Who Cares What’s Standard?

I was a junior lawyer, fewer than two years qualified, when I was asked a question during a contract negotiation that I should have been able to answer.

My client was a small software company engaged to deliver software that was to be used to deliver Wi-Fi to the customers of one of Sydney’s public services. The contract included a liability clause that I had drafted and that the lawyers for the other side had amended. The amendment had the effect of shifting significantly more risk onto my client than I considered to be either fair or normal.

“That isn’t market standard,” I told them.

The conversation progressed, and I said it again. “What you’re asking for isn’t consistent with the market standard.”

Eventually, the senior commercial representative for the customer had heard enough.

“I don’t give a f—k about what the market standard is. I care about my deal. Why should I care about what’s standard?”

And damn it, I didn’t know.

The reference to the market standard had been a reliable tool for me in the past. I knew what the market standard position was, and I knew when I was being presented with something that was materially different, and the people I had negotiated against previously seemed to care when I pointed it out. It didn’t always end an argument for me, but it usually did a bit of the lifting.

We moved forward, and the deal closed, but the questions hung around in my mind. Why did it matter? If it was said to me, why would I care?

It was probably about a week or two later when the penny dropped. I worked out what I’d been trying to say. The argument that I was making was a reference to the alternatives my counterparties could access. If the position that I was defending was ordinary for the market, it’s unlikely that my counterparty would enjoy a better offer elsewhere. In all likelihood, another supplier would resist the request, too.

My counterparty could challenge my understanding of the market, or they could claim that their deal justified a deviation from the standard, but there was a strong commercial basis for the position. The phrase communicated that, and by understanding it, I gained access to something that enabled me to consider it properly against the actual transactions that I was working on.

Before that point, I was only borrowing the authority of the phrase. I didn’t own it, because I didn’t understand where the authority came from.

That story is one that I tell often. All of my colleagues have heard it. But I’ve been thinking about it more often lately. When I read about proposals to automate legal judgement with AI, using defined guardrails and records of accepted positions, the exchange taps my shoulder with a new feeling.

A historical record of completed contracts can provide the information of what was agreed, but the reasons and judgement that were applied are much harder to review. Concessions may reflect deal pressure, careful consideration of risk, something deemed not worth worrying about, or even mistakes. All of them land in an archive wearing the same signatures.

Even in instances where a decision is recorded together with its justification, applying the justification again requires another judgement. The new context matters, and the persuasive elements giving rise to the original decision may be less significant when applied to different circumstances.

Getting that judgement wrong in an automated system carries a commercial cost that I think is underappreciated. Sometimes departing from standard is what makes the business possible. I don’t mean a concession made to get the other side over the line. I mean cases where the non-standard term is part of what is being sold.

Years later, I was asked to help integrate an acquired software product into the software suite of a much larger organisation. The product functioned as an information exchange between businesses working on a shared project, but without direct contractual relationships. The contracting model needed to accommodate participants that weren’t paying customers but who still needed rights in relation to the software.

Think of a builder who pays for the platform, and needs the architect and the engineer to share information. The architect is engaged to share different files with the builder and the engineer, and won’t want to trust either of them to manage all of the files in the best interest of the architect.

That was a fundamental problem for the acquiring business’s standard contracting model.

I had available to me an obvious and easy way to make it work. The business could contract with the customer, give the customer authorisation rights with respect to third parties, and make the customer responsible for the use of the product by the third parties as if they were customer users. The software would still work, without losing any technical functionality, under that model.

The challenge was the effect that that would have on the relationship between the participants. The blocker that the software was intended to address related to information sharing between parties that couldn’t trust each other because of a lack of a direct contractual relationship. All parties could trust the software platform if all parties had independent rights in relation to their use of the platform. The supplier taking on a direct relationship with the non-paying party, and the risk that came with it, was a major part of what made the product attractive to the paying customer.

A conventional and market standard contract would have eliminated a key reason to use the product at all.

My expectation was that attempting to introduce an alternative contracting model into an organisation the size of the one I was working for would be near impossible. I fully anticipated being asked to find another option. Instead I was surprised by how accommodating the business was. There was no meeting where a brilliant argument was made in favour of the request and everyone cheered with enthusiasm. It was a simple “Yes.” following the first assessment of the natural justification.

The business approvers understood that the request carried additional risk. The exposure created by giving rights to a non-paying party does not disappear just because that exposure is part of the product. But it does give the business a reason to accept the risk on this occasion.

The decision to accept that type of risk is very awkward to standardise. The departure from the standard position is easy to identify, and the explanation for why the ordinary position should prevail is clear. You can take both as references for automated decision making and reach a conclusion that destroys the product’s commercial value.

