Open a study’s context for its setting, limits and a proposed ABM application. Research, surveys and technical guidance are labelled separately.
NeurIPS research paper · 2020
Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
Patrick Lewis and co-authors
Retrieval-augmented models produced more factual generation than the tested parametric-only baseline.
Study context and ABM application
- Study setting
- Knowledge-intensive question answering and generation benchmarks
- Boundary of the evidence
- Wikipedia-based benchmark results do not establish accuracy for a private account database.
- What we would test in ABM
- Test retrieval and claim support separately before accepting a brief.
EMNLP research paper · 2023
Enabling Large Language Models to Generate Text with Citations
Tianyu Gao, Howard Yen, Jiatong Yu and Danqi Chen
The study evaluates fluency, correctness and citation quality as separate dimensions.
Study context and ABM application
- Study setting
- ALCE question-answering and citation benchmark
- Boundary of the evidence
- A plausible citation can still fail to support the sentence beside it.
- What we would test in ABM
- Check whether each material assertion is entailed by its cited passage.
Peer-reviewed TACL research paper · 2024
Lost in the Middle: How Language Models Use Long Contexts
Nelson F. Liu and co-authors
Performance depended on where relevant information appeared in the tested contexts.
Study context and ABM application
- Study setting
- Multi-document question answering and key-value retrieval
- Boundary of the evidence
- The tested model generation does not determine the behaviour of every current model.
- What we would test in ABM
- Move critical constraints around a representative brief and test recall.
ICLR research paper · 2023
ReAct: Synergizing Reasoning and Acting in Language Models
Shunyu Yao and co-authors
Interleaving reasoning and actions improved results on the studied benchmarks.
Study context and ABM application
- Study setting
- Question answering, fact verification and interactive task benchmarks
- Boundary of the evidence
- Benchmark success is not permission to operate a production marketing system autonomously.
- What we would test in ABM
- Design observable tool steps with bounded actions and stop conditions.
Security research paper · 2023
Not what you have signed up for: Indirect Prompt Injection
Kai Greshake and co-authors
Instructions embedded in retrieved data could redirect application behaviour.
Study context and ABM application
- Study setting
- Demonstrations against LLM-integrated applications
- Boundary of the evidence
- The paper demonstrates attack mechanisms rather than a current marketing incident rate.
- What we would test in ABM
- Treat retrieved pages as data and enforce tool permissions outside the model.
Peer-reviewed systematic review and meta-analysis · 2024
When combinations of humans and AI are useful
Michelle Vaccaro, Abdullah Almaatouq and Thomas W. Malone
Human–AI combinations underperformed the better standalone participant on average; outcomes varied by task.
Study context and ABM application
- Study setting
- 106 experiments; 370 effect sizes
- Boundary of the evidence
- The pooled result covers heterogeneous tasks and earlier systems, not one ABM workflow.
- What we would test in ABM
- Compare human, agent and combined versions of the same task.
Peer-reviewed randomised writing experiment · 2023
Experimental evidence on the productivity effects of generative artificial intelligence
Shakked Noy and Whitney Zhang
Average task time fell 40% and assessed writing quality rose 18% with ChatGPT.
Study context and ABM application
- Study setting
- 453 college-educated professionals
- Boundary of the evidence
- Professional writing tasks are different from factual account research and qualified meetings.
- What we would test in ABM
- Measure accepted message quality and total production time before testing response.
Peer-reviewed marketing research · 2015
Unraveling the personalization paradox
Elizabeth Aguirre and co-authors
Responses to personalised advertising depended on how information was collected and on trust cues.
Study context and ABM application
- Study setting
- Field evidence and three consumer advertising experiments
- Boundary of the evidence
- Consumer advertising findings do not establish enterprise reply rates or legal compliance.
- What we would test in ABM
- Use relevant business context without suggesting undisclosed personal surveillance.
Peer-reviewed advertising field experiment · 2013
When Does Retargeting Work? Information Specificity in Online Advertising
Anja Lambrecht and Catherine Tucker
Specific retargeted ads were less effective on average; preference development changed their usefulness.
