Advanced Invoicing Techniques for Ai & Machine Learning

Advanced Invoicing Techniques for AI & Machine Learning Freelancers

• Segment your projects: Break down large AI/ML projects into smaller, manageable phases (e.g., Data Exploration, Model Prototyping, Model Training, Deployment, Monitoring). Invoice per phase completion to provide clear payment milestones, which can also help track progress when working remotely from places like Taipei or Lisbon.
• Document everything religiously: Keep detailed records of time spent on data cleaning, model experiments, client communications, and any scope changes. This documentation is your strongest ally when invoicing for unexpected work or justifying cost adjustments. Tools for time tracking are discussed in our article, The Best Productivity Tools for Remote Teams.
• Establish clear change order processes: Define how scope changes will be handled and priced before the project begins. This avoids awkward conversations later and ensures you're compensated for additional work. By acknowledging and proactively addressing these unique challenges, AI/ML freelancers can create invoicing strategies that accurately reflect their effort, expertise, and the inherent value of their specialized services. ## Structuring AI/ML Project Agreements for Optimal Invoicing The invoice isn't just a bill; it's a reflection of the underlying project agreement. For AI/ML projects, a well-structured agreement is paramount to facilitate smooth invoicing and prevent disputes. Standard contracts often fall short in accounting for the technical nuances, IP considerations, and iterative nature of AI/ML work. Therefore, crafting agreements that specifically address these factors is essential for any digital nomad or freelancer in this field, whether they are based in Berlin or working from a beachfront in Bali. Before any code is written or data is touched, a Statement of Work (SOW) or a detailed service agreement should be established. This document needs to go beyond typical project descriptions. It should clearly define the scope of work in terms of AI/ML specific deliverables, such as "deliver a trained machine learning model capable of X prediction with Y accuracy," rather than vague objectives. It should also outline the data requirements and responsibilities, specifying who provides what data, its expected format, quality, and privacy considerations. Clauses around data preparation time and potential surcharges for data needing extensive cleaning are crucial. Milestone-based invoicing is particularly effective for AI/ML projects. Instead of a single upfront payment or continuous hourly billing, break the project into distinct, measurable stages with associated payments. Examples of relevant milestones might include:

1. Data Procurement & Preparation Complete: Payment upon successful ingestion and initial cleaning of client-provided data.

2. Exploratory Data Analysis (EDA) & Feature Engineering Complete: Payment after delivering insights from data and defining key features.

3. Model Prototyping & Baseline Performance Established: Payment upon demonstrating a working model prototype and initial performance metrics.

4. Model Training & Optimization Complete: Payment for reaching agreed-upon performance targets in a controlled environment.

5. Model Deployment & Integration Support: Payment upon successful integration of the model into the client's system.

6. Post-Deployment Monitoring & Iteration Plan: Final payment or recurring payment for ongoing support. This approach provides both the freelancer and the client with clear financial checkpoints and encourages focus on achieving specific project outcomes. It also helps manage cash flow effectively, especially for longer engagements. Another critical element is the inclusion of change order clauses. AI/ML projects often evolve. Clients might request new features, different prediction targets, or changes to evaluation metrics based on early results. A clear process for documenting, agreeing upon, and pricing these changes before committing to them is vital. This usually involves a written change request, an updated SOW addendum, and agreement on additional fees and timeline adjustments. Without such a clause, scope creep can quickly erode profitability. For guidance on managing project scope, refer to our article on Project Management for Remote Teams. Finally, the agreement must explicitly address intellectual property (IP) ownership. Will the client own the entire trained model, including the custom-written training scripts and algorithms? Or will the freelancer retain ownership of proprietary algorithms developed during the project, licensing them to the client? These details have significant financial implications and must be ironed out upfront. For instance, a freelancer might charge a higher initial fee if they are completely relinquishing IP, or a recurring license fee if maintaining ownership of core components. This is a topic thoroughly discussed in our guide on Legal Considerations for Remote Workers. ### Example Agreement Structure Elements: * Project Title & Objectives: Clear, concise problem statement and desired outcomes.

