and targeted at least five.
Signed enterprise expansion
Six agreements · Three-year total contract value
US$18MUS$3M above target
2.4
2.7
2.8
3.0
3.3
3.8
Three-year value per agreement · US$M
Target ≥US$15MTarget met
What the result changed
Six agreements beat the target by US$3M in three-year value. I agreed supported features and staffing with commercial leads before confirming the work.
See the measurement details
Six signed expansion agreements total US$18M in contract value over three years, against a US$15M target. The baseline of zero marks the start of this selected programme; revenue recognition and collection follow each agreement.
Research time and expert acceptance guided release scope. Paid adoption and signed value guided client commitments; repeated work and unresolved answers remained under review.
Product context & measurement notes
How the products fit together
CoCounsel Legal supports research and document review. Knowledge Search connects relevant sources; Tabular Analysis compares contract answers with source references. Thomson became a model behind Tabular Analysis in 2026. ONESOURCE+, introduced in 2025, is the wider portfolio linking corporate tax, trade, legal and risk solutions.
Reading the measures
The 46-team research study, 1,250-case answer review and six-account commercial programme have separate scopes. The model study used 1,100 paired comparisons across five groups. Repeated effort fell by 192 person-days, or 30.97%, over nine months. Expansion values cover three years; each result includes its period, definition and responsible owner.