AI Product & Operations Leader

Khalid Alali

I build and ship the AI that runs the operation.

AI product and operations leader with 14+ years redesigning cross-functional workflows, shipping AI into production operations, and driving adoption across support, payments, finance, and partner teams.

At Fixlane (formerly RepairWise) I own AI and internal-systems strategy end to end: identify the highest-ROI workflow, redesign the process, build or buy the system, and drive adoption across repair ops, finance, support, and partners.

Head of Business Operations and AI Product · Fixlane (formerly RepairWise)

$800K
partner-portal revenue

Launched a partner portal across 15 shops and 400+ repair jobs, enabling $800K in revenue.

$3M
CRM migration savings

Turo: 1,900-user CRM migration covering 12M cases and 5 integrations.

$150K
annual AP savings

LLM-powered AP automation at Fixlane (formerly RepairWise).

20%
faster coding & payouts

AI labor-coding prediction engine with a weekly evaluation loop.

OperationsAIFinanceSupportPartnersPlatformShops

Career & leadership

2011–present

From writing the systems to deciding which ones get built — operations, internal products, and AI in production.

Fixlane (formerly RepairWise) · Menlo Park2023–Present

Head of Business Operations and AI Product

AI-enabled operations platform for EV repair. Owns AI and internal-systems strategy end to end.

  • Identifies the highest-ROI workflows, redesigns the process, builds or buys the system, and drives adoption across repair ops, finance, support, and partners.
  • Built an AI labor-coding prediction engine with a weekly evaluation loop (20% faster coding and payouts), and a RAG technician copilot that drafts customer responses and recommends knowledge-base articles.
  • Built and deployed a production MCP server that exposes core repair workflows as tools AI agents can call.
  • Launched a partner portal across 15 shops and 400+ repair jobs, enabling $800K in revenue.
  • Built LLM-powered AP automation ($150K annual savings), a multi-party inventory system that uncovered $200K+ in excess parts, and two internal CRMs with human-in-the-loop shop-record enrichment.
Turo · San Francisco2018–2022

Senior Business Operations Manager (Head of Internal Products)

First BizOps hire. Managed 3 data analysts and 4 contract engineers through 400%+ company growth.

  • Owned the ops technology roadmap across support, claims, payments, and shared ops.
  • Led a 1,900-user CRM migration covering 12M cases and 5 integrations, saving $3M.
  • Launched global payouts processing $120M+/year; cut payout time 99% and costs by $200K.
  • Deployed AI-enabled contact flows that reduced support contacts 30%.
iMENA Group / OpenSooq · Saudi Arabia2015–2016

Country Head

Built and led a 13-person sales and ops team across 2 countries and 4 cities.

  • Built Salesforce CRM, workflows, and dashboards; onboarded 350 dealers and reached ~90% share.
  • Launched Saudi Arabia and Kuwait.
Rocket Internet · Lamudi Saudi Arabia2014–2015

Managing Director and Co-Founder

Scaled a real-estate marketplace from 500 to 5,000 listings.

McKinsey & Company · Dubai2011–2014

Business Analyst

Top 15% performer; banks and financial institutions.

Education

  • UC Berkeley HaasMBA, Strategy and Technology · 2018
  • Concordia UniversityBCom, Finance and International Business · 2010

Flagship work

Context → role → decisions → outcome

Three systems with measured outcomes, then the AI-judgment work that does not yet have a public accuracy number. Same pattern: pick the workflow, decide where the model belongs, ship it, stay with it.

Fixlane (formerly RepairWise) · partner networkShipped

Partner portal for the shop network

A portal for partner shops to run repair jobs across the EV repair network.

Screenshot or 60-second walkthrough forthcoming.

Context

Fixlane (formerly RepairWise) runs an AI-enabled operations platform for EV repair. Partner shops needed a surface to take and complete repair jobs without living inside internal ops tools.

My role

As Head of Business Operations and AI Product I own AI and internal-systems strategy end to end. For the partner network that meant identifying the job-flow as a high-ROI workflow, building the portal, and driving adoption across shops.

Decisions

Identify the partner job-flow as a high-ROI workflow, build the portal, and drive adoption across the shop network.

Measured outcome

Launched across 15 shops and 400+ repair jobs, enabling $800K in revenue.

Systems
Partner portal
Fixlane (formerly RepairWise) · repair ops & payoutsShipped

AI labor-coding prediction engine

Predict labor codes for repair jobs, then pay shops faster — with a weekly evaluation loop instead of a one-shot model.

Screenshot or 60-second walkthrough forthcoming.

Context

Repair jobs have to be coded before shops get paid. Slow or inconsistent labor coding delays payouts and creates rework between technicians, ops, and finance.

