An early example of where things are headed.

In Post 228, Paula Doyle, Chief Legal Innovation Officer at the World Commerce and Contracting Association (WorldCC), made the claim that inefficiencies in the current commercial contracting process likely cost the global economy more than $1 trillion annually. We reach this figure by adding up the massive second-order effects caused by excessive contract complexity and poor process:

  • Too many steps involving too many people (>50% transaction costs),
  • Excessive wait times (9.7-day difference in cycle time between average and top performers),
  • Buyers to give up and go elsewhere (abandonment rates of up to 57%),
  • Post-execution losses (average of 9% contract leakage due to poor contract management).

Obviously, contract optimization provides lawyers and law departments with a golden opportunity to impact the overall performance of the business.

Post (236) is a case study of how a group of industry leaders banded together to run a pilot exploring how they might use AI and Big Data to streamline the contracting process.

The consortium’s members come from all segments of the legal ecosystem and include corporate law departments and procurement groups from NetApp, Travelers Insurance, Liberty Mutual Insurance, Code42, and a large telecom company, legal tech/data companies (LexCheck, TermScout), law firms (Bryan Cave Leighton Paisner; Keesal, Young & Logan), a law company (Quislex), and a hybrid tech/law company (KP Labs).

Connie Brenton

Connie Brenton, VP of Law, Technology and Operations at NetApp, chaired the consortium.  TermScout, a contracts analytics company, did much of the heavy lifting in collecting and analyzing the dataset used in the pilot.   In all, over fifty people from more than a dozen companies/firms contributed to the pilot.

Disclosure: As Chief Product Officer of TermScout, I played a significant role in the design and execution of the pilot.

In Phase 1 of the pilot, the consortium collected and analyzed data regarding how contract negotiations happen today and used that data to explore where digital transformation opportunities might exist to streamline the process.  The data revealed significant opportunities to:

  1. Avoid negotiating certain contracts altogether,
  2. Employ an automated approach to negotiate certain matters,
  3. Limit the terms negotiated to those that present meaningful opportunities for changing risk allocation.

In total, there is a realistic possibility that the companies involved in this case study could reduce the cost of handling low-risk contracts by one-third or more.

Post 236 is organized into four parts.  Part I defines the problem—one that is likely familiar to many readers.  Part II summarizes the key data. Part III describes the opportunities identified for meaningful cost and cycle time reduction.  Part IV discusses some of the practical steps related to implementation.

I.  Problem: assume unknown levels of risk or endure delay?

To better understand the consortium’s impetus for running this pilot, it is useful to consider the following scenario that law departments and procurement organizations face daily.

You are one of many professionals in your organization who works on commercial contracts.  One of your business groups comes to you wanting to enter into a contract for $25,000 worth of SaaS services.  Although these services are very important to this particular business group, the overall risk to the company is low.  The business group, naturally, wants to conclude the contract as soon as possible so that it can get started.

The vendor has provided you with its standard terms and conditions.  The document, including all nested terms, runs to over 100 pages.  Your department is understaffed, and you have larger and more strategic agreements on your plate.

Data collected by WorldCC show that the typical cost for you to review and process the vendor’s paper will be $6,900 (and substantially higher if this were a higher-risk contract). See Post 228 (presenting WorldCC data). In this case, the cost of contracting for the service is more than 25% of the contract value. Moreover, the contracting cost does not include the impact of diverting your attention away from the larger and more strategic agreements, and the cost of the delays the business experiences in getting started.  Is it really worth spending this level of effort on a low-risk agreement?

Faced with this dilemma, companies commonly take one of two approaches.

  1. Sign “as is.”  As discussed in greater detail below, this means accepting an agreement that has a high likelihood of being unbalanced and may even contain terms that could negatively impact the broader business (broad rights for vendor use of customer’s data, imposition of exclusivity or non-compete, etc.).  Although you know that the likelihood of such provisions causing problems in practice is very low, everyone remembers that deal from years ago where a catastrophic failure occurred. You really do not want to be “that guy” who approved the contract.
  2. Full review and processing.  This is the option that will likely anger internal stakeholders. In addition to being costly, this can take weeks to complete, often bringing the related business project to a standstill.