My enthusiasm for using AI to change how legal work gets done should be no secret to anyone. Most of the work that I currently contribute value through is work that I would very happily redesign for the current era of technology. But the question of how much authority a system should get is one that gets difficult very quickly.

Recently I asked an AI tool to review a contract that I was negotiating. One of the clauses under negotiation related to indemnities. The tool, set up with the appropriate context for legal work, suggested that I should accept the customer’s request to mutualise indemnity provisions so that both parties would accept the same risks expressed originally as the supplier indemnity.

In the contract that I was working with, the customer’s indemnity addressed claims that arose from the customer’s conduct and the content that they uploaded to the supplier’s platform. The supplier’s indemnity addressed third party intellectual property claims relating to the software. In both cases, the indemnity addressed items that were material risks to the other party, that the other party could not control, and that the other party should expect the indemnifying party to be fully accountable for.

Making them mutual meant replacing the customer’s indemnity with a mirror of the supplier’s. Each party would cover the other for IP claims, and the customer’s promise to answer for its own conduct and content would fall away. The supplier would lose its protection against exactly the risk it could not control.

The recommendation sounds reasonable if the goal is balance. Here, it missed the point.

I saw my younger self in that output. The tool was making a recognisable negotiating argument without the reason that would justify it in this deal. Its confidence came from the familiarity of the pattern, not from the transaction in front of it.

To respond to this risk, the natural inclination of the business is to seek to tighten the instructions of the tools. More positions, more history, more reasons, better definitions, explicit limitation disclosure requirements, clearer red line definitions, source reporting. The list is endless. People relying on these tools need to be able to trust what they recommend.

And yet, the tighter instructions bring with them another problem. Take the platform contract. That contracting model would almost certainly sit outside a tightly instructed tool’s permitted answers. Even a tool that understood the commercial case perfectly would have no authority to accept it.

This is where the hunt for reliably cautious automation becomes an expense nobody sees, least of all the person with authority to approve the exception. A bad concession leaves an exposure someone can point to. A good arrangement rejected before anyone considered it leaves almost nothing. The request is declined, and the deal never happens.

A tool administrator sees a working process.

This challenge existed before AI. Research relating to automated tenant screening has indicated that discretion doesn’t disappear; it moves into rules written in advance. The exceptions can remain available, but there is a problem if all qualifying circumstances are determined in advance and must be presented in a recognisable format. An unanticipated explanation has no category to go into. A familiar one slots neatly into its category and is handled by protocol.

In the commercial context, I find the challenge to be equally troubling. Picture a small business approaching a much larger one with an offer that needs non-standard commitments, a different relationship between the parties, or a deeper appreciation of what the product is actually for. The reason why the offer carries value may be directly related to the reason why it fits poorly into the established model.

If the larger business’s system can only negotiate inside established boundaries, the smaller business may have to change its offer to get a deal. The larger organisation’s contracting assumptions start to decide what everyone else can build.

Humans already disappoint in this respect. In my experience, getting attention and consideration for unusual requests can often depend far too much on reaching the right person on the right day. I have been both surprised and frustrated by responses that I’ve seen or received. That inconsistency alone is a legitimate justification to seek out something better.

Maybe AI could help. It could make arguments clearer, identify reasons for restrictions, and reduce the effort required to process a departure request. Little in my experience suggests an AI system is incapable of useful judgement.

But my problem is that even if AI judgement improves, someone at an enterprise must still decide what the system is actually allowed to do with that judgement. A path of escalation is really only useful if there exists someone who is capable of reconsidering the rule, and who also has the inclination to do so. Listing every known acceptable exception will always omit the next unfamiliar opportunity that falls outside of the list.

And a general instruction to take chances in response to exciting-sounding commercial offers is not a serious option. Delegating authority in such a way would be a concerning surrender of calibrated judgement. Not all persuasive proposals should succeed, and some well-measured risks still go wrong. Explanations that justify departures for one transaction can be totally inappropriate for another.

We still have to make the judgement.

Two evenly matched tennis players can trade thirty excellent forehands in a rally: controlled, repeatable, with good margins. To end the point, one of them usually has to see an opening and go closer to the line.

They might miss. Choosing the shot means accepting that. A coach who banned it would cut the misses, and remove the shot that wins the point.

What I want is for the legal profession to be involved in deciding how these systems get built. I believe the choices extend much further than determining how quickly documents are produced. They determine which proposals are considered, which risks can be taken, and how businesses discover that existing positions should change.

The commercial representative negotiating the Wi-Fi contract cared about his deal. I gave him the standard answer. I eventually came to understand why my answer mattered, and it enabled me to think about when it might cease to matter enough.

I would like the systems we build next to be capable of having that same conversation.

Especially when nobody has agreed to the deal before.