Robot mediators? A short review of SmartSettle One

Bruce Greig
October 7, 2020

There are AI tools out there to help manage a negotiation. I take a look at SmartSettle One. How intelligent is it?

The robots are coming.

There are a few tools out there which use artificial intelligence to help navigate a negotiation.

SmartSettle One is one such tool. At a glance, it just guides each party to communicate a series of settlement offers to the other side … no rocket science there … but it has two extra twists which help the process along:

Secret offers

Each party can make a public offer and a secret offer. You might be asking £500k, when you’d actually settle for £250k. But you don’t want to offer £250k publicly for fear of getting beaten down further from there. The other side might be denying your claim completely, offering £zero. They know they risk £250k in costs from going to court, which they can’t be sure of recovering, so they might offer £250k to settle, but they don’t want to do that publicly yet either.

You are poles apart with your opening offers, but Smartsettle allows you to secretly share what you’d really settle for with the system. In my example, you’d both be willing to settle for £250k, and agreement is reached. And the secret offers remain secret: each knows that £250k was in the zone of possible agreement, but not what the precise amount each side offered.

Reward early effort

If secret bids overlap then how do you decide the settlement figure? Any figure within the overlap will in theory be acceptable to both parties. SmartSettle uses an algorithm to “reward early effort”. The claimant who moves their secret bid further early on gets a larger share of the overlap. For example, consider if the claimant above secretly moved from £500k to £250k then down to £200k, and the other side stuck at £0 until shifting to £250k at the last minute. Any figure between £200k and £250k is acceptable in theory, but the Reward Early Effort formula will push the settlement figure nearer to £250k, to reward the claimant who moved further earlier.

It strikes me that this does create opportunities to game the algorithm. For example, what if you start with a very high offer, immediately jump your secret offer way lower, can you lock-in the early effort reward that way? And, actually, the reward is for the party with the smallest last move. Not the largest early move, but the smallest last move, on the basis that if their last move was small, they must have already moved further early on. That seems to encourage tedious tiny moves towards the end of a negotiation?

Digging into the detail it looks like what is billed as an “intelligent algorithm” actually just calculates a weighted average of the two offers, weighted by the size of the final moves:

Settlement = (Sc * Fc + Sd * Fd) / (Sc + Sd)


Fc = final secret bid of Claimant

Fd = final secret bid of Defendant

Sc = size of Claimant’s move in the last session

Sd = size of Defendant's move in the last session

So in the example above, settlement figure would have been:

(50*200+250*250)/(250+50) = £242k

I’m a technologist, I always use the technology solution for a task which computers can do better than humans. But is this tool really better than humans? Or is it a solution looking for a problem? Hard to know.

More here:

Technical explanation of the algorithms here: Rewarding Good Negotiating Behaviour