Study context and ABM application
- Study setting
- Online travel advertising field experiment
- Boundary of the evidence
- Travel-product advertising is not an enterprise software procurement experiment.
- What we would test in ABM
- Match message specificity to observed evaluation readiness.
Peer-reviewed B2B survey research · 2025
The influence of key account management on competitive advantage and firm performance
Farbod Fakhreddin, Pantea Foroudi and Kaouther Kooli
Survey modelling connected key-account orientation, capabilities, competitive advantage and performance.
Study context and ABM application
- Study setting
- 568 European B2B supplier firms
- Boundary of the evidence
- Associations in a survey cannot establish the causal return of an ABM campaign.
- What we would test in ABM
- Inspect relationship capability and delivery ownership alongside campaign execution.
Publisher-reported B2B buyer survey · 2026
The surprising economics of B2B growth
McKinsey & Company
Buyers reported using an average of ten channels across the purchasing journey.
Study context and ABM application
- Study setting
- Nearly 4,000 decision-makers across 13 countries
- Boundary of the evidence
- Reported survey patterns are not a causal estimate of adding channels to your programme.
- What we would test in ABM
- Carry the same decision context across the channels your buyers actually use.
Peer-reviewed advertising measurement research · 2017
Ghost Ads: Improving the Economics of Measuring Online Ad Effectiveness
Garrett A. Johnson, Randall A. Lewis and Elmar I. Nubbemeyer
Ghost ads identify control counterparts of exposed users to improve experimental measurement.
Study context and ABM application
- Study setting
- Randomised advertising experiments with counterfactual exposure records
- Boundary of the evidence
- This requires platform support; ordinary CRM records cannot recreate ghost-ad allocation.
- What we would test in ABM
- Preserve assignment and eligibility before interpreting campaign exposure.
ICML research paper · 2017
On Calibration of Modern Neural Networks
Chuan Guo, Geoff Pleiss, Yu Sun and Kilian Q. Weinberger
Classification confidence and actual correctness were misaligned; post-processing could improve calibration.
Study context and ABM application
- Study setting
- Image and document classification experiments
- Boundary of the evidence
- These datasets do not validate a particular ABM propensity model.
- What we would test in ABM
- Test whether score bands correspond to observed outcomes in later cohorts.
Peer-reviewed evaluation-method research · 2015
The Precision-Recall Plot Is More Informative than the ROC Plot When Evaluating Binary Classifiers on Imbalanced Datasets
Takaya Saito and Marc Rehmsmeier
Precision–recall exposes the reliability of positive predictions under class imbalance.
Study context and ABM application
- Study setting
- Imbalanced binary-classification examples
- Boundary of the evidence
- Precision depends on prevalence; comparisons need the same evaluation population.
- What we would test in ABM
- Report useful accounts found within the sales team’s available capacity.
Technical specification overview · 2013
PROV-Overview
World Wide Web Consortium
PROV represents relationships among entities, activities and responsible agents.
Study context and ABM application
- Study setting
- The W3C PROV family of provenance specifications
- Boundary of the evidence
- This is a data specification, not an experiment demonstrating commercial performance.
- What we would test in ABM
- Record the source, transformation and owner behind an account assertion.
Peer-reviewed statistical methods paper · 2019
Metalearners for estimating heterogeneous treatment effects using machine learning
Sören R. Künzel, Jasjeet S. Sekhon, Peter J. Bickel and Bin Yu
The paper develops approaches for estimating how treatment effects differ across contexts.
Study context and ABM application
- Study setting
- Methods and examples for conditional treatment-effect estimation
- Boundary of the evidence
- Causal identification assumptions and adequate data remain necessary.
- What we would test in ABM
- Distinguish likely converters from accounts whose outcomes contact might change.
Cross-sectoral risk-management guidance · 2024
Generative Artificial Intelligence Profile: NIST AI 600-1
National Institute of Standards and Technology
The profile describes generative-AI risks and proposed management actions.