Members of the consortium all believed that there must be a better way to handle these tradeoffs, particularly in a field that is embracing AI and big data.

II. Data collected and analyzed

The consortium ran Phase 1 of the pilot by assembling a database of 60 contract sets, involving a total of approximately 60 vendors and five buyers.  Each contract set is comprised of the final negotiated agreement and the starting document.  In most, but not all, cases the starting document is the vendor’s standard, publicly available terms and conditions.  All collected contracts involved software and/or SaaS offerings.

Each contract in the database was run through the TermScout tool to ascertain how the contract allocated risk between the parties on over 750 data points.  See Post 211 (providing Termscout example). The resulting data set provided a foundation for the consortium to generate a variety of significant insights as discussed below.

Phase 1 of the pilot sought to answer two key questions:

  1. Do current AI tools make it possible to efficiently and confidently identify incoming vendor contracts that can be accepted as-is without material adverse impact to the buyer’s risk profile?
  2. For contracts that must be negotiated, can current AI tools enable buyers to (a) reduce the time and cost associated with completing the negotiations, and (b) achieve better outcomes?

The data generated in Phase 1 of the pilot is promising on both fronts.

A. What percentage of vendor contracts might be acceptable as-is?

Multiple vendors have made significant efforts to make their terms reasonable. See, e.g., Post 211 (discussing platform-as-a-service (PaaS) providers). The consortium sought to test vendor IT agreements in the database to see how many of them were something that the consortium members, as buyers, might potentially be able to accept as is.

The first step in this analysis involved creating a filter of 19 deal-breaker provisions (example), the presence of any one of which would cause the consortium members to refer the contract for manual negotiation.  The filter used for the exercise represented the lowest common denominator of items that virtually all companies would consider to be deal-breakers.  (Buyers in the consortium each had a smaller list of additional terms they considered to be deal-breakers because of their individual business models, industry sectors, etc.  These additional items were not considered in this exercise.)  The consortium validated the reasonableness of the filter by requiring that each trigger item reflect a position that at least one-third of the top 100 software companies offer to any customer without negotiation.

Next, the consortium used the TermScout tool to apply the filter to each of the 60 starting-point agreements, generating a Report Card (example) showing how the contract performed on each trigger item.  Results are presented in Figure 1:

At one extreme, 17% of the 60 vendor contracts made it through the filter cleanly, making them strong candidates for potential acceptance as-is.

At the opposite extreme, 27% of the contracts failed on at least five terms, indicating that they likely would require full negotiation.

Perhaps most interesting are the 57% of contracts that failed on no more than four terms.  These contracts came relatively close to passing, and there appears to be a strong possibility that at least some of these vendors would voluntarily address these fail points if the buyer brought them to the vendor’s attention with credible evidence that the vendor’s current position was outside of market norms.  Certain members of the consortium intend to test an automated approach for generating these Self-Correction Notices (example) and seeing if these contracts can be closed without resort to traditional negotiations.

These results paint a picture of a significant opportunity for buyers to streamline their contracting processes and make their businesses more agile.  17% of all contracts have the potential, in low-risk settings, to be signed as-is.  Another 57% have the potential to be resolved entirely through automation.  Collectively, this represents an opportunity to eliminate contract negotiations in as much as 74% of the cases—all while adhering strictly to the company’s desired risk profile.

B. How much do negotiations matter?

Even if a company realizes all the above-described potential for negotiation avoidance, a significant number of contracts still are going to require negotiation.  Given these large volumes, even small improvements in the efficiency of the negotiating process stand to yield significant savings, create efficiencies, and drive value to the business.

Unfortunately, identifying opportunities for improvement always run into a brick wall. This is because the only data on the value—or lack thereof—that negotiations bring to the company are the subjective impressions of the lawyers and procurement professionals who do the negotiating.

