A 16-person AI agency in Portland had been growing revenue steadily for two years. Revenue was up 40% year-over-year. But profit margins were shrinking. The founder could not figure out why. When they finally implemented time tracking and analyzed three months of data, the answer was obvious: their three largest clients were consuming 70% of the team's capacity but generating only 45% of revenue. Two of those clients had fixed-fee contracts that were massively underpriced because nobody had tracked how much time the projects actually required. One project that was sold for $60,000 had consumed $93,000 in labor costs. That single project wiped out the profit from two other successful engagements.
Within six months of implementing time tracking, the agency renegotiated two contracts, adjusted their pricing model for new clients, and saw profit margins increase from 12% to 24%. The data from time tracking did not just reveal the problem. It gave them the evidence they needed to fix it.
Why AI Agencies Resist Time Tracking (And Why They Are Wrong)
Let us address the elephant in the room. Many AI agency founders and team members dislike time tracking. The objections are predictable:
"We are not a traditional agency. Time tracking is for law firms." AI agencies face the same fundamental challenge as any professional services firm: your primary cost is people, and you need to understand how those people spend their time to run a profitable business. The industry does not matter. The math does.
"It kills creativity. Engineers and data scientists need unstructured time." Good time tracking does not mean accounting for every minute. It means understanding, at a project level, where effort goes. Tracking to the half-hour or hour level is sufficient. Nobody is asking your ML engineers to log bathroom breaks.
"Our people will feel micromanaged." This is a communication problem, not a time tracking problem. If you frame time tracking as surveillance, people will resist it. If you frame it as a tool that helps the agency price projects correctly, allocate resources fairly, and identify overloaded team members before they burn out, they will support it. The framing is everything.
"We bill fixed fees, so hours do not matter." Fixed fees make time tracking more important, not less. Without tracking, you have no idea whether your fixed fees are profitable. You are running your business on gut feel, and gut feel does not scale.
What to Track (And What Not To)
Effective time tracking requires the right level of granularity. Too detailed and nobody will do it consistently. Too coarse and the data is not useful.
Track at the Project and Category Level
For every entry, capture three things:
- Which project (or internal initiative) the time was spent on
- Which category of work it was (see categories below)
- How long (in half-hour increments, rounded)
That is it. No task-level tracking. No minute-by-minute logs. Project and category is the sweet spot.
Standard Time Categories for AI Agencies
Define consistent categories that every team member uses:
- Discovery and scoping: Client meetings, requirements gathering, data assessment, proposal development
- Data engineering: Data collection, cleaning, transformation, pipeline development, data quality work
- Model development: Architecture design, training, evaluation, iteration, experiment management
- Application engineering: API development, UI development, integration, testing
- Infrastructure: Cloud setup, CI/CD, monitoring, security, DevOps
- Project management: Planning, status updates, internal coordination, documentation
- Client communication: Client meetings, email, presentations, demos (outside of discovery)
- Quality assurance: Testing, code review, performance validation, security review
- Internal operations: Hiring, training, process improvement, tool management
- Business development: Sales calls, proposals, networking, marketing content
- Learning and development: Training, conferences, research, skill building
Every hour should fit into one of these categories. If it does not, add a category. But resist the urge to create 30 categories because adoption drops with complexity.
What Not to Track
- Breaks and personal time. You are tracking work effort, not attendance.
- General overhead. If someone spends 15 minutes reading industry news, that does not need its own time entry. It rounds into learning and development or gets absorbed.
- Granularity below 30 minutes. A 10-minute Slack conversation with a client is not worth a separate entry. Batch small activities into the dominant category for that time block.
Choosing the Right Time Tracking Tool
The best time tracking tool is the one your team will actually use. Prioritize ease of entry over feature richness.
Recommended Tools for AI Agencies
Harvest. The most popular choice for professional services firms. Clean interface, good project and category structure, built-in invoicing, solid reporting. Integrates with most project management and accounting tools. Around $11 per user per month.
