Autonomous Tendering Is Coming for the Static Routing Guide

The Analysis

I read the 23 June Logistics Viewpoints analysis on autonomous tendering the way a buyer reads anything about freight: which line of my landed cost does this touch, and who pays for it. Standing on the buyer's side of the table, the answer to the second question never changes. I pay. The argument is that the static routing guide, the ranked ladder of primary, secondary and backup carriers that most transportation teams tender down, is giving way to systems that weigh real-time capacity, carrier acceptance rates, spot alternatives and facility constraints at the moment of tender and then choose. Contract rates do not disappear. Human oversight does not disappear. What changes is that the thing doing the choosing is no longer a list I wrote nine months ago; it is a decision policy that runs on every single load.

Before I get anywhere near a signature I want to be honest about one thing. This is not a story about technology arriving to save money. It is a story about who writes the rules. The analysis is explicit that autonomous tendering does not eliminate procurement expertise; it embeds that expertise into automated guardrails. If that is true, then the guardrails are the deliverable, and a guardrail is a commercial document, not a software setting. Somebody has to write the maximum spot premium the system may pay, the lanes where a human must approve, the volume floor that keeps a second carrier alive, and the reason code that lets me reconstruct a decision six months later. If I do not write those numbers, they get written by whoever configures the tool, and I will be paying the bill for someone else's defaults.

Let me pin down what the analysis actually claims, because most of the noise around it is not in the piece. The traditional routing guide is built for a stable market. You run an annual or semiannual bid, carriers are awarded lanes, primary, secondary and backup carriers are ranked, and the transportation management system tenders down that hierarchy. That works when capacity and commitments hold. It stops working when capacity tightens, spot rates move, service performance changes, facilities congest, or a weather or labour event changes the economics of a load in the middle of a quarter. The analysis is careful about the distinction that matters: autonomous tendering is not automated tender sequencing, which has existed for years. It is decision automation. At the moment of tender the system weighs historical acceptance rates, recent lane-level performance, real-time capacity, cost and service trade-offs, facility constraints, appointment availability, customer priority, spot market alternatives, emissions considerations and exception risk, and then it chooses.

None of that is a freight discount. Standing on the buyer's side, I care about exactly one thing: which line of my cost structure moves, and by how much. So let me take the cost structure apart before anybody gets excited. A move I buy contains the contracted line-haul rate, accessorials, the spot premium when the ladder fails, the internal labour of chasing a tender, the cost of an expedite, the inventory carrying cost of a late arrival, claims, and the cost of a service failure that reaches a customer. Autonomous tendering touches the decision made at one instant, and that decision feeds three of those lines: how often I fall through to spot, how much internal time each fall-through eats, and how often a load arrives late.

There is a number in this story that I do not trust, and it happens to be the one most transportation teams still report: routing guide compliance. The analysis makes the point plainly. Compliance made sense when the guide was the best available plan. In a market this dynamic, the better question is whether the shipment was executed according to the best available decision at the time. That is a big change for a buyer, because compliance is easy to measure and easy to hide behind. If I pay for a system that hits 98% compliance by tendering down a stale ladder and then covering 8% of the book in the spot market at a 25% premium, I have bought a metric, not an outcome. I would rather pay for the outcome.

So let me do the arithmetic, and let me state every assumption out loud, because not one of these figures comes from the analysis. Assume I run 1,200 tenders a month, which is 14,400 a year. Assume my average contracted rate is $1,000 per move, used purely as an anchor so the percentages have somewhere to land. Assume 8% of tenders fall through the entire ladder and get covered in the spot market, which is 96 moves a month and 1,152 a year. Assume spot on those lanes runs 25% above my contract rate, so $250 per move. That is $24,000 a month and $288,000 a year of premium I never put in a budget. Now add the labour: assume each fall-through eats 45 minutes of an analyst at a fully loaded $60 an hour, which is $45, or $51,840 a year. The two together are $339,840 a year sitting inside one decision.