In Phase 1, we wanted to uncloak this mystery.  To do so, the consortium sought to identify:

  1. How much negotiation occurs today?
  2. Which terms are getting negotiated?
  3. What are opportunities for improvement through negotiation?
  4. How successful buyers were when they negotiated?

(Ideally, these data would be coupled with data showing how often terms get litigated, and to what effect, but that falls outside the scope of the pilot.)

1. Number of Terms Being Negotiated

As a starting point, the consortium used the TermScout tool to compare each starting agreement with the corresponding final executed agreement to identify each point that had changed in a substantive manner.

On average, the parties were negotiating approximately 35 terms per contract.

This number likely understates the true amount of negotiation occurring in a potentially significant way in that it reflects only those negotiations that yielded a substantive change to a term and does not include cases where the parties negotiated the wording, but not the substance, of a provision.

2. Materiality of Terms Being Negotiated

Next, the consortium looked to identify what percentage of the negotiations involved terms that were clearly material, as opposed to items of secondary import.

For purposes of the pilot, the consortium treated all filter items as material (for obvious reasons) and non-filter terms as secondary.  (As mentioned above, in practice each company is likely to have its own filter that will include both the pilot filter terms and a smaller number of additional terms that it also considers to be deal-breakers.)  Results are presented in Figure 2:

Surprisingly, over 80% of negotiations involved non-filter terms.  While some of these terms likely were material, most were likely of secondary import.

Diving a bit deeper, the consortium identified the ten most negotiated provisions.

Any progress made on these points clearly will have an outsized impact on negotiation cycle times and costs.

3. Opportunities for improvement

In addition to assessing the materiality of the terms being negotiated, the consortium also looked at the starting point for each negotiated term, to ascertain how many opportunities for improvement existed.  Specifically, what percentage of the negotiations involved terms where the starting position was truly unfavorable as opposed to something that was already balanced or better?

Results are presented in Figure 3:

Over 43% of the terms negotiated had a starting point of balanced or better, meaning that the opportunity for improving one’s position in these cases was relatively limited.

4. Gains achieved through negotiation

Finally, the consortium looked at the extent to which negotiations enabled the buyer to improve the risk allocation associated with a term.  Using the TermScout tool, the consortium assigned a score of 1-5 to each provision, with 1 being most favorable to the vendor and 5 being the most favorable to the buyer.  We then used these scores to assess the amount of movement that negotiations achieved.

Results are presented in Figure 4:

Negotiations yielded positive results for the buyer in over half the cases.  In a third of the cases, the results were neutral, indicating that the buyer changed the term substantively but did not improve its position overall.  One example of this would be negotiating an indemnity provision and trading an offered indemnity for a different one of equal value.

Most interesting, however, is the fact that in over 11% of the cases negotiations caused the buyer to move backward rather than forward, netting a final position that was worse than what the vendor had offered in the first place.

We were unable to tell conclusively from the data why this occurred.  Possibilities include (a) negotiating out a clearly drafted initial provision and replacing it with a poorly drafted substitute provision that could be interpreted in a negative manner, and (b) a conscious decision by the negotiator to trade away a favorable starting position on one term to obtain a term important to the negotiator.

Either way, the message is clear that negotiations is an opportunity for things to go both backward as well as forward if one is not careful. And being careful takes time, which is an expensive resource in short supply.

III. Opportunities to improve the contracting process, ROI

Drawing upon the Phase 1 results, Figure 5 below summarizes the opportunities to improve the contracting process.

Let’s break down each part of the workflow:

Step 1. Contract Analysis and Filtering.  Run all incoming contracts through an AI contracting tool (in this case, TermScout) to analyze and score the contracts, comparing results against a filter of deal-breaker terms.  The resulting Report Card (example) is used to determine how the contract should be routed.  If the contract passes the filter, it can be executed as-is, with no legal involvement.  If the contract fails, it gets routed for additional treatment.  Data collected in the pilot shows that 17% of contracts potentially will make it through the filter cleanly.  This number likely will grow in the future as vendors become attuned to this approach and the opportunity that it presents to make sales on an expedited basis.