My role

I built the AI labor-coding prediction engine and put a weekly evaluation loop around it so the model is judged on coding and payout speed, not on a demo.

Decisions

Prediction is the product, not a chatbot. Keep a weekly evaluation loop so the engine stays tied to coding and payout time. Do not treat a one-time accuracy snapshot as done.

Measured outcome

20% faster coding and payouts.

Systems
Labor Code Prediction API
PythonRender
Fixlane (formerly RepairWise) · shops, finance, APShipped · portal live · advisor beta

Multi-party inventory and AP automation

See parts across shops and parties, and take the manual work out of accounts payable — with humans confirming what the model proposes.

Screenshot or 60-second walkthrough forthcoming.

Context

Parts sit across shops and counterparties. Attestation, keep-versus-return, and AP lived in separate flows, so excess stock and invoice work lagged.

My role

I built a multi-party inventory system and LLM-powered AP automation, and I designed keep/return as a recommendation layer: shops still confirm; the model proposes.

Decisions

Attestation is the source of truth. Keep/return is a recommendation, not an auto-action. Batches of 20 parts per Claude call kept structured JSON reliable enough for automatic RMA handoff without waiting on a perfect catalog. AP is a model-assisted workflow with finance in the loop.

Measured outcome

The multi-party inventory system uncovered $200K+ in excess parts. LLM-powered AP automation produced $150K in annual savings.

What changed

Before: weekly status in one place, returns and requests elsewhere, RMA started by hand. After: portal attestation → keep/return proposal → RMA path, with requests and large-part email routing on the same loop.

Systems
Parts Inventory PortalInventory Keep/Return AdvisorParts Request FormRMA ParserShipping Label Printer
Next.jsSupabaseClaude APIMotherDuckFilloutAirtable
Fixlane (formerly RepairWise) · finance operationsShipped

AI invoice matching & reconciliation

Match OEM invoice line items against orders and surface billing discrepancies for finance review — the model proposes, finance confirms.

Context

OEM invoices included incorrect shipping charges and taxes, charges belonging to other accounts, and duplicate billing. Finance needed to reconcile those charges against order records and establish which discrepancies required escalation.

My role

I built an AI-assisted reconciliation workflow in a Next.js app on Supabase and MotherDuck, with human validation before anything became an OEM dispute.

Decisions

Use AI to parse invoices and match line items to orders, with human validation before escalating discrepancies to the OEM. The model proposes matches and flags; finance confirms before anything becomes an OEM dispute.

Measured outcome

Uncovered $200k in incorrect OEM invoice charges, validated through human review and escalated for resolution. Separate from the $200K+ in excess parts uncovered by the multi-party inventory system.

What changed

Pipeline shape: ingest OEM invoice → parse line items → match against order records → flag shipping / tax / wrong-account / duplicate mismatches → finance validates → escalate confirmed errors to the OEM.

Systems
Parts Invoice ReconciliationAttachment ViewerAMP Tracker & Invoice Generator
Next.jsSupabaseMotherDuckClaudeVercel
Fixlane (formerly RepairWise) · technician supportBeta · CRM-integrated

RAG technician copilot

AI-assisted replies and workflow recommendations for shop messages on active repair orders. Beta — no live accuracy figure to publish.

Context

Partner shops ask technicians about repair steps, parts, software updates, and warranty eligibility through the CRM. Answering requires both the conversation’s current context and relevant repair documentation. Routine messages compete for attention with questions that need deeper technical judgment.

My role

I built a RAG technician copilot that drafts customer responses and recommends knowledge-base articles. I extracted the pipeline from the CRM into a dedicated asynchronous service and decided when the model is allowed to speak.

Decisions

I separated three decisions: whether a message needs a reply, which workflow actions to propose, and whether the response should be sent automatically or drafted for review. Courtesy messages bypass the model. Known response patterns follow deterministic rules. Other replies use retrieved repair documentation and a grounding verifier.

Measured outcome

No live copilot accuracy or adoption figure. The implemented pipeline produces a reply decision and structured workflow recommendations from the same inbound message. It is in beta, CRM-integrated. Research findings below are not live accuracy.

Evaluation and failure handling

Courtesy and acknowledgement messages are filtered with deterministic rules before any model call. Known safe patterns (for example parts-status or upload-confirmation replies) can auto-send; other drafts run through a grounding verifier that checks claims against retrieved docs and order state. If the verifier flags unsupported sentences, the system regenerates once and holds the reply for review when flags remain. Workflow actions are scored in separate fan-out stages from reply quality, then merged so already-ordered parts and conflicting signals do not become duplicate recommendations. Predictions persist before the CRM callback, so duplicate webhooks can reconcile against stored state instead of inventing a second draft.

