Overview

This November 2026 update reassesses AI-driven smart execution versus deterministic VWAP/TWAP for large crypto orders using expanded data and industry developments through October 2026. We re‑ran controlled backtests (Jan–Oct 2026), interviewed execution desks and relays, and incorporated new market plumbing changes—subscription relays, standardized builder fee APIs and wider encrypted‑mempool uptake—to produce a current decision framework for when AI slicing produces real net savings and when simple VWAP/TWAP remains preferable.

Background: what changed since June 2026

Since our June 2026 update, three practical shifts have altered execution economics:

  • Private submission commercial models matured. Multiple relays now offer subscription pricing (flat monthly + reduced per‑slice fees) and standardized builder fee APIs. That reduced per‑slice uncertainty for some institutional flows but raised the effective cost floor where previously builder fees were discretionary.
  • Encrypted mempool adoption widened. More rollups and several third‑party relays adopted encrypted submission by default for institutional channels. This materially reduced naïve sandwich leakage for private‑relay users, but did not eliminate builder‑level ordering and allocation economics.
  • Builder fees and auction dynamics hardened. As competition for priority access increased, builder-side fee floors rose in congested windows. That made AI strategies that depend on frequent on‑chain interaction more sensitive to relay economics and added variance to net outcomes.

Data and evidence: expanded backtests (Jan–Oct 2026)

Methodology: we expanded the original study window to Jan 1–Oct 31, 2026. Inputs: tick‑level CEX orderbooks, on‑chain traces across major L2s, public and private mempool captures, and relay/builder fee schedules gathered from market vendors (anonymized). Assets: BTC, ETH and a representative high‑liquidity L2 token (ARB). Parent orders: buys and sells sized 0.25%–2.0% of daily volume, executed under three modes:

  1. Classic TWAP (uniform slices)
  2. VWAP (historical intraday volume profile)
  3. AI engine reoptimizing sub‑second (200–800 ms) with short‑run impact forecasts and mempool prediction, able to submit via public mempool, subscription relays, or builder APIs

Cost components: realized slippage (arrival-price benchmark), exchange/AMM fees, gas + priority fees, detected sandwich MEV (open heuristics), builder/relay payments, and operational overheads. We stress‑tested for gas spikes and relay fee surges observed in 2026.

Key quantitative findings (summary)

  • Realized slippage: AI still reduces median slippage versus TWAP, but magnitudes shifted:
    • BTC: median improvement narrowed to ~6–12% (relative reduction in slippage for 0.5%–1% ADV orders). Deep CEX liquidity and improved matching engines reduced incremental AI gains compared with early 2026.
    • ETH: median improvement moved to ~15–25%. AI captures L2 routing windows and CEX imbalances but is now more sensitive to builder fee spikes that occur in congested blocks.
    • ARB (L2 token): median improvement remained sizable at ~18–33%; fragmented pool liquidity and concentrated liquidity strategies continue to offer routing gains to dynamic engines.
  • MEV exposure and relay economics:
    • Public‑mempool AI routing continued to increase detected sandwich/priority losses (roughly +4–12 bps vs TWAP for ETH/ARB in our runs), particularly for aggressive, frequent slicing.
    • Subscription relays with encrypted submission reduced detected sandwich cost to near‑zero in most tests; however, relay subscription models and builder fee floors raised per‑slice effective costs. Relay/ builder payments in our scenarios ranged from ~1–12 bps depending on congestion and subscription structure (flat + per‑slice).
    • Where relays offered explicit allocation tradeoffs (discounted fees for preferred flow), potential conflicts with best‑execution obligations emerged in our compliance interviews—traders must treat these as execution economics, not free rent.
  • Net total cost (slippage + fees + gas + MEV + builder payments):
    • BTC: AI produced modest net gains when off‑chain routing dominated (net savings ~2–8 bps). When AI increased on‑chain activity unnecessarily, net gains vanished.
    • ETH: AI still commonly outperformed VWAP/TWAP, but the distribution tightened: net savings commonly ranged from ~4–18 bps. During builder fee spikes or gas surges the advantage narrowed and occasionally reversed.
    • ARB: AI yielded the largest net savings in many configurations (8–28 bps) when private submission and subscription relay economics aligned with expected slippage improvement; public mempool routing often eliminated those gains.

Case example (updated): a simulated 1% ADV aggressive buy of ETH executed on Sept 14, 2026 (a date we used to stress gas/relay dynamics). The AI engine used sub‑second slicing, opportunistic L2 routing and a subscription relay. Results vs VWAP: realized slippage −18% (relative), detected sandwich MEV ≈ 0 bps (encrypted relay), relay/builder payments 6 bps under the subscription plan, net cost improvement ≈ 12 bps. The identical engine submitted via public mempool lost ~7–11 bps to sandwich/priority extraction in the same window, turning net improvement negative.

Multiple perspectives

Execution desks: Several sell‑side and agency desks report that subscription relays materially reduce variance in execution economics compared with ad‑hoc builder payments. Their caution: subscription models shift costs from per‑slice volatility to a fixed component—good for high volume but less attractive for one‑off institutional trades.