Step 2. Automated Negotiation.  Contracts that narrowly miss passing the filter are good candidates for an alternative approach to negotiation.  One possible approach is to have the TermScout analysis tool auto-generate a Self-Correction Notice (example) to the vendor identifying the few terms where the vendor is off-market and giving them the opportunity to proceed with the sale on an expedited basis if they self-correct the problem by agreeing to a simple addendum.  Data collected in the pilot indicates that 57% of contracts may lend themselves to this approach.

Step 3. Negotiation Boundaries. Contracts that fail to make it through either of the first two steps will have to be negotiated.  Data from the pilot shows that 43% of negotiations currently involve terms where limited opportunity exists to achieve a meaningful gain in risk allocation due to the favorable nature of the vendor’s standard term and/or the lower materiality of the provision.

The TermScout contract analysis tool can generate a Negotiation Check List (example) that identifies the top-tier and middle-tier terms for negotiation and instructs the negotiators not to spend time on anything else.  The tool also can populate the checklist with information regarding what is market for each term on the list, along with data regarding which companies offer the requested provision.  This should streamline the negotiations both by eliminating time spent on low-value points and enabling the negotiator to win more often on the high-priority points.

The results of Phase 1 show that all three steps above are part of the current state of the art.

Step 4. AI-Generated Redlines. Negotiators currently generate most redlines manually, working from a playbook.  This approach is both time-consuming and prone to error.  Data collected by LexCheck shows that companies can reduce cycle times and error rates significantly by using an AI tool to auto-generate redlines.

Some participants in the consortium have begun Phase 2 of the pilot, field testing the effectiveness of the LexCheck AI tool to auto-generate red-lined documents based on the buyer’s individual playbook.

Return on Investment.  In reviewing the above four steps, a key question to ask here is whether the ROI is worth the time and expense of building this type of tech-enabled contracting process.

Having identified the potential opportunity, the consortium then sought to assess the potential payoff associated with moving to this model.  The data indicates a total potential opportunity for cost savings exceeding 80% of current costs.  Some of these potential savings stem from efficiency gains that are entirely within the buyer’s control, such as deciding to accept contracts that pass the filter as-is.  Others, such as using auto-generated self-correction notices to resolve issues, require the vendor to play ball.

Taking all of this into account, as summarized in Figure 6, the data indicate that in most companies/firms, the cost of handling low-risk contracts can be reduced by a third.

IV.  Implementing the model

Finally, the consortium explored what it might take to implement the above-described solution.  With NetApp serving as a laboratory for this aspect of the exercise, the team identified the following requirements in terms of people, process, and technology.


Not surprisingly, the team identified change management as the single greatest challenge to the project.  Implementing the above-described model requires significant changes from the current approach to negotiations.

Under the new model, a contract is reviewed and assessed by TermScout, which generates a “Report Card” (example). This report card provides a pass/fail assessment for key terms associated with that contract type and an overall risk score.  If that risk score is above an acceptable risk target, no negotiation is required.  If the contract falls short of that risk target, the TermScout report identifies those topics that require more acceptable terms.  The LexCheck tool then provides the redline, inserting more acceptable language.  (For companies looking to use standardized responses the TermScout tool can generate Self-Correction Notices similar to this example.) These tools do much of the initial heavy lifting that had previously been done manually by contract specialists and attorneys.  If negotiation is required, the issues are narrowed and there is objective data showing why the other side’s terms were unbalanced.

The scope of use of these tools is a key initial decision point.  Each organization will have to determine its own comfort levels.

At NetApp, there is an initial assessment of a contract to determine if its tier—high, medium, or low—taking into consideration risk, dollar value, and strategic significance. The AI assessments and reviews are adapted to each tier.