• Scope of Work (Detailed): Specific AI/ML tasks, deliverables, and exclusion criteria.
• Data Requirements: Client responsibilities for data provision (format, volume, quality). Penalties or extra fees for non-compliance.
• Milestones & Deliverables: Itemized list of stages, specific outputs for each, and associated payment schedule.
• Performance Metrics & Acceptance Criteria: How success will be measured (e.g., 90% accuracy, F1 score of 0.85).
• Change Management Process: Procedure for requesting, approving, and pricing scope changes.
• Intellectual Property Rights: Clear definition of ownership for data, algorithms, models, and code.
• Payment Terms: Net 30, payment methods, late payment penalties.
• Termination Clauses: Conditions under which either party can end the agreement. By diligently structuring these agreements, AI/ML freelancers can ensure that their invoicing process is transparent, fair, and aligned with the value they provide, protecting both their interests and their client relationships. ## Advanced Billing Models for AI/ML Engagements Moving beyond straightforward hourly or fixed-price models is crucial for AI/ML freelancers. The unpredictable nature, high value, and iterative processes of these projects demand more sophisticated billing approaches. These advanced models not only ensure fair compensation but can also incentivize performance and align interests more closely with clients. ### 1. Value-Based Pricing Instead of charging for time or components, value-based pricing focuses on the measurable business impact your AI/ML solution delivers. This is particularly relevant when your work directly contributes to significant revenue increases, cost reductions, or improved efficiency for the client. For instance, if your prediction model helps an e-commerce company increase conversion rates by 5%, you might negotiate a fee that is a percentage of that increased revenue, or a substantial fixed fee reflecting the expected gains. * Pros: Potentially much higher income if successful, strong alignment with client goals, positions you as a strategic partner.
• Cons: Requires clear metrics for success, client willingness to share sensitive financial data, and a higher risk for the freelancer if the project underperforms.
• Actionable Advice: Before pitching value-based pricing, conduct a thorough analysis of the client's potential ROI. Clearly define the KPIs and how your model's impact will be measured. Consider a tiered structure where a baseline fee is paid, with a bonus upon achieving specific performance milestones. ### 2. Retainer-Based Models with Scope Flexibility For ongoing AI/ML support, maintenance, model retraining, or continuous refinement, a retainer-based model works well. This guarantees the freelancer a consistent income stream and provides the client with consistent access to specialized expertise. Unlike a simple monthly fee, an AI/ML retainer often includes a block of hours and specific deliverables, with flexibility for minor scope adjustments inherent to the field. * Pros: Predictable income for the freelancer, consistent support for the client, encourages long-term relationships, can cover R&D time effectively.
• Cons: Can be challenging to define the exact "scope" covered by the retainer, requiring careful management to avoid expectation misalignment.
• Actionable Advice: Clearly define what the retainer covers (e.g., 40 hours/month for model monitoring and minor feature enhancements, or 20 hours/month for algorithm optimization and weekly reports). Establish an hourly rate for work exceeding the retainer's scope. Consider different retainer tiers based on the level of engagement desired by the client. For managing these recurring tasks, explore tools mentioned in Automating Your Freelance Business. ### 3. Performance-Based Payments / Royalty Models This model involves a lower fixed fee, or even no upfront fee, combined with a performance-based bonus or a royalty derived from the use or success of the AI/ML solution. This is especially suitable for projects where the outcome is uncertain but potentially very lucrative. Examples include a percentage of sales generated directly by a recommendation engine, a per-transaction fee for a fraud detection system, or a share of cost savings from an optimization algorithm. * Pros: Extremely high alignment of interests, low upfront risk for the client, potentially very rewarding for the freelancer.
• Cons: Significant financial risk for the freelancer, requires tracking and reporting mechanisms, client might be hesitant to share revenue data.
• Actionable Advice: This model requires a very high level of trust and transparent data sharing. Ensure detailed legal agreements are in place defining how performance will be measured, reported, and how royalties or bonuses will be calculated and paid. Only consider this for projects where you have high confidence in your ability to deliver measurable, significant results. ### 4. Hybrid Models Often, the most effective approach is a hybrid model combining elements of the above. For instance: * Fixed Fee for Development + Performance Bonus: A fixed price for building and deploying a baseline model, with a bonus if it achieves higher-than-expected performance metrics.