MEV researchers: Researchers we consulted noted that encrypted submission reduced observable sandwich events but MEV migrated to allocation and priority design choices inside builders. They emphasized that simple sandwich heuristics undercount allocation‑level extraction and advised richer detection metrics (e.g., allocation patterns, bundle ordering analysis).

Compliance and legal: Compliance teams stressed documenting relay/subscription terms, disclosing builder fees in pre‑trade cost estimates, and preserving audit trails. When relays offer allocation incentives, firms must evaluate conflicts with best‑execution obligations and client disclosure requirements.

Implications for traders

The Nov 2026 evidence reinforces a pragmatic, hybrid approach and adds tactical considerations around relay economics:

  • Prefer AI-driven execution when:
    • Order size >0.5% ADV and liquidity is fragmented across L2s and CEXs; pre‑trade sims show slippage gains exceed relay/subscription costs.
    • Your firm has access to encrypted or subscription relays with transparent fee schedules and you can quantify per‑slice vs fixed cost tradeoffs.
    • Infra supports sub‑second decisioning, continuous telemetry, and governance controls (fee caps, rapid fallback triggers).
  • Prefer VWAP/TWAP when:
    • Order is small (0.25% ADV) or you lack reliable private submission access.
    • Relay subscription economics are unfavorable for one‑off trades or builder fee floors spike during your target window.
    • Your team cannot instrument allocation‑level MEV detection and quick fallbacks.

Updated best practices and implementation checklist

  • Run relay‑aware pre‑trade sims: include subscription pricing, builder fee floors and gas spike scenarios. Model both per‑slice and amortized subscription cost.
  • Negotiate transparent relay terms: prefer relays that publish builder fee schedules, allocation policies and offer per‑slice telemetry for auditing.
  • Set builder‑fee caps and allocation knobs: enforce hard per‑slice fee ceilings and trade size limits to prevent runaway auction costs during congestion.
  • Expand MEV detection beyond sandwiches: instrument allocation‑level signals (repeated beneficiary patterns, bundle ordering shifts) and reconcile with relay reports.
  • Hybrid routing architecture: use AI for opportunistic slices plus deterministic VWAP/TWAP baseline slices; route only clearly opportunistic on‑chain activity through paid relays.
  • Compliance and bookkeeping: log relay contracts, builder payments and allocation receipts; include these in pre/post‑trade reports for client disclosure and audit trails.

Outlook — next 12–18 months

Expect these likely developments with reasonable probability:

  • Relay subscription growth: subscription models will expand; institutions with steady flow will find lower variance in execution costs while occasional traders may prefer pay‑per‑slice alternatives.
  • Standardized builder APIs and reporting: more relays will adopt standardized fee and allocation reporting; this will improve pre‑trade modeling but also crystallize the cost of privacy.
  • Regulatory focus intensifies: regulators and industry groups are increasingly focused on auction transparency and client disclosure of builder/relay payments; expect more formalized guidance that affects product design.
  • AI execution commoditizes, ops remains differentiator: basic short‑run forecasting will be widely available; edge will come from data quality, relay access, governance and execution operations excellence.

Bottom line

Through October 2026, AI‑driven execution remains a net positive in many scenarios—especially for ETH and L2 tokens—when combined with encrypted or subscription relay access and disciplined builder‑fee governance. The core change since June is economic: private submission now has clearer, often non‑negligible costs that must be modeled up front. For trading teams: run relay‑aware pre‑trade sims, negotiate transparent subscription terms, cap builder fees in live execution, and retain deterministic VWAP/TWAP fallbacks.

FAQ

How much can AI execution save on net, realistically in Nov 2026?

In our Jan–Oct 2026 backtests, net savings (slippage + fees + gas + MEV + builder payments) for AI vs VWAP/TWAP ranged: BTC ~2–8 bps, ETH ~4–18 bps, and L2 tokens like ARB ~8–28 bps. Results are scenario‑sensitive—run relay‑aware pre‑trade sims on your exact venue set.

Do subscription relays always make AI execution better?

No. Subscription relays reduce variance but introduce a fixed cost component. They typically help high‑volume, repeat flow owners. For one‑off or low‑volume orders, subscription amortization may be unfavorable compared with pay‑per‑slice or deterministic off‑chain routing.

Does private submission eliminate MEV risk entirely?

No. Encrypted relays dramatically reduce observable sandwiching, but MEV migrates to builder allocation and priority economics. Effective mitigation requires fee caps, allocation‑level monitoring and contractual transparency with relays/builders.

When should I fall back to VWAP/TWAP?

Use deterministic fallbacks when private submission is unavailable, relay subscription economics aren't justified, gas/builder fees spike, monitoring flags model drift or latency issues, or when regulatory/audit simplicity is required.

How often should I recalibrate AI execution models?

Cadence depends on regime volatility. For active ETH/L2 flows, weekly recalibration for short‑run impact models plus continuous telemetry for drift detection is prudent. Maintain monthly governance reviews and versioned decision‑logic for audits.