  • The high-tier contracts have higher acceptable risk thresholds and require additional human touch.  The AI assessments are still helpful but serve more to identify issues than to resolve them.
  • Medium-tier contracts have a lower risk target threshold and help narrow negotiations to key issues.  The AI tools are used more liberally to lower the amount of human touch for these contracts.
  • Low-tier contracts have the lowest risk target thresholds and rely mostly on AI tools to move these contracts to completion.

The change management challenge is to put trust in these tools and use them to reduce the level of human touch for medium and low-value contracts.  When used in this manner, the human focus is directed toward the most significant contracts, which is the best use of highly skilled resources.

For lawyers and procurement specialists who now review and negotiate every aspect of every agreement, this is a huge change, even though it stands to lessen their crushing workloads and allow them to focus on higher-value matters.  Making this change requires having a strong executive champion and a stream of wins confirming the value of the model.

Wins will be possible only if the model is set up properly.  This means developing appropriate risk profiles and quality materials to implement them.  Doing this is a complex task: to be effective, the risk profile must strike a delicate balance between protecting the company and driving business agility.

Developing risk profiles and implementing materials is somewhat analogous to writing the rules that enable TurboTax to generate a compliant tax return.  This is a critical task that requires a sophisticated contract lawyer or team of lawyers with a strong understanding of the company’s business who can work effectively with the AI teams.  This legal team will need to compare the AI results to what a lawyer would do, provide feedback, and help the software developers continually improve their product.

Andy Banquer

This type of set-up and development work could be handled by internal lawyers, an alternative legal service provider that is part of the client ecosystem (Andy Banquer of Quislex filled this role admirably in the pilot), or a law firm.  Over time, it seems likely that law firms and/or ALSPs (such as those participating in the pilot) will offer these services to their clients, as they have the potential to achieve economies of scale and scope that are not possible within a legal department.

Finally, implementing this model will require the project management and process oversight associated with any automation project.


Implementing this model involves optimizing the company’s current contracting process and updating the company’s workflows accordingly.  These changes can include:

  1. Updating the intake form to get the information required to classify a contract as low, medium, or high value;
  2. Routing it to a machine for initial review and analysis, and
  3. Routing the results to the initial requester.

Based on the rules adopted, a contract passing a target threshold could be auto-routed for signature with notification to stakeholders.  For contracts not passing that threshold, the workflow would then (a) direct the TermScout tool to generate a Self-Correction Notice or (b) route the results to the LexCheck AI tool for redlining.  The seamless interaction among the workflow tool, the AI tools, and approval and signature tools will significantly truncate cycle times and the need for heavy human touch.

Jeff Marple

Like all processes, this one must be appropriately resourced to work as intended.  This could well include having a person serve as a front gate to take the information provided by the intake form and the machine and route the matter appropriately.  Below is an example of the high-level process map developed for the pilot by Jeff Marple of KP Labs;


The instance of the solution created as part of the pilot employs three key pieces of technology.

Workflow:  While it is possible to run workflows manually, having an automated workflow facilitates things considerably.  The team used TAP (by Mitratech) to automate the workflow.

Contract Analysis and Rating:  This solution rests upon a foundation of being able to use a machine to analyze contracts at a granular level, apply filters, and generate report cards, auto-negotiation letters, and negotiation checks.  The team used TermScout to perform this function.

Redlines:  Manually generated redlines work but can be both time-consuming to create and error-prone.  Companies can address these issues by using an AI tool to automate redline generation.  The tool should be trained using the risk profiles and implementing materials described above. The team used LexCheck to perform this function.

Next Steps

The backward-looking data (i.e., signed contracts) examined during Phase 1 of the pilot reveals a potentially massive opportunity for cost savings and improved business agility.

In the next phase of the pilot, members of the consortium intend to implement various aspects of the solution, use them to handle live agreements, and measure realized ROI.  If you are interested in joining the consortium or learning more, feel free to contact Connie Brenton or the author.