deepsense.ai helped an American real estate developer identify where AI automation could deliver practical value across accounting operations, then turned that analysis into working PoCs. In 3.5 weeks, the team mapped four processes, prioritized 10 automation opportunities, built two working automations, and established a repeatable method the client’s employees could use to develop further AI-assisted workflows.
Meet our client
Client:
Industry:
Market:
Technology:
In a Nutshell
Client’s Challenge
The client’s accounting team relied on workflows spanning PDFs, Excel, email and specialist systems. The resulting manual handoffs required employees to repeatedly extract, transfer and re-enter the same information. In one process, employees entered the same invoice data three times.
Automating these workflows was not simply an integration exercise. Key systems presented integration constraints, external portals lacked self-service APIs, document formats varied, and accounting decisions still required human control. The client needed to identify where automation was technically feasible and valuable without introducing unnecessary system changes or removing critical approval gates.
Our Solution
deepsense.ai delivered a monthly process discovery, AI advisory, prioritization, and hands-on PoC development.
The team mapped 4 accounting processes end-to-end, assessed 10 automation candidates, and built and validated two working automations using real monthly data. Rather than forcing deep integration with systems that did not readily support it, deepsense.ai focused automation on the file layer–working across statements, invoices, PDFs, email, and Excel while keeping people in control of consequential decisions.
In parallel, we ran two company-wide enablement sessions and developed a repeatable “session → skill” methodology, supported by a three-wave roadmap for further adoption.
Client’s Benefits
The engagement delivered two working automations at MVP scope, including an Amex reconciliation workflow whose preparation stage went from hours to roughly 20 minutes.
The client also gained documented process knowledge, a prioritized automation backlog, designs for additional workflows, step-by-step run guides, and a repeatable method for developing further automations.
A Deep Dive
1. Overview
We helped a US real estate developer explore AI automation in its accounting operations. The team mapped processes, assessed opportunities, built and tested two PoCs with real monthly data, and created a roadmap for further automation, all within the client’s existing systems.
All in less than a full month.
Engagement Highlights
During this project, deepsense.ai delivered:
- 4 accounting processes mapped end-to-end
- 10 automation candidates assessed and prioritized
- 2 working automations at PoC scope
- 2 company-wide AI enablement sessions
- A three-wave automation roadmap
- A reusable no-code “session → skill” methodology
The monthly card-reconciliation automation reduced the preparation stage from hours to roughly 20 minutes while keeping authority and approval with the user.
2. Client
The client is a US residential developer with residential projects in construction and sales.
The company is operating in a shifting real estate market where AI-assisted operations are becoming a baseline expectation rather than an advantage. The engagement started from a strategic position – keep pace with where the industry is moving – rather than from a single broken process. We chose accounting as the proving ground because the repetitive work there was obvious and measurable
Its operations include in-house accounting, construction, sales and marketing, acquisitions and finance, and a debt fund. The opportunity assessment therefore had potential relevance beyond the initial accounting engagement, although the PoC work described here remained focused on accounting processes.
The in-house team had deep domain knowledge of the accounting processes but no analytical or AI capability to assess which parts could be automated, which is precisely what the engagement provided.
The client uses construction accounting and project management systems alongside Microsoft 365.
3. Challenge
The accounting workflows depended on several systems, but significant operational effort occurred in the gaps between them. Information moved repeatedly between PDFs, spreadsheets, email, and specialist applications. In one workflow, employees entered the same invoice data into the AIA, then a budget tracker, then the invoice system.
Re-keying consumed employee time and created room for errors, but the obvious fix was not available: the core systems offered limited integration options, and some steps involved judgment or authority that needed to remain with employees.
The challenge was therefore not simply to automate existing steps. It was to determine where automation should sit in the workflow, what AI could reliably handle, what needed to remain human-controlled, and which opportunities justified implementation first.
4. Solution
deepsense.ai delivered a combined advisory and PoC engagement, bringing together process discovery, automation opportunity assessment, AI engineering, validation, enablement, and roadmap development.
Five principles shaped the approach:
- Enter once, reuse everywhere. One upstream extraction should support multiple downstream uses.
- Target PDF ↔ Excel workflows. Many of the highest-friction steps involved transferring information between documents and spreadsheets.
- Address the source of repetitive work. Some downstream tasks, such as chasing Amex documentation, resulted from inconsistent upstream intake.
- Keep authority gates human. AI prepares, extracts and validates; people approve.
- Automate where formats support reliability. Structured and consistent documents were stronger automation candidates, while highly variable formats remained assisted workflows.
Core Components
Amex Combine & Reconciliation Preparation
deepsense.ai built a user-facing skill that combines statements from four cardholders into a single sorted, color-coded spreadsheet.
The automation matches charges against stored receipts using filename metadata and flags missing documentation. The workflow deliberately avoids unnecessary OCR where filename information is sufficient.
Amex Backup Search & Follow-up
A connected workflow uses Microsoft 365 to search email for missing invoices.
