The quote has not moved because effort and price are not the same thing, and an SI prices on the second. Systems integrator fees are 40 to 60% of program cost, per ERP Research, so a saving in SI hours can only ever move part of the bill. Agents also remove junior hours before senior ones, so the blended rate rises while the hour count falls, and the price drops by less than the hours do. Even the McKinsey projection of 50% less effort is worth 20 to 30% of budget, and only when the contract forces the SI to pass it through. This article is the margin arithmetic behind that gap.
SI fees are 40 to 60% of the program, so the saving starts from a slice
The number you heard in the sales meeting is an effort number. The McKinsey article from May 11, 2026 says agents could cut implementation effort by at least 50%, with conditional wording and no completed program behind it. BCG, in April 2025, put the figure lower at 20 to 40%. Both describe the hours an SI works, not the budget you sign.
Those hours are a part of the program, not the whole. ERP Research puts SI fees at 40 to 60% of implementation cost. The rest is software subscription, internal staff and backfill, data cleanup, change management, infrastructure, and the dual-run period that runs 12 to 24 months. Agents touch none of that. They do not shorten the months your finance team spends cleaning vendor master data, and they do not reduce the subscription.
So the saving is bounded before any agent runs. On a $5M single-region tier 1 program, SI fees are $2M to $3M. If the SI halves its hours and passes every dollar through, you save $1M to $1.5M, which is 20 to 30% of the program. If the real effect is BCG's 20 to 40%, you save $400,000 to $1.2M, which is 8 to 24%. That is a figure worth negotiating for. It is not half, and the pillar on AI-assisted ERP implementation traces why the half reached you with its qualifiers stripped off.
Agents remove junior hours first, so the blended rate rises while hours fall
Here is the mechanism that eats the rest of the saving. When an agent makes a work package cheaper, it sheds the routine hours first. Those are the junior hours. The senior hours that review the agent's output, own the design decision, and sign off the configuration stay where they are. The hour count falls, the mix shifts toward senior, and the blended rate goes up. The bill moves less than the hours do.
The rate ranges make the effect concrete. KORE1, a staffing firm, puts 2026 boutique SI rates at $150 to $250 an hour for Dynamics 365, $150 to $275 or more for NetSuite, and $200 to $325 or more for SAP. Those are confidence C figures from a firm that places the consultants, and KORE1's own pages disagree with each other, so treat them as ranges, not fixed prices. The spread inside each product is what matters: the top of the range is roughly 1.5x to 1.7x the bottom, and that gap is the junior-to-senior split.
Take a Dynamics 365 bid of 10,000 hours, split 6,000 junior at $150 and 4,000 senior at $250. The bid is $1.9M at a blended $190. Suppose agents remove 40% of the junior hours and none of the senior ones. The arithmetic is below.
| Line | Before agents | After agents | Change |
|---|---|---|---|
| Junior hours (at $150) | 6,000 | 3,600 | down 40% |
| Senior hours (at $250) | 4,000 | 4,000 | no change |
| Total hours | 10,000 | 7,600 | down 24% |
| Bill | $1.9M | $1.54M | down 19% |
| Blended rate | $190 | $203 | up 7% |
The hours fall 24%. The bill falls 19%. The blended rate climbs 7%. A real and well-run efficiency gain turned a 40% cut in junior hours into a 19% cut in price, and the difference is the rising rate. This is an illustration built on the KORE1 ranges, not a quote from a program, but the direction holds on any bid where agents hit routine work first. On the wider SAP range the effect is larger, because the gap between junior and senior rates is wider.
None of this is the SI behaving badly. Opkey's own description of its tools says consultants shift toward advisory work, which is exactly a shift toward senior hours. That is good for quality. It is not a discount unless the contract names one.
Design is 10 to 20% of hours, so even an 80% design cut moves the total by 8 to 16 points
The second reason the quote holds is where the agents work. The loudest vendor claim is on design and configuration. Oracle quoted one customer at an 80% cut in "certain design and configuration efforts." Set that against the size of the package it applies to.
Design, the fit-gap work, is 10 to 20% of implementation hours, per ERP Research. An 80% cut applied to a package that is 10 to 20% of the total removes 8 to 16 points of total effort at best, and only if the cut is as deep as the headline and the package is as large as the top of the range. The table shows the span.
| Design as share of hours | 80% cut removes | Effect on total hours |
|---|---|---|
| 10% (low end) | 80% of 10% | 8 points |
| 15% (midpoint) | 80% of 15% | 12 points |
| 20% (high end) | 80% of 20% | 16 points |
Testing is the package with the most production evidence, and it is larger, commonly a fifth to a quarter of implementation hours. A 60% cut there removes 12 to 15 points of total effort, which is why testing is the right place to press first. The point stands either way: a dramatic percentage on one package is a modest percentage on the program, and the headline attaches to the package, not the total.
Cap the blended rate or fix-price the agent-heavy packages
Two contract moves convert an effort saving into a price saving. Both belong in the statement of work, not in a conversation after you have selected the SI.
The first is a blended-rate cap. Fix the blended rate at or below the bid's starting rate, so the SI cannot hold the price up by shedding junior hours and keeping the mix senior. In the Dynamics 365 illustration above, a cap at $190 would have passed the full 24% hour reduction to the bill instead of 19%. The cap is the single clause that answers the rising-rate problem directly.
The second is a fixed price on the packages the agents touch most. Testing and documentation are the two with the clearest agent evidence and the most repeatable output, which makes them the easiest to price as a fixed package rather than by the hour. Fix-price those packages and the SI carries the risk of its own tooling. If the agents work, the SI keeps the upside on a fixed scope. If they do not, the bill does not move and the overrun is the SI's, not yours.
A blended-rate cap and a fixed price on the agent-heavy packages are not alternatives to reading the contract closely. The twin article on negotiating the SI contract covers the clauses in depth. Here the point is narrow: without one of these two levers, the arithmetic above says the saving stays with the SI as margin.
Re-baseline the price at the end of design, because that is when the saving is visible or gone
The saving, if it is real, shows up early. Design is 10 to 20% of hours and sits at the front of the program, so by the end of design the SI has run its agents on your configuration and knows what they removed. If the promised reduction has not appeared in the design and build hours by then, it will not appear later.
Write a re-baseline point into the contract at the end of design. At that gate, the SI shows actual hours by work package against the baseline it quoted, and the price reopens to match. This is the single clause that protects a buyer who could not force a full hours table before signing. It moves the proof to the moment the proof exists.
The gate also disciplines the claim. An SI that expects its agents to deliver will accept a re-baseline, because it expects the hours to fall and is happy to be measured. An SI that resists the gate is telling you the AI content was a slide. In our experience a tier 1 program needs 36 months from decision to go-live, and the end of design arrives in the first several of those months, early enough to act on what the gate shows.