Research that informed the product

Reconstructed shop-guidance episodes (~2,000) showed about 76% resolve as mark-complete and about 62% of new guidance tasks are formulaic FOTA software-update ladders. Those findings motivated structured automation and retrieval; they are not live copilot accuracy.

What changed

Example shape (sanitized): shop asks whether a software update is needed → retrieve FRS/playbook evidence for the order → propose a FOTA ladder task if criteria match → draft a reply citing the retrieved steps → verifier checks claims → CRM receives draft or auto-send with category and confidence.

Systems
Shop Message CopilotTSB Ingest PipelineLabor Code Prediction API
Next.jsTypeScriptTrigger.devClaudeMotherDuckPostgresVercel

Also shipped

Fixlane (formerly RepairWise) · platformShipped · production

Production MCP server

Built and deployed a production MCP server that exposes core repair workflows as tools AI agents can call. No public volume or adoption figure.

Repair Workflow MCP ServerMCP
Fixlane (formerly RepairWise) · operations & OEMShipped · maps live · router beta

Network assignment & coverage

Partner shops were assigned by hand, and leadership lacked a clear coverage number for owners near active shops. Distance-first nearest-neighbor assignment over Census ZCTAs, plus Mapbox coverage and OEM map surfaces. 89% historical assignment agreement in backtesting; 68.4% of Tesla owners within 15 miles of an active shop.

Auto-Assignment RouterShop Coverage MapNetwork MapPythonMapboxMotherDuck
Fixlane (formerly RepairWise) · finance opsShipped

AP credit-memo pipeline

Credit memos arrived as email PDF attachments that Airtable could not ingest cleanly. Make mailhook → PDF filter → Drive → Airtable, with an attachment viewer for the resulting files. AP could open credit-memo PDFs without manually uploading attachments.

AP Credit Memo PipelineAttachment ViewerMakeDriveAirtableNext.js
Fixlane (formerly RepairWise) · shop-facing serviceShipped

Expert Assist

Shops needed Tesla technical help during the job. A form that never became a callback was useless. Paid consultation workflow: shop intake and ticket creation, Dialpad callback, and Zoho Lens video. Published pricing $50 per 20 minutes / $150 per hour. Shops could go from intake to ticket to call to video on a live paid product. Separate from the technician copilot. No adoption or revenue figure beyond that pricing.

Expert AssistNext.jsTypeScriptDialpadZoho Lens

Systems atlas

38 systems

Inventory behind the flagship work: 33 live, 4 beta, 1 draft. Status is labeled per row. Experiments stay visible; they do not compete with production.