Once the user confirms the relevant documents, the workflow can file them and prepare grouped follow-up emails for outstanding documentation. The employee remains responsible for sending the communication.
Draw Package Preparation
deepsense.ai built an automation that uses structured Procore exports to populate the AIA, calculate retainage by line, validate roll-ups and check backup completeness before sign-off.
Invoice Data Extraction
A shared capability extracts information from non-Procore invoice PDFs into the Excel schedule, with validation rather than automatic acceptance of extracted information.
The same capability can support subsequent invoice system entry without requiring a separate extraction workflow.
Automation Prioritization
Ten candidates across four workflows were evaluated using an Impact × Time-to-Value × Feasibility model.
This gave the client a structured basis for determining which automations justified attention first instead of prioritizing opportunities based solely on theoretical impact.
Session → Skill Methodology
deepsense.ai established a six-step procedure that non-technical employees could reuse:
watch the work → write it in plain English → split rules from judgment → tell Claude → test and correct → save and reuse
This turned the engagement from a set of individual automation builds into a repeatable approach to identifying and developing additional workflows.
Process
The engagement progressed through four connected phases. Discovery ran in parallel across all four processes. We cooperated with the team and analyzed their daily tasks. After that, we developed solutions aligned with the team’s needs and iterated based on their feedback.
The resulting candidates were then scored and prioritized, and the two highest-ranked automations were built one after the other, each validated against real monthly data before the next began.
1. Enablement
deepsense.ai ran two company-wide sessions covering Claude fundamentals and prompting, followed by hands-on work with Claude Cowork.
2. Discovery
The team conducted deep dives into four high-value accounting processes and mapped each workflow end to end.
A particularly important part of the method was observing the person who actually performed the work. Rather than relying exclusively on documented procedures, deepsense.ai examined the real sequence–including the manual steps and exceptions that determine whether an automation will work in practice.
We followed the same approach during development: showing them the automation and iterating based on their feedback.
3. Build and Validation
The team built and tested two skills using real monthly data.
Client process experts compared the results with how they would perform the work themselves, while deepsense.ai iteratively corrected and refined the workflows.
4. Roadmap and Adoption Model
The engagement concluded with a three-wave rollout plan and an adoption operating model incorporating department champions, a shared skill library, Safe-AI guidelines and a monthly build cadence.
5. Outcome
The 3.5-week engagement resulted in two working automations at PoC scope, automation designs for additional workflows, documented process maps, a prioritized automation backlog, and a roadmap for further adoption.
The team captured process knowledge that had previously depended on individual employees in documented workflows. The client also gained a repeatable method that non-technical employees can use to turn processes into AI-assisted automations.
Crucially, we achieved this without replacing the client’s transactional systems or removing employees from key decision and approval points.
Results
The engagement delivered:
- two working automations at PoC scope,
- four end-to-end process analyses,
- 10 scored and prioritized automation candidates,
- automation designs for invoice system and real-estate tax workflows,
- two company-wide enablement sessions,
- a reusable skill-building methodology,
- step-by-step run guides,
- and a three-wave rollout roadmap.
For monthly card reconciliation, the preparation stage, which previously took hours, was reduced to about 20 minutes.
Business Impact
The immediate impact was a reduction in repetitive manual work within the selected accounting workflows.
The automations also reduced reliance on repeated re-keying of the same information, addressing an associated source of error while keeping authority gates with employees.
The broader outcome was organizational capability. Instead of leaving the client with two isolated PoCs, deepsense.ai provided a repeatable method, documented processes, a prioritized opportunity backlog and an adoption model that created a foundation for exploring automation opportunities beyond the initial accounting use cases.
Lessons Learned
The engagement demonstrated that valuable automation opportunities do not necessarily require deep system integration.
In this environment, much of the repetitive work existed in the manual bridges between systems. Addressing the file layer allowed deepsense.ai to target that effort without waiting for changes to transactional platforms.
The project also reinforced that automation prioritization needs to account for feasibility and time-to-value, not theoretical impact alone. The strongest initial candidates were those that could be implemented using the available environment without significant external dependencies.
Finally, successful workflow automation depended on understanding how work was actually performed. Observing process experts carrying out real tasks exposed details and judgment points that a written procedure alone could miss.
6. Summary
deepsense.ai helped a US real-estate developer move from identifying AI opportunities to testing them in real accounting workflows.
Over 3.5 weeks, the team mapped four processes, assessed and prioritized 10 automation candidates, built and validated two working PoCs, and established a repeatable method for developing additional AI-assisted workflows.
The engagement combined process analysis, AI engineering, validation, employee enablement, and adoption planning. By focusing automation on the manual bridges between existing systems and keeping consequential decisions human-controlled, deepsense.ai demonstrated a practical path to reducing repetitive accounting work without requiring a broader systems transformation.
The result was two working automations. In addition, the client gained a structured approach to deciding where AI is worth applying, how to turn real processes into working automations, and how to keep building beyond the initial PoCs.