• Hourly for R&D + Value-Based for Production: Hourly billing for the initial exploratory data analysis and algorithm prototyping, transitioning to a value-based or performance-based model once a production-ready solution is identified and deployed. This can be suitable for long-term projects with evolving requirements, perhaps involving teams spread across cities like Mexico City and Bangkok.
• Retainer for Core Maintenance + Project-Based for New Features: A monthly retainer for ongoing model monitoring, basic maintenance, and bug fixes, combined with separate project-based pricing for developing and integrating new features or significant model upgrades. ### Choosing the Right Model: The best billing model depends heavily on the project's specifics, the client's risk appetite, and your confidence in delivering measurable results. Always discuss these options transparently with your client, explaining the pros and cons of each, and choose the model that best aligns with the project's goals and encourages a successful partnership. For more on tailoring your services, see our article Niche Specialization for Digital Nomads. ## Invoicing for Data-Related Services Data is the lifeblood of AI and Machine Learning. Without quality data, even the most sophisticated algorithms are useless. As an AI/ML freelancer, a significant portion of your time might be dedicated to data-related tasks: data acquisition, cleaning, labeling, transformation, and feature engineering. Standard invoicing often overlooks the true effort involved in these critical preparatory steps. Therefore, developing specific invoicing strategies for data services is paramount, especially when working on complex projects from locations like Barcelona or Ho Chi Minh City. The first step is to clearly itemize data-related services on your invoice. Do not simply bundle "model development" into one line item. Break it down so the client understands where the effort is going. ### Key Data Services to Itemize: 1. Data Acquisition & Collection Strategy: If you're involved in advising on how to collect data, sourcing external datasets, or setting up data pipelines, this needs to be billed. This could involve exploring publicly available datasets, strategizing with the client on internal data generation, or integrating with APIs. Billing Method: Often hourly or as a fixed project fee if a clear strategy report is the deliverable. Example Invoice Line: "Data acquisition strategy consultation & external dataset research" 2. Data Cleaning & Preprocessing: This is often the most time-consuming part of an AI/ML project. It involves handling missing values, standardizing formats, removing outliers, correcting inconsistencies, and dealing with duplicates. Billing Method: Typically hourly. For very large datasets or particularly messy ones, you might propose a fixed fee per chunk of data (e.g., "$X per 100,000 rows cleaned"). Example Invoice Line: "Data cleaning and standardization for customer transaction dataset (1M records)" 3. Data Labeling & Annotation: For supervised learning, data needs to be meticulously labeled. If you or your team are performing this, or managing a labeling service, this is a distinct, billable service. Billing Method: Can be hourly, per item/unit labeled, or as a fixed fee for a specific batch. Example Invoice Line: "Image annotation for object detection model (5,000 images @ $0.50/image)" 4. Feature Engineering: This is where raw data is transformed into features that can be used by a machine learning algorithm, often requiring deep domain expertise and creativity. This can involve creating new variables, combining existing ones, or applying complex transformations. Billing Method: Typically hourly, as it often involves experimentation and iteration. Example Invoice Line: "Feature engineering for sales prediction: creation of lagging indicators & seasonal variables" 5. Data Augmentation: For tasks like image or text processing, artificially expanding the training dataset by creating modified versions of existing data can be crucial for model performance. Billing Method: Usually hourly, or a fixed fee for a specific augmentation pipeline. Example Invoice Line: "Image data augmentation for anomaly detection model (rotation, scaling, noise addition)" 6. Data Privacy & Compliance Consultation: Advising on GDPR, CCPA, or other data privacy regulations related to the dataset being used, or setting up anonymization techniques, is a highly specialized skill. Billing Method: Hourly, or a fixed fee for a compliance report/plan. Example Invoice Line: "Consultation on GDPR compliance for customer data processing" ### Practical Tips for Invoicing Data Services: * Estimate Realistically: Data cleaning and preparation almost always take longer than expected. Pad your estimates for these activities.
• Data Audit First: If possible, conduct a small pre-project data audit to understand the data's quality and complexity. This allows for a more accurate quote. For deep dives into data, consider specialist roles from our Talent pool.
• Set Clear Data Deliverables: For data cleaning, specify the "clean" state (e.g., "no missing values in critical columns A, B, C; all dates standardized").
• Client Responsibilities Clearly Documented: If the client is responsible for data provision, clearly state the expected format, quality, and timelines. Include clauses for additional charges if the provided data falls short of expectations, requiring you to spend unexpected time on remediation.
• Use Sub-Invoices: For large data projects, consider submitting separate "Data Preparation" invoices before moving onto "Model Development" to show progress and manage cash flow. This is a common practice for freelancers working globally, whether they are in Sydney or Rio de Janeiro. By meticulously itemizing and accurately pricing your data-related services, you ensure that you are compensated for the foundational work that makes AI/ML projects possible, rather than having it disappear into a generic "development" cost. ## Incorporating Intellectual Property (IP) and Licensing Fees For AI/ML freelancers, Intellectual Property (IP) is often the crown jewel of their work. The algorithms, models, unique data processing techniques, and even custom-developed frameworks can be highly valuable assets. Failing to address IP ownership and licensing explicitly in your contracts and subsequent invoicing can lead to significant financial loss and legal disputes. This section delves into how to incorporate IP considerations and potential licensing fees into your advanced invoicing practices. The foundational principle is that, by default, the creator of intellectual property owns it. This means that when you develop an algorithm or train a model for a client, you, as the freelancer, generally own that IP unless a contract explicitly states otherwise. Most clients, however, will want to own the work they pay for. This tension needs to be resolved transparently and commercially in your agreement, which then dictates your invoicing. ### Types of IP Ownership & Licensing Arrangements: 1. Work for Hire (Client Owns All IP): Description: The most common scenario where the contract states that all IP created during the project becomes the sole property of the client. Invoicing Impact: You should charge a higher overall project fee to compensate for relinquishing all ownership. This higher fee reflects the transfer of a valuable asset. The invoice itself won't have separate IP line items, but the project cost is elevated. Actionable Advice: Ensure your contract explicitly uses "Work for Hire" language. Make sure the compensation reflects the full value of the IP. 2. Freelancer Retains IP, Grants Client a License: Description: You retain ownership of your core algorithms, unique model architectures, or generalizable code components. You then grant the client a non-exclusive, perpetual, worldwide license to use the specific implementation you delivered for their business purposes. Invoicing Impact: This is where separate IP-related charges appear. Initial Licensing Fee: A one-time fee for the right to use your IP. This can be substantial. Recurring License Fee: An annual or monthly fee for continued use of the licensed components, similar to software as a service (SaaS) model. Usage-Based Licensing: Fees based on the model's usage (e.g., per prediction API call, per user served). Actionable Advice: Clearly define the scope of the license (e.g., internal use only, specific product lines, non-transferable). This model works well for building generalizable AI tools or components you can re-use. Your invoice would show "Model Development" and then "Perpetual License for [Algorithm Name]" or "Annual License Fee for Predictive Engine." 3. Hybrid Approach / Core Components vs. Customization: Description: You might use your proprietary frameworks or base models (which you own) and then customize them for a client's specific data and needs (which the client owns). Invoicing Impact: The invoice might include a fee for customizing the model (work for hire) and a lower recurring license fee for the underlying proprietary components. Actionable Advice: This is a sophisticated approach. Document which parts are your core IP and which are client-specific customizations. This is particularly relevant when specializing in a certain niche, as discussed in Building a Niche for Remote Work. ### Practical Tips for Invoicing with IP Considerations: * Explicit Contract Language: Never assume. Ensure your service agreement clearly defines IP ownership. Consult a legal professional to draft or review these clauses; this is one of the important tips covered in Legal Considerations for Remote Workers.
• Itemize on Invoices: If charging for licenses, break it out clearly. An invoice should not just say "AI Solution Development." It should differentiate between "Development Services (Work for Hire)" and "Licensing Fee for X Algorithm (Freelancer's IP)."
• Valuation of IP: How do you price your IP or the right to own it? Consider: Market demand: How unique or powerful is your algorithm? Future revenue potential: How much money could the client make from this IP? Your R&D investment: How much time and effort went into creating the underlying IP? Exclusivity: If the client wants exclusive rights, the fee will be much higher.
• Protect Your IP: Register copyrights where applicable (e.g., for unique code), and consider non-disclosure agreements (NDAs) early in the client engagement process.
• Review Regularly: For recurring licenses, have review clauses to adjust fees periodically (e.g., every 2-3 years) to account for inflation or increased value/usage. By strategically incorporating IP ownership and licensing into your project agreements and invoicing, AI/ML freelancers can secure better financial compensation for their expertise and the valuable digital assets they create. This not only protects your interests but also positions you as a savvy business professional who understands the full commercial value of your highly specialized skills. Consider how global locations, such as Dubai or Vancouver, might have different IP laws affecting your agreements. ## Performance-Based Clauses and Bonuses In the results-driven world of AI and Machine Learning, simply billing for hours or fixed deliverables might not fully capture the value of your contribution. Implementing performance-based clauses and bonuses directly ties your compensation to the success and impact of the AI/ML solution you develop. This approach not only incentivizes you to deliver exceptional results but also aligns your financial interests directly with the client's business objectives, fostering deeper partnerships. This is particularly appealing to clients who are looking for demonstrable ROI from their AI investments. A performance-based clause isn't about charging extra for simply doing your job; it's about being rewarded when your work exceeds expectations or produces exceptional, measurable business outcomes. ### How to Implement Performance-Based Clauses: 1. Define Clear, Measurable KPIs (Key Performance Indicators): This is the cornerstone. Before agreeing to any performance clauses, you and the client must jointly define what "success" looks like and how it will be quantitatively measured. Examples of AI/ML KPIs: Accuracy/F1 Score: For classification or prediction models. Conversion Rate Increase: For recommendation engines or personalization AI. Cost Reduction: For optimization models (e.g., logistics, energy consumption). Fraud Detection Rate: For financial anomaly detection systems. Customer Churn Reduction: For predictive customer retention analytics. Processing Time Reduction: For automation or efficiency-focused AI. Revenue Uplift: Attributed directly to the AI system. Actionable Advice: Be specific. Instead of "improved accuracy," state "achieve minimum 90% F1 score on unseen test data, or 15% increase in conversion rate over baseline." 2. Establish a Baseline: Before your AI model is implemented, you need to know the current state. This establishes the "baseline" or control against which your AI's performance will be measured. Actionable Advice: Conduct thorough initial analysis to document the current performance. This might be a billable "Discovery & Baseline Analysis" phase. 3. Set Bonus Triggers and Tiers: Define thresholds for performance that, when met or exceeded, trigger specific bonus payments. Example Tiers: If conversion rate increases by 5%: $5,000 bonus. If conversion rate increases by 7.5%: $10,000 bonus. If conversion rate increases by 10% or more: $20,000 bonus. Actionable Advice: Make the bonus structure clear, transparent, and easy to calculate. Consider caps on bonuses to manage client expectations and potential financial exposure. 4. Define Measurement Period and Reporting: How long will the AI model run before performance is assessed? How frequently will results be reported and verified? Actionable Advice: Specify whether performance will be measured over one month, three months, or six months post-deployment. Outline the reporting cadence (e.g., weekly performance dashboards, monthly review meetings). 5. Include Verification and Audit Clauses: Clients might need internal teams to verify your reported performance. Include clauses about data access for verification and dispute resolution. Actionable Advice: Maintain detailed logs of model performance, A/B test results, and any contributing factors. Be prepared to share anonymized data or aggregated performance metrics. ### Invoicing Impact: * Separate Line Item: Bonuses should appear as a distinct line item on your invoice, clearly labeled "Performance Bonus - [Specific KPI Met]" with a reference to the signed agreement.
• Timing: The bonus invoice would be submitted after the agreed-upon measurement period and verification, separate from your regular project invoices.
• Contractual Basis: Crucially, your initial contract or SOW must explicitly detail these performance clauses, the KPIs, the bonus amounts, and the measurement process. This protects both parties. ### Example Scenario:

You are developing a fraud detection model.

• Base Project Cost: $25,000 for development and deployment.
• KPI: Reduce false positive rate by 20% while maintaining the current true positive rate.
• Bonus Structure: If false positive rate reduces by 15-19.99%: $3,000 bonus. If false positive rate reduces by 20-25%: $7,500 bonus. * If false positive rate reduces by >25%: $15,000 bonus.
• Measurement: After 3 months of live deployment, verified by client's security team. If, after 3 months, your model reduces false positives by 22%, you would submit an invoice for your initial $25,000 (if not already fully paid) plus a separate invoice for $7,500 labeled "Performance Bonus for Fraud Detection Model (false positive reduction 22%)". This approach works particularly well for experienced AI/ML professionals confident in their abilities and willing to tie their income to tangible business outcomes. It transforms your service from a cost center into a direct driver of profitability and innovation for your clients, regardless of whether you're working from Tokyo or Bogota. For more on establishing your expertise, check out Building a Strong Personal Brand. ## Itemization and Transparency for Complex Projects For complex AI/ML projects, a generic invoice simply won't cut it. Clients, especially those less familiar with the intricacies of AI development, need to understand exactly what they are paying for. Detailed itemization and transparency in your invoicing is not just good practice; it's essential for building trust, preventing disputes, and justifying your pricing. It transforms your invoice from a simple request for payment into a clear statement of value delivered. Think of your invoice as a mini Statement of Work that outlines the completed activities and their corresponding costs. This approach is particularly valuable for remote professionals who may not have daily face-to-face interactions with clients. ### What to Itemize and How: 1. Project Phases/Milestones: If you've structured your project into phases (and you should, as discussed earlier), each phase completion should be a primary item on your invoice. Example: "Phase 1: Data Exploration & Feasibility Study Complete" - $X "Phase 2: Model Prototyping & Baseline Development" - $Y 2. Specific AI/ML Tasks: Within each phase, break down the actual work performed into clear, understandable tasks. Data-Related Tasks: "Data Cleansing & Preprocessing (Client Dataset 'Alpha')" - Z hours @ $rate "Feature Engineering for [Specific Model]" - A hours @ $rate "Data Labeling Oversight & Quality Control" - B hours @ $rate Model Development Tasks: "Algorithm Selection & Experimentation (e.g., Logistic Regression, SVM, Neural Network)" - C hours @ $rate "Model Training & Hyperparameter Tuning" - D hours @ $rate "Model Evaluation & Performance Reporting (e.g., F1 score, AUC analysis)" - E hours @ $rate Deployment & Integration: "Model API Development & Integration into Client's CRM" - F hours @ $rate "Containerization (Docker) for Deployment" - G hours @ $rate "Post-Deployment Monitoring Setup" - H hours @ $rate Research & Development: "Literature Review & Custom Algorithm Adaptation for Edge Cases" - I hours @ $rate "Experimentation with Novel Machine Learning Techniques" - J hours @ $rate 3. Tooling & Software Costs (if applicable and agreed upon): If you are passing on costs for specialized software licenses, cloud computing resources (e.g., AWS, GCP, Azure), or data labeling platforms, these must be itemized. Example: "AWS GPU Instance Usage (Training Environment)" - $X (Attach usage report) "Data Labeling Platform Subscription (Month of Oct)" - $Y 4. Consultation & Communication: While often part of project management, for AI/ML, client education and regular syncing are critical. Example: "Weekly Sync Meetings & Progress Reporting" - X hours @ $rate "Client Education Session: Understanding Model Interpretability" - Y hours @ $rate 5. Documentation & Knowledge Transfer: Delivering documentation is a crucial output for AI/ML projects. Example: "Model Documentation & Technical Specifications" - X hours @ $rate "Knowledge Transfer Session for Client's Internal Team" - Y hours @ $rate ### Practical Tips for Enhanced Transparency: Use Descriptive Language: Avoid jargon where possible, or briefly explain it. Instead of "Hyperparam tune," write "Optimizing model settings (e.g., learning rate, epochs) for best performance."
• Reference Timesheets: If billing hourly, state that your invoice is based on detailed timesheets, and offer to provide them upon request. Use a reliable time-tracking tool. Many tools are explored in Productivity Hacks for Remote Workers.
• Attach Supporting Documents: For cloud costs or third-party tools, attach screenshots of usage, invoices from providers, or detailed reports.
• Clearly State Payment Terms: Include payment due date, accepted payment methods, and late payment policies. For international payments, consider platforms that handle multiple currencies, a common need when working from destinations like Prague or Medellin.
• Regular Updates: Don't let your first invoice be the first time the client sees a breakdown of costs. Provide regular progress reports that outline completed tasks and anticipated costs.
• Consistent Formatting: Use a professional, consistent invoice template. This reinforces your professionalism. By providing this level of detail, you demonstrate professionalism, accountability, and a deep understanding of the work being done. It helps clients understand the investment they're making into a complex technology and reduces the likelihood of billing disputes. This detailed approach establishes you as a credible and trustworthy AI/ML expert, essential for securing ongoing projects. ## Managing Retainers and Recurring Services Beyond