SystemAudienceJobStatus
Shop Status ConsoleRailway · MotherDuck · Supabase
Ops / network managersTiered partner-network status and weekly cadence reviewlive
Fixlane CRMNext.js · Supabase · Places API
OpsContacts, locations, accounts; Places enrichment into CRMlive
FSE Visit ReportsFillout · Airtable · Vercel
Ops / OEM field serviceVisit report form; notify shop + Fixlane; Airtable recordlive
Attachment ViewerNext.js · Vercel · Make · Airtable
Ops / financeBrowse credit-memo and OEM invoice PDFs for a given orderlive
Auto-Assignment RouterPython · ZCTA
OpsNearest-neighbor shop assignment over Census ZCTA (beta: backtested, not broadly rolled out)beta
RMA ParserResend · Vercel
Ops / logisticsInbound webhook → email routing; large-parts classificationlive
Shop Onboarding PortalNext.js · Supabase
Ops / shopsB2B onboarding checklist: docs, status, contract acklive
AMP Tracker & Invoice GeneratorNext.js · Google SSO · Vercel
Ops / financeRental AMP tracking and invoice generationlive
Parts Inventory PortalNext.js · Supabase · Claude API
Shops + coordinatorsWeekly inventory attestation; keep/return path; RMA handofflive
Inventory Keep/Return AdvisorClaude API · MotherDuck
ShopsBatched LLM keep/return (20 parts/call); structured JSON; auto-RMA (beta pilot)beta
Parts Request FormFillout · Airtable
Shops / opsStructured parts requests; 518 entries; defect flagslive
Expert AssistNext.js · Dialpad · Zoho Lens
ShopsPaid Tesla consult: intake → Dialpad → Zoho Lens ($50/20min, $150/hr)live
Shop Message CopilotNext.js · Trigger.dev · Claude · MotherDuck · Postgres
Shops / techniciansAsync AI replies + workflow actions on CRM shop messages; grounded in KB/TSB/playbooksbeta
Partner portal
Partner shopsPortal across 15 shops and 400+ repair jobslive
VinFast Knowledge BaseReplit
ShopsVinFast technical knowledge baselive
Network MapNext.js · Mapbox · Vercel
OEM / internalVinFast shop network coverage visualizationlive
VinFast Collision MapMapbox · Supabase · Vercel
Public / OEMMapbox map of VinFast collision shopslive
VinFast Tow MapMapbox · Supabase · Vercel
OEM agents / internalTow capacity and agent coveragelive
VinFast Zero-Pay TrackerMetabase · Postgres
OEM / finance analysts$0 warranty/goodwill service orders by shop, month, partlive
Parts Invoice ReconciliationNext.js · Supabase · MotherDuck
Ops / financeAI-assisted OEM invoice line matching to orders; discrepancy review and escalationlive
AP Credit Memo PipelineMake · Drive · Airtable
Finance / APMailhook → PDF filter → Drive → Airtable for credit memoslive
Collections WorkflowBraze
Finance / collectionsAging-bucket reminders and collections handofflive
Shop Coverage MapMapbox GL · ZCTA · MotherDuck
Ops / dataZCTA Mapbox coverage; Tesla owner proximity analyticslive
Diagnostic Bucket TrackerMetabase · DuckDB
Data / opsTesla repair orders across 12 diagnostic bucketslive
Health Check FunnelPostHog · HogQL
Growth / dataFunnel by state and acquisition source (HogQL)live
Pre-Appointment Messaging AnalyzerDuckDB · Python
Ops / dataPre-service message-volume cohortsbeta
Labor Code Prediction APIPython · Render
Ops / techniciansLabor-code prediction with a weekly evaluation looplive
Repair Workflow MCP ServerMCP
Internal / AI agentsExposes core repair workflows as tools AI agents can calllive
LLM-powered AP automation
Finance / APLLM-powered accounts payable automationlive
TSB Ingest PipelinePython · Claude
Internalpdfplumber + Claude → structured repair steps from TSB/Field Action PDFslive
Location Enrichment PipelineSupabase · Places API
Internal / CRMGoogle Places enrichment into CRM location recordslive
Urgently Tow IngestDocker · Python · Render
InternalRoadside/tow data ingest into Postgreslive
Zoho Sign WebhooksNode · Zoho Sign · Render
InternalSign events → Airtable; recall → update → resubmit contract flowlive
GCS UploaderNode · GCS · Render
InternalPDF upload to GCS for knowledge-base contentlive
Customer IntakeNode · Render
Consumers / opsPre-drop-off concern and VIN capture for repair orderslive
Consumer Help CenterReplit
Drivers / ownersPublic help centerlive
Directus ProxyNode · Directus · Render
Internal / contentProxy articles and search over Directus (+ GCS PDFs)live
Case Enrichment AgentClaude · pgvector
InternalPrototype: warranty pre-screen + rework detectiondraft

How I lead

Teams and adoption

Highest-ROI workflow first

Identify the workflow, redesign the process, build or buy the system, and drive adoption across repair ops, finance, support, and partners.

Stand up the function

First BizOps hire at Turo. Managed 3 data analysts and 4 contract engineers through 400%+ company growth, and owned the ops technology roadmap across support, claims, payments, and shared ops.

Build the team in market

At iMENA / OpenSooq, built and led a 13-person sales and ops team across 2 countries and 4 cities, and launched Saudi Arabia and Kuwait.

Keep humans in the loop

Shipped two internal CRMs with human-in-the-loop shop-record enrichment. The model proposes; operators confirm before the record or the customer-facing action is final.

How I work

Builder, with judgment

Cross the stack

I take an ops problem from an unclear request to a shipped UI, data flow, integrations, and deploy — then stay with it in production.

Build for operations

I design for weekly partner reviews, shop floors, finance handoffs, and the places field businesses usually break.

Control the model

I decide when deterministic rules are enough, when to retrieve and ground, and when a draft must wait for verification before it reaches the shop. Courtesy messages bypass the model. A grounding verifier checks claims against retrieved docs and order state.

Ship, then evaluate

I ship a thin first version, then run an evaluation loop — weekly, in the case of the labor-coding engine — before I widen automation.

Technical range

By capability

Product

  • Next.js
  • React
  • TypeScript
  • Mapbox

Data

  • Supabase / Postgres
  • MotherDuck
  • DuckDB
  • Metabase
  • PostHog

Automation

  • Python
  • webhooks
  • Make
  • Airtable
  • Fillout
  • Braze

AI

  • retrieval and grounding
  • structured predictions
  • multi-stage orchestration
  • output verification

Infrastructure

  • Vercel
  • Render
  • Railway
  • GCS
  • Google Places
  • Dialpad
  • Zoho

Get in touch

I'm looking for Head or Director roles where I own the AI that runs the operation — AI operations, platform and internal products, business technology, or GTM/ops systems.

Seeking · Head / Director of AI Operations · Platform / Internal Products · Business Technology · GTM / Ops Systems

I personally build and ship the AI into operations — not a slide-layer over someone else's stack.