Study context and ABM application
- Study setting
- Companion profile to the AI Risk Management Framework
- Boundary of the evidence
- Guidance is not a product certification or an ABM performance benchmark.
- What we would test in ABM
- Connect material failure modes to owners, checks and monitored evidence.
Peer-reviewed marketing research synthesis · 2016
Understanding Customer Experience Throughout the Customer Journey
Katherine N. Lemon and Peter C. Verhoef
The review connects experience over time with multiple touchpoints and organisational functions.
Study context and ABM application
- Study setting
- Synthesis of customer-experience and journey research
- Boundary of the evidence
- A research synthesis does not provide a fixed enterprise conversion formula.
- What we would test in ABM
- Design evidence and ownership around a buyer task across touchpoints.
Peer-reviewed field study · 2025
Generative AI at Work
Erik Brynjolfsson, Danielle Li and Lindsey Raymond
AI assistance increased issues resolved per hour by 15% on average.
Study context and ABM application
- Study setting
- 5,172 customer-support agents
- Boundary of the evidence
- Customer support with substantial differences between workers. This is not an ABM revenue lift or an Outsell result.
- What we would test in ABM
- Test accepted account work per hour, including verification and rework.
Peer-reviewed randomised experiment · 2026
Navigating the Jagged Technological Frontier
Fabrizio Dell’Acqua and co-authors
Within the tested AI capability boundary, participants completed 12.2% more tasks and finished 25.1% more quickly. Outside it, correctness fell by 19 percentage points.
Study context and ABM application
- Study setting
- 758 BCG consultants
- Boundary of the evidence
- Consulting tasks with GPT-4 in 2023; final paper published March 2026. Task productivity is not pipeline growth.
- What we would test in ABM
- Evaluate extraction, drafting and strategic inference separately.
Randomised controlled trial · 2025
Early-2025 AI and Experienced Developer Productivity
Joel Becker, Nate Rush, Beth Barnes and David Rein
Developers took 19% longer with the early-2025 AI tools in the studied repositories.
Study context and ABM application
- Study setting
- 16 experienced developers; 246 tasks
- Boundary of the evidence
- A narrow coding setting. A February 2026 follow-up reports possible speedups but serious selection and measurement limits. This is not a timeless verdict on AI.
- What we would test in ABM
- Measure elapsed work rather than whether automation feels faster.
Peer-reviewed comparison with field experiments · 2019
A Comparison of Approaches to Advertising Measurement
Brett Gordon, Florian Zettelmeyer, Neha Bhargava and Dan Chapsky
Observational methods often did not recover the effects found by randomised experiments.
Study context and ABM application
- Study setting
- 15 US Facebook advertising experiments
- Boundary of the evidence
- Consumer advertising on one platform. This informs measurement design, not the effect size of an enterprise ABM programme.
- What we would test in ABM
- Separate touchpoint attribution from incrementality; use account-level holdouts where feasible.
KDD research paper and benchmark · 2024
GEO: Generative Engine Optimization
Pranjal Aggarwal and co-authors
Content interventions improved the benchmark’s visibility measure by up to 40%, with effects varying by domain.
Study context and ABM application
- Study setting
- GEO-bench across multiple query domains
- Boundary of the evidence
- A benchmark result, not a forecast of traffic, live AI citations or Google rankings for this website.
- What we would test in ABM
- Make answers extractable and verifiable, then observe actual citations and referrals.
HICSS conference paper; critical discourse analysis · 2022
AI Agents, Humans and Untangling the Marketing of Artificial Intelligence in Learning Environments
Isabel Pedersen and Ann Hill Duin
The paper examines how promotional language presents AI agency and human roles in learning.
Study context and ABM application
- Study setting
- Corporate representations of AI in learning environments
- Boundary of the evidence
- It is not an agency performance experiment and provides no Outsell-versus-competitor or 2× result.
- What we would test in ABM
- Describe the actual agent task, evidence and decision owner.