That figure is not a saving. It is the size of the prize if the fall-through never happened again, and it will happen again. The useful question is what improvement pays for the capability. Take a cost you must fill in from your own quote rather than from this article, and for illustration assume $90,000 a year all in, licence plus internal labour. Each avoided fall-through is worth $295, the $250 premium plus the $45 of labour. So the break-even is 90,000 divided by 295, which is about 305 avoided fall-throughs a year, or 25 a month, or 2.1 percentage points on a 1,200-tender book. My threshold is therefore concrete: the fall-through rate has to come down from 8.0% to 5.9% or better. If a proposal cannot explain how it gets there, it is not a proposal, it is a demo.

Run the same arithmetic on one lane where I already suspect a problem. Assume 200 tenders a month on it. Assume my primary carrier accepts 82%, which is 36 rejections a month. Assume another qualified carrier on that lane has been accepting 90%, so 20 rejections. The difference is 16 moves a month that never have to be re-tendered or covered in the spot market. At the $250 premium I assumed, that is $4,000 a month, or $48,000 a year, on one lane, from one reordering of a ranking. I am not claiming the system will find it. I am saying that this is the size of the thing worth looking for, and that 82% versus 90% is a number I can check in my own tender history before I believe anyone's model.

This is where the deal actually gets written. The analysis says autonomous tendering should not be confused with unmanaged automation, and that shippers will need guardrails, approval thresholds, exception rules and auditability. I agree, and I would add that those four things are commercial terms. A guardrail on price is a purchasing decision. A threshold that forces human approval is a risk decision. An exception rule is a service promise made to a customer. Auditability is what lets me defend a decision to my own finance team. The analysis also says the best model is human-supervised autonomy: managers define strategy, constraints and escalation, and the system executes inside those boundaries. Fine. Then the boundaries are my job, and I want them in a document with my name on it.

Concretely, here is what I would write down before anything goes live. Cap the spot premium the system may pay at 15% above the contract rate, or $150 per move on my $1,000 anchor, whichever is lower, and require a named human on anything above that. Require human approval on any move above $2,500, any expedite above $500, and any shipment for a customer I have named as strategic. Hold a minimum tender share of 10% for every qualified backup carrier on a lane, so the second and third names stay real. And require a reason code on every automated award, retained for 24 months, that I can export without asking the vendor for a favour.

That 10% floor is the one people argue with, so let me say why I insist on it. A routing guide is also a benchmark. The second and third names on the ladder are what let me test the first name's price at the next bid. If an optimiser is free to send everything to whoever looks best this week, it will gradually starve the backups, and the next time I go to market I will be negotiating with one carrier that knows it has no competitor and a second one that has stopped caring. I have watched that happen with nothing more sophisticated than a dispatcher's habit. Autonomy will do it faster and with better maths. The floor is cheap insurance. I would rather spend 10% of a lane's tenders than lose the benchmark.

One more requirement comes straight out of the analysis, and I would not sign without it. It says a strong system should be able to explain why a carrier was selected, and should learn from tender rejections, service failures and changing market conditions. Explain why is the phrase I care about. If the system cannot produce a reason code I can read, I cannot audit it, I cannot defend it to the carrier that lost the load, and I cannot improve it. And the learning loop runs on my data: my tenders, my rejections, my lanes. I want it stated in the contract that I own the tender log and can take it with me, because that log is the evidence base for every bid I will run from here on.

So how would I sign this if I were the buyer? I would keep the bid. The analysis is careful on this point: contract rates and carrier commitments still matter, and the routing guide is not dead, it is being redefined as one input into a broader optimisation model. I would keep the annual or semiannual bid, keep the contracted rate, and keep the ranking. What I would change is the status of the ranking. In the contract I would add a decision-policy exhibit saying the order is an input and not an obligation, that the system may reorder inside the guardrails above, and that either party can revert to static tendering for one quarter on 30 days' notice if the numbers move the wrong way.

I would also change how commitment is defined. A volume commitment written as a fixed number of loads fights the whole idea; I would write it as a share of the lane's tenders offered first, with acceptance measured on tenders offered inside an agreed window rather than on a 4pm request for a 6am pickup. Carriers have a legitimate complaint on that point and I am not going to pretend otherwise.

