SERVICE LEVEL AGREEMENT (SLA)

1. Purpose of the Agreement

This Service Level Agreement (SLA) establishes the terms, conditions, and service quality commitments that DeepDo (a service line of Deeptechnology S.A.S.) offers to its corporate customers (hereinafter “the Customer”) in connection with the use of the Deepdoers platform (AI Employees). In addition to traditional availability and support standards, this document incorporates specific metrics for Artificial Intelligence services: model accuracy, response time, autonomous resolution rate, and model update policies, aligned with international industry best practices.

2. Scope of Application

This SLA applies to the following services provided by DeepDo:

• DeepDo Platform: web application for managing AI Employees.
• Conversational AI Employees (LLM-based).
• Automations and complex AI workflows.
• DeepDo integration API.
• Administration dashboard.
• Technical support and customer care.
• Data storage and processing.

Integrated third-party services (WhatsApp Business API, OpenAI/Anthropic LLM models, etc.) whose availability is governed by their own agreements are excluded from scope.

3. Key Definitions

4. Service Availability

This section defines the guaranteed uptime, calculation formula, scheduled maintenance, and initial calibration period applicable to the DeepDo platform.

4.1. Guaranteed Uptime

DeepDo guarantees a Monthly Uptime Percentage during any billing cycle according to the plan thresholds described below. In the event of non-compliance, the Customer shall be entitled to receive a Service Credit pursuant to section 7. This does not include cases where the Customer turns off the agent in the process, or scheduled maintenance.

The effective credit percentage depends on the level of non-compliance, according to the tiered table in section 7.1. Interruptions caused by infrastructure providers or third-party platforms (including Meta, WhatsApp Business API, language model providers, cloud services, etc.) are excluded from the guaranteed availability calculation and do not give rise to credits for uptime non-compliance. Nevertheless, DeepDo undertakes to notify the Customer within 2–4 hours of detecting the incident, to maintain periodic communication during the contingency, and to actively manage the resolution with the affected provider.

4.2. Availability Calculation Formula

Monthly availability (DI) is calculated using the following formula: DI = [ (N × 24 − T) / (N × 24) ] × 100 Where:

• N = Number of days in the calendar month.
• T = Total hours of unavailability during which the service was completely inoperative (Priority P1).

Only total service unavailability (P1) is counted. Partial degradation incidents (P2) do not count toward the availability calculation. Downtime is counted from DeepDo’s formal confirmation of the incident until service normalization. Scheduled maintenance communicated with 72 hours’ notice is NOT added to the unavailability calculation.

4.3. Scheduled Maintenance

DeepDo will perform scheduled maintenance with a minimum of 72 hours’ prior notice by email and in-platform notification. Maintenance will preferably be executed on Sundays between 2:00 a.m. and 6:00 a.m. (Colombia time). AI model retraining will be notified with 48 hours’ advance notice.

4.4. Initial Calibration Period

During the first 60 calendar days after service go-live, no metric under this SLA —including platform availability (section 4.1) and AI Employee performance metrics (section 5)— will generate credits for non-compliance. This calibration period is necessary for the platform, integrations, and AI model to adjust to the Customer’s operating environment. During this period, DeepDo will work jointly with the Customer to calibrate the AI Employee and reach the guaranteed thresholds.

5. AI Employee Quality and Performance

This section sets out the AI service quality commitment, continuous monitoring, and the model drift and model update policy.

5.1. AI Service Quality Commitment

DeepDo undertakes to keep the AI Employee operating so that it responds in a coherent, timely manner aligned with the knowledge base supplied by the Customer. AI Employee performance depends on factors inherent to DeepDo’s service and on external factors, including the quality and currency of information provided by the Customer, the behavior of third-party language models, and interaction volume. Accordingly, DeepDo does not guarantee absolute accuracy levels, but rather a continuous, good-faith effort aimed at keeping the AI Employee in optimal operating condition.\n\nWithin that commitment, DeepDo will continuously perform the following actions:

• Monitor the overall behavior of the AI Employee and identify response patterns that require adjustment.
• Apply model improvements and adjustments when relevant deviations from expected behavior are identified, within the scope defined with the Customer.
• Inform the Customer of situations that materially affect service quality and that require the Customer’s collaboration to be corrected.

A one-off failure to meet performance expectations does not by itself constitute a breach of this SLA or give rise to credits, unless sustained and documented degradation is verified pursuant to section 5.2.

5.2. Monitoring and Continuous Improvement

For Business and Enterprise plans, DeepDo will conduct periodic reviews of AI Employee performance jointly with the Customer, with the goal of identifying improvement opportunities and adjusting the service to operational needs. These reviews may include conversation analysis, Customer feedback, and knowledge base adjustments. The frequency of formal reviews will be agreed between the parties during onboarding.\n\nIf, as a result of joint monitoring, sustained degradation attributable to factors within DeepDo’s control is identified, the parties will agree on an improvement plan with reasonable deadlines. Only if DeepDo fails to execute such agreed plan within the deadlines agreed in writing may the Customer request a credit pursuant to section 7.2.

5.3. Model Drift and Model Update Policy

AI models may degrade in performance when data patterns change over time (model drift). DeepDo undertakes to:

• Continuously monitor AI Employee behavior to detect signs of degradation.
• Where applicable, initiate a retraining or adjustment process when sustained degradation attributable to drift is confirmed, within a reasonable timeframe according to the complexity of the required adjustment.
• Proactively notify the Customer when significant model degradation is detected.
• Document all model changes (versions, dates, improvements) and share them with the Customer upon request.

Any credits for sustained model degradation are governed by section 7.2 and require a prior written agreement regarding the breached improvement plan.

6. Technical Support

This section defines support channels, hours by plan, incident classification, target resolution times, and the escalation policy.

6.1. Support Channels

• Email: support@deepdo.ai
• Live chat: available from the platform.
• WhatsApp Business and phone line: 315 408 6102 (same number handles calls and WhatsApp messages). Available for Business and Enterprise plans; 24/7 priority for the Enterprise plan.

For critical priority incidents (P1), the recommended channel for Business and Enterprise plans is WhatsApp or the phone line, ensuring immediate attention. For the Starter plan, the P1 incident channel is live chat or email during support hours.

6.2. Support Hours by Plan

6.3. Incident Classification and Response Times

Incidents are classified using an Urgency × Impact Matrix (industry standard):

6.4. Target Resolution Times

6.5. Escalation Policy

When a ticket is not resolved within the target time, the next escalation level is activated automatically:

7. Credits for Non-Compliance

This section establishes the service credits applicable for platform availability non-compliance and for breach of a written AI improvement plan.

7.1. Platform Availability Non-Compliance

If actual monthly availability (DI) is below the Guaranteed Uptime applicable to the Customer’s plan, a Service Credit will be applied according to the following tiered table:

This table applies uniformly to all three plans (Starter, Business, Enterprise). Each plan’s guaranteed uptime (section 4.1) determines the threshold from which credits are activated, but the compensation scale is the same. The maximum credit accumulable per calendar month may not exceed 25 % of the monthly value of the contracted plan, regardless of the number of incidents in that period. Credits are the sole available compensation and are not transferable across months.

7.2. Credit for Breach of an AI Improvement Plan

The only circumstance that gives rise to a credit related to AI Employee performance is DeepDo’s failure to comply with an improvement plan agreed in writing between the parties pursuant to section 5.2. In that case, and provided the breach is attributable to factors within DeepDo’s control, a credit of up to 10 % of the monthly value of the contracted plan will be applied, at DeepDo’s discretion according to the severity and duration of the breach. This credit is not cumulative with platform availability credits (section 7.1), and the combined maximum credit for any reason in a calendar month may not exceed 25 % of the monthly value of the contracted plan.

The absence of a written improvement plan means there is no enforceable threshold under this section; accordingly, no credit will apply for AI Employee performance.

8. Responsibilities of the Parties

This section describes the obligations of DeepDo and the Customer, as well as the preconditions for meeting AI metrics.

8.1. DeepDo’s Responsibilities

• Proactively monitor the AI Employees platform.
• Notify the Customer of scheduled maintenance with at least 72 hours’ notice.
• Proactively notify of AI model degradation (model drift) when detected.
• Execute model retraining or adjustments within the committed timeframes.
• Document all model version changes and share them upon request.
• Ensure that Customer data is used exclusively to operate the contracted services.
• Configure the AI Employee in a closed manner so that it responds solely based on information supplied by the Customer, applying hallucination mitigation mechanisms (RAG, guardrails, contextual validations).

8.2. Customer’s Responsibilities

• Keep contact details and billing information up to date.
• Designate an account administrator and a technical point of contact.
• Report incidents exclusively through official support channels.
• Perform preliminary tests on its own infrastructure before escalating incidents to DeepDo.
• Notify DeepDo of the need for changes in data flows or integrations that may affect AI Employee performance.
• Not share access credentials among unauthorized users.
• Supply, keep updated, and curate the knowledge base and data with which the AI Employee is trained and operated. The quality, accuracy, and currency of this information are the Customer’s exclusive responsibility.
• Where access is available, not modify prompts, flows, or critical AI Employee configurations without coordination with DeepDo, under penalty of suspension of AI metrics guarantees.

8.3. Preconditions for AI Metrics Compliance

The AI metrics thresholds guaranteed in section 5 apply only if the following conditions are met by the Customer:

• The knowledge base supplied is complete, up to date, and free of internal contradictions.
• The Customer has completed the Calibration Period (section 4.4).
• No uncoordinated changes to prompts or AI Employee configurations have been made during the evaluated month.

9. Maintenance vs. Development

To ensure billing transparency and avoid ambiguity regarding the scope of the monthly fee, this section clearly establishes which activities are covered by the Monthly Service (maintenance) and which constitute New Requirements (additional billable development).

9.1. General Criterion

Maintenance means any activity aimed at preserving service operability as originally agreed. Development means any activity that expands, alters, or adapts the service to needs or contexts not contemplated in the original contract scope. As a reference rule, requests requiring more than 4 cumulative engineering hours are classified as development and require a prior quotation.

9.2. What the Monthly Service DOES Include (Maintenance)

The recurring fee covers operational assurance and minor content adjustments:

• Correction of technical errors (bugs) that prevent the originally agreed functionality.
• Updates to text responses (FAQs) that do not alter the conversational flow.
• Diagnosis of disconnections with already integrated third-party APIs.
• Service restoration in the event of server or infrastructure outages.
• Monitoring, minor performance adjustments, and application of security patches.
• Model retraining and adjustments due to model drift, pursuant to section 5.3.

9.3. What It Does NOT Include (New Requirements)

The following requests require quotation and additional billing:

• Structural changes: modification of the decision tree or conversational logic of the AI Employee.
• New business rules: changes to operational validations (e.g., order cutoff times, coverage zones, discount policies, approval thresholds, data requests).
• Development: creation of new AI intents, new data-collection steps, new integrations with external systems, new flows, or new channels.
• Reprogramming due to third-party changes: if an already integrated external API (e.g., Siigo, payment gateways, CRM, WhatsApp Business) changes its technical documentation, deprecates endpoints, or modifies its data contracts, reprogramming the AI Employee code is quoted separately.
• Forced technology migration: if an essential provider discontinues its service, retires model versions (e.g., LLM model deprecation), or radically changes connection protocols, the engineering required to migrate to a new provider or version will be billed as an update project.
• Experience redesign: changes to the AI Employee’s tone, personality, base language, or identity beyond minor copy adjustments.
• Custom reports: reports or dashboards outside the standard set included in the contracted plan.

9.4. Quotation Process

When the Customer requests a New Requirement, DeepDo will follow this process:

• Acknowledgment of receipt of the request within 1 business day.
• Delivery of a formal quotation (scope, estimated effort, delivery time, and price) within 5 business days for standard requirements, or up to 10 business days for complex requirements requiring extended technical analysis.
• Development starts only after the Customer’s written approval of the quotation.
• Developments are billed separately from the monthly fee and do not generate cross-discounts with SLA credits.

9.5. Classification of Ambiguous Reports

When there is doubt as to whether a Customer report constitutes a bug (maintenance) or a scope change (development), DeepDo will perform an initial analysis and deliver its formal classification within 2 business days. It is considered a bug when current behavior differs from what was originally agreed in the contract scope or configuration documentation. It is considered a scope change when current behavior matches what was agreed but the Customer requires different behavior.

Changes, reconfigurations, or adjustments made by the Customer directly on the AI Employee without coordination with DeepDo are not considered bugs, and their correction, if required, will be billed as development.

10. SLA Exclusions

Deeptechnology / DeepDo shall not be liable for SLA non-compliance caused by:

• Force majeure or events beyond reasonable control (disasters, mass internet outages, etc.).
• Failures in the Customer’s infrastructure, connectivity providers (ISP), or third-party tools not managed by DeepDo.
• AI model degradation caused by changes in Customer data that were not communicated to DeepDo.
• Changes to prompts, flows, or configurations made by the Customer without coordination with DeepDo.
• Scheduled maintenance communicated with the established advance notice.
• Denial-of-service (DDoS) attacks of exceptional magnitude.
• Interruptions, capacity limitations, performance degradation, or unavailability of third-party language models (including, without limitation, OpenAI, Anthropic, Google, or other AI infrastructure providers) that affect the service, regardless of duration or frequency. DeepDo does not guarantee AI Employee availability or performance during periods in which such providers experience failures, and such periods will not count as downtime attributable to DeepDo.
• Unpredictable behavior, hallucinations, biases, or incorrect responses generated by third-party language models, whose architecture, training, and parameters are outside DeepDo’s technical control. DeepDo assumes no responsibility for the intrinsic capabilities of such providers or for unilateral changes they make to their models.
• Customer actions that violate the service Terms of Use.
• Errors or inaccuracies arising from the quality of data, the knowledge base, or instructions configured by the Customer.

11. Security, Privacy, and AI Data

DeepDo implements and maintains the following security measures:

• Encryption of data in transit.
• Continuous infrastructure monitoring and threat detection.
• Backups.
• Compliance with Law 1581 of 2012 (Habeas Data Colombia) and applicable regulations.
• The Customer retains ownership of all its data.
• Data deletion procedure available within a maximum of 30 business days after contract termination.

11.1. AI-Generated Content and Hallucinations

DeepDo acknowledges that large language models (LLMs) may occasionally generate inaccurate, incomplete, or incorrect information, a phenomenon known as a “hallucination.” To mitigate this risk, DeepDo implements engineering techniques such as Retrieval-Augmented Generation (RAG), conversational guardrails, contextual validations, and closed configuration of the AI Employee so that it responds exclusively based on information supplied by the Customer.\n\nNotwithstanding the foregoing, the Customer acknowledges and accepts that:

• The accuracy and currency of the information with which the AI Employee is trained and operated are the Customer’s exclusive responsibility.
• DeepDo does not guarantee the absolute accuracy of all responses generated by the AI Employee; it guarantees only the statistical accuracy levels defined in section 5.
• Ultimate responsibility for decisions made based on AI Employee responses rests with the Customer and its end users, especially in sensitive use cases (legal, medical, financial, etc.).
• The Customer is responsible for adequately informing its end users that they are interacting with an Artificial Intelligence system.

12. Term and Modifications

This SLA enters into force as of the service start date and remains in effect throughout the license period. DeepDo may update the terms with at least 30 calendar days’ prior notice. Modifications will be communicated to the account administrator by email and published in the then-current version of the document at www.deepdo.ai. After such 30-day period, if the Customer has not terminated the contract in writing, the modifications shall be deemed automatically accepted and binding on both parties without the need to sign a new document. Failure to meet service levels shall not by itself be grounds for contract termination. It may be considered grounds for early termination only if repeated non-compliance with the same indicator is proven for six (6) or more consecutive months, and only if DeepDo has not submitted an improvement plan accepted in writing within that period.

13. Acceptance

The parties state that they have read, understood, and accepted the full content of this Service Level Agreement, including the specific metrics for Artificial Intelligence services.