Toggl Track. Excellent for teams that prefer timer-based tracking. Start a timer when you begin working, stop it when you switch tasks. The browser extension and desktop app make this frictionless. Better for teams that context-switch frequently. Around $10 per user per month.
Clockify. Free for unlimited users with basic features. The paid plans add reporting and project management features. Good starting point for agencies that want to test time tracking before committing to a paid tool.
Float. Combines time tracking with resource planning. Shows you not just where time was spent but how it compares to planned allocations. Valuable for agencies with 10+ people where resource planning becomes complex.
Integration Requirements
Your time tracking tool should integrate with:
- Your project management tool (Asana, Linear, Jira, Monday) so team members can track time against specific projects without duplicating data
- Your accounting software (QuickBooks, Xero) for labor cost calculations
- Your payroll system for hourly employee and contractor payments
- Your invoicing system (if you bill hourly) for direct invoice generation from time entries
Building the Time Tracking Habit
The biggest challenge with time tracking is not the tool or the process. It is building the habit. Here is a phased approach that works.
Week 1-2: Introduction and Grace Period
- Announce time tracking to the team. Explain why you are doing it (profitability insight, better pricing, workload visibility) and why it benefits them (fairer workload distribution, evidence for hiring, data for pricing).
- Set up accounts for everyone. Do a 30-minute live walkthrough of the tool.
- Ask everyone to track their time for two weeks. Commit to not analyzing or acting on the data during this period. This is purely about building the habit.
- Send a friendly daily reminder at 4:30 PM: "Quick reminder to log your time for today."
Week 3-4: Refinement
- Review adoption rates. Who is tracking consistently? Who is not? Talk to the people who are struggling and find out why.
- Adjust categories based on feedback. If a category is too broad or too narrow, fix it.
- Start a weekly five-minute time entry review where each person verifies their entries for the week are complete and accurate.
Month 2: Integration Into Workflow
- Remove the daily reminders and shift to a weekly reminder for anyone who has missing entries.
- Begin sharing aggregated data with the team. Show project-level breakdowns. Let people see how their time maps to projects.
- Start using the data in project retrospectives. Compare estimated hours to actual hours for completed projects.
Month 3 and Beyond: Operational Standard
- Time tracking is now a normal part of operations. New hires learn it during onboarding.
- Data feeds into project budgeting, pricing decisions, and capacity planning.
- Monthly profitability analysis uses time data as a primary input.
Dealing with Resistance
Some team members will resist longer than others. Common patterns and responses:
The "I forgot" pattern. Someone who consistently forgets is not opposed to tracking. They need a better trigger. Suggest they track time when they switch contexts (close a project tab, open a new one) rather than trying to reconstruct the day at 5 PM.
The "too busy" pattern. Ironic but common. The busiest people resist tracking because they feel overwhelmed. Point out that tracking takes 2-3 minutes per day and that the data will help identify their overload so the team can redistribute work.
The "this is beneath me" pattern. Senior engineers or data scientists sometimes view time tracking as administrative busywork. Reframe it: you are not tracking their time because you do not trust them. You are collecting data that helps the business make decisions that affect everyone, including them.
The principled objector. Some people have a genuine philosophical opposition to time tracking. Hear them out. If their objection is reasonable, accommodate where you can (broader categories, end-of-day logging). If they simply refuse, this is a management conversation about team expectations, not a time tracking conversation.
Analyzing Time Data for Profitability
Once you have 8-12 weeks of data, you can start making meaningful business decisions. Here are the analyses that matter most.
Project Profitability Analysis
For each completed or active project, calculate:
- Total hours invested (from time tracking)
- Blended cost per hour (total compensation for involved team members, divided by their available hours)
- Total labor cost (hours multiplied by blended cost)
- Revenue from the project (contract value or billed amount)
- Gross margin (revenue minus labor cost, divided by revenue)
Healthy AI agency project margins range from 40% to 60%. Below 30% is a problem. Below 20% means you are losing money when you factor in overhead.