And I would tell the carriers what is happening before they hear it from a portal. If a carrier is going to be judged on acceptance rate, it deserves to know the definition, the window and the consequence. If a carrier is going to lose volume to an algorithm, it deserves to know the rule and the appeal path. Every carrier I have ever bought from accepts a hard rule far more easily than a silent one. The analysis notes that autonomous tendering exposes uncomfortable realities: some guides are stale, some rankings reflect old assumptions, and some preferred carriers are preferred because they won a bid rather than because they are the best choice today. That is going to be an uncomfortable conversation. It is still better to have it in July than in the middle of the fourth quarter when the spot premium is already sitting on my invoice.

On timing I work in quarters, not weeks, and this is a two-quarter job. In the quarter that ends 30 September 2026 I want the baseline: tender count, fall-through rate, spot premium over contract, acceptance rate by carrier and lane, and internal minutes per fall-through, all frozen with one named owner per measure. I want the four guardrails drafted and walked through with transportation and finance. Over the following quarter I want one lane live, not the whole network, with the same three numbers measured weekly against the baseline and a written review in the week of 14 December 2026. The gate to expand is simple: fall-through at 5.9% or lower, weighted acceptance at 90% or better, and cost per move versus contract within 3%. Miss any one of the three and the lane goes back to static.

There is one more commercial detail I would settle before the first live month, and it is the one most likely to turn into an argument later: what counts as an offer, and what counts as a rejection. If the system fires a tender and the carrier does not answer inside the window, is that a rejection, a timeout, or a non-event? The answer moves the acceptance rate, and the acceptance rate is about to become the number a carrier is judged on and possibly paid against. I would write the definition into the same exhibit as the guardrails, with the timeout length, the business hours it applies to, and what happens to a tender sent outside them. Carriers are not being difficult when they ask this. They are being sensible, and if a carrier is measured on a number it had no hand in defining, the first quarterly review will be spent arguing about arithmetic instead of service.

I would also change what I ask in the bid itself. Three questions, and none of them is on the standard template. Where do you think my guide is wrong? What would you change about the way work reaches you, if you could change one thing? And what data do you need from me to accept more of what I offer? That last one is the useful one, because acceptance is a two-sided problem. A carrier rejecting tenders on a lane may be telling me something true about the lane, about the facility, or about the rate I won. If I only measure the rejection, I will spend the year squeezing a carrier for a problem I created. If I ask, I occasionally get an answer worth more than the premium.

The ways this goes wrong are unglamorous. An optimiser told to minimise cost will quietly buy service failures, and you will see it in expedite spend rather than in freight spend, so put expedite on the same page as the freight number. Bad inputs are worse than no model: if my acceptance history is missing the tenders my own team cancelled before they were ever offered, the system will learn the wrong lesson. A pilot run inside one quarter will overfit to that quarter, which is why I want a full quarter and two different market conditions before I believe a result. And there is a people risk. A transportation manager who spent nine months building the guide is going to read this as an audit of their judgement. Sometimes it is. Handle it by making that person the author of the guardrails rather than the subject of them.

I am not going to use any of this to beat up a carrier. The analysis makes a point I do not enjoy and do agree with: some rankings reflect old assumptions, and some preferred carriers are preferred because they won a bid, not because they are the best choice today. That is my failure as a buyer, not theirs. I wrote the ladder. I let it go stale. I kept tendering into a facility that has been congested since February without touching the ranking. If a system surfaces that, it is doing me a favour, and the honest response is to fix my own document, not to squeeze the carrier who happened to be standing nearest.

There is also a budgeting point hidden inside that $288,000, and it is the reason most of these projects never show a return. In most organisations that money is not a line. It appears as a variance against a freight budget, it gets explained as market conditions, and nobody owns it. If I am going to pay for a capability whose entire purpose is to attack that number, the number has to exist first, with a name against it. Otherwise whatever I save is indistinguishable from a soft quarter, and I will be asked in twelve months what the capability cost and I will have nothing to point at.

I will say plainly what I do not know. The analysis says the trend is expected to accelerate as freight volatility persists. It does not give me an adoption rate, a saving, or a date, and I am not going to invent one. What I do know is that the decision is moving out of a document I control and into a policy I have to write down, and that writing it down is a purchasing job. Standing on the buyer's side, that is the whole of it. The money is mine either way. I would much rather be the person who wrote the number than the person who has to explain it later.

↑ Back to the key points

— By Grace Yang