Run this analysis monthly for all active projects. Rank projects by margin. The bottom three projects tell you where to focus your attention.
Client Profitability Analysis
Some clients are more profitable than others, not just because of pricing but because of the effort they require:
- Time spent on client communication relative to project size. High-maintenance clients consume disproportionate PM and communication hours.
- Scope creep measurement. Compare hours budgeted to hours consumed at each project stage. Clients who consistently require more than budgeted are costing you.
- Rework percentage. What fraction of time on a client's projects is rework versus forward progress?
Use this analysis to segment clients into tiers. Your most profitable clients deserve priority access to your best team members. Your least profitable clients need contract renegotiation, scope discipline, or graceful offboarding.
Role and Department Analysis
Understand how different roles spend their time:
- Billable percentage by role. What percentage of each person's time is on client projects versus internal work? Data scientists should be 65-75% billable. Project managers might be 50-60% billable. Founders are often 20-30% billable and that is fine.
- Category distribution by role. Are your ML engineers spending 40% of their time on data cleaning? Maybe you need a dedicated data engineer. Is your project manager spending 30% of time on technical troubleshooting? That is a process problem.
- Overtime and overwork. Who is consistently logging more than 40-45 hours? This is your burnout early warning system.
Estimation Accuracy Analysis
For completed projects, compare estimated hours to actual hours at the category level:
- We estimated 80 hours of data engineering and spent 120 hours. Our data engineering estimates are 33% low.
- We estimated 60 hours of model development and spent 55 hours. Our model development estimates are pretty accurate.
- We estimated 20 hours of client communication and spent 45 hours. We are dramatically underestimating communication overhead.
These insights directly improve your budgeting accuracy for future projects.
Common Time Tracking Mistakes
Tracking time but never analyzing it. If nobody looks at the data, nobody will bother entering it accurately. Commit to monthly analysis and share the findings.
Making time tracking punitive. If you use time data to criticize people for being "too slow," you will destroy the system. People will either stop tracking honestly or pad their entries. Use the data for process improvement and pricing, not individual performance judgment.
Tracking at too granular a level. Task-level time tracking for AI work is unreliable and creates resistance. Stick to project and category.
Ignoring non-billable time. Internal operations, business development, learning, and administrative work are real time consumers. If you only track billable time, you underestimate true project costs by 20-30%.
Waiting for perfect data before acting. Your first few months of data will have gaps and inconsistencies. That is fine. Even imperfect data is dramatically better than no data. Act on the directional insights and improve data quality over time.
Treating all hours as equal. An hour from your most senior ML engineer costs significantly more than an hour from a junior developer. Use role-specific rates in your profitability calculations.
Setting Utilization Targets
Utilization rate, the percentage of available hours spent on billable client work, is your most important operational metric. Set realistic targets by role.
- Data scientists and ML engineers: 65-75% utilization target. The remaining time goes to learning, internal tools, and mentoring.
- Software engineers: 70-80% utilization target. More predictable work allows higher utilization.
- Project managers: 50-65% utilization target. PM work includes significant non-billable coordination and process work.
- Solutions architects: 40-55% utilization target. Heavy involvement in pre-sales, scoping, and cross-project work.
- Founders and leaders: 20-40% utilization target. Significant time goes to sales, strategy, hiring, and operations.
Agency-wide target: Most healthy AI agencies target 60-70% average utilization across the team. Below 55% means you are overstaffed or underselling. Above 80% means people are burning out and you need to hire.
Track utilization monthly and discuss it in your operations review. It is the leading indicator of both profitability and team health.
Your Next Step
Pick a time tracking tool today, set up your projects and categories, and start tracking this week. Do not wait for perfect setup. Do not spend two weeks configuring the tool. Create your project list, create the eleven categories listed above, invite your team, and start logging. After 30 days of data, run the project profitability analysis on your three largest current projects. The numbers will either confirm that your pricing is working or reveal exactly where you are leaking money. Either way, you will have information that changes how you run your business. The only way to get that information